The Diagnostic Error in Medicine 14th Annual International Conference

Is There Evidence of Provider Disparities in Responses to USMLE Step 3 Multiple Choice Questions?

C. Runyon ¹, K. Swygert¹*, S. McEllhenney¹, K. DeRuchie¹, M. Paniagua¹

*(Presenting Author)

¹NBME, Philadelphia, PA

Purpose/Problem: Implicit biases of physicians can negatively impact the diagnosis and treatment of patients based on demographics and social characteristics. In an examination setting, these biases may result in different responses to vignette-based multiple-choice items if the patient characteristics are mentioned. The present study investigates the extent to which varying patient characteristics results in different response distributions to vignette-based items on the USMLE Step 3 exam.

Description of Program, Assessment, or Study: A convenience set of 20 USMLE Step 3 vignette-based multiple-choice items had variants created where the patient race and/or gender were systematically altered; original items did not mention race or gender. The 20 items and 115 variants were included on Step 3 examinations delivered throughout 2019. No examinee saw both an original item and a variant, nor multiple variants. Chi-square tests were used to examine differences between responses to the original item and variants. A non-significant test indicates no difference in response distributions across the original item and all variants. A significant test identifies when there is a change in item difficulty or if a particular incorrect response option (distractor) becomes more/less attractive.

Outcomes: A total of 57,866 item responses across 28,933 examinees were collected for the study; each item was viewed approximately 430 times. Seventeen of the 20 chi-square tests were not statistically significant. Two of the 3 statistically significant chi-square tests were significant purely due to a statistical artifact and not a substantive reason. The third statistically significant chi-square test revealed a difference in preference across the distractors depending on the gender of the patient; this pattern was not related to the gender of the examinee.

Discussion: These results indicate that changes to the patient demographic characteristics do not appear to influence the distribution of examinee responses across response options. The content in the item accounting for the only observed discrepancy warrants further study to better understand why the significant result was observed, but this singular result across 20 items is not seen as an indicator of systematic differences in examinee responses.

Significance of Findings: The inclusion of patient characteristics did not change the distribution of examinee responses on a set of USMLE Step 3 vignette-based multiple-choice items. No differences in response options were observed based on the patient characteristics of gender and/or race, indicating that we did not observe differential treatment strategies based on patient characteristics in this examinee population on this assessment.

A Qualitative Root Cause Analysis of Barriers for Diagnostic Error Reduction in Japan

T. Harada ¹, T. Miyagami², T. Watari³*, T. Kawahigashi⁴, Y. Harada⁵, K. Shikino⁶, T. Shimizu⁵

*(Presenting Author)

¹Showa University, Japan

²Juntendo University, Japan

³Shimane University, Izumo, Japan

⁴Yamato Tokushukai Hospital, Japan

⁵Dokkyo Medical University, Tochigi, Japan

⁶Chiba University Hospital, Chiba, Japan

Background: Diagnostic errors are also a major problem in Japan. Two academic societies, the Japanese Society of Internal Medicine and the Japanese Society of General Hospital Medicine have established working groups addressing diagnostic errors. Despite these steps and increased awareness, nationwide action to improve diagnostic processes has yet to be implemented. To better understand the current situation, we, the volunteer members of the “Diagnostic Errors Working Group” of the Japanese Society of General Hospital Medicine, conducted a root-cause analysis to identify barriers to implementing a comprehensive diagnostic error reduction strategy in Japan.

Methods: Six Japanese physicians met online in February 2021 and listed potential hurdles in spreading the movement to diagnostic error prevention in Japan using the affinity diagram approach, as per Kawakita Jiro’s method, through this process, a list of 100 possible problems and causes was generated. After duplicate items were deleted, the remaining problems/causes were then grouped and labeled.

Results: We identified seven groups of contributory factors (Figure); Education, system, environment and culture, policies, paramedical staff, physician, and patient-related factors. There are few opportunities for diagnostic quality and safety to be taught in Japanese medical schools. Not only concepts regarding safety and error but the need for reflection and feedback are seldom addressed. In addition, the number of educators qualified to teach these subjects is also small, another challenge for Japanese medical education. Regarding the healthcare system, a major problem concerns electronic health record systems in Japan. There are also few medical institutions where mortality and morbidity conferences take place to discuss and learn from cases of diagnostic errors. In addition, Japan’s culture of “shame” may have slowed down efforts to counter diagnostic errors. As Physician-level factors, we concluded that inertia, paternalism, perfectionism, and hierarchy can contribute to diagnostic error tolerance. Finally, there is no organization in Japan that focuses on patient safety at the national level, such as the Agency for Healthcare Research and Quality (AHRQ) in the United States.

Conclusion: Identifying these seven primary groups can hopefully improve understanding of the present situation and provide a starting point for further discussion and planning towards the primary goal of improving diagnosis and patient outcomes.

Diagnostic Errors Among Patients/Families with Limited English-language Health Literacy or Disadvantaged Socio-economic Position

S. Bell ¹, Z. Dong¹, L. Ngo¹, P. McGaffigan², E. Thomas³, F. Bourgeois⁴

¹Beth Israel Deaconess Medical Center, Boston, MA

²Institute for Healthcare Improvement, Boston, MA

³UT Houston – Memorial Hermann, Houston, TX

⁴Boston Children’s Hospital, Boston, MA

Background: Vulnerable patient populations may be particularly susceptible to diagnostic errors because of language barrier, reduced self-advocacy, lower health literacy, or biased care. Although communication is a key contributing factor in safety events, little is known about how patients and families with limited English-language health literacy (LEHL) or disadvantaged socio-economic position (SEP) experience diagnostic errors.

Method: We analyzed the 2017 IHI U.S. population-based survey to compare patient-reported diagnostic errors, contributing factors, and long-term (>1 month) impacts between respondents with LEHL (Spanish-preferring or </=“somewhat confident” completing medical forms alone) or disadvantaged SEP (income of <$50,000/year or <high school education) and their counterparts. The survey was conducted in English or Spanish, online or by telephone. We used chi-sq test to compare participant responses by LEHL/SEP group, and stepwise logistic regression to model contributing factors to diagnostic error.

Results: 596 respondents reported a diagnostic error, comprising the study population. 381 participants met criteria for either LEHL or disadvantaged SEP. The most common contributing factor to diagnostic error reported by all participants was “not feeling listened to” (69%). After adjusting for sex, race and ethnicity, and physical health, individuals reporting LEHL/disadvantaged SEP were more likely to identify the following contributing factors compared to their counterparts: No qualified translator or healthcare provider that spoke the patient/family’s language (Odds ratio and 95% CI: 4.4 (1.3,14.9)); not understanding the follow-up plan (1.9 (1.1,3.1); too many providers but no clear leader (1.8 (95%CI 1.2,2.7)); inability to keep follow-up appointments (1.9 (1.1,3.2); not being able to pay for care (2.5 (1.4,4.4); and out-of-date or incorrect medical records (2.6 (1.4,4.8). Over two-thirds of individuals with LEHL/disadvantaged SEP reported long-term physical and long-term emotional impacts, and half reported long-term financial impacts from the error. Participants with LEHL/disadvantaged SEP were significantly more likely to report long-term effects on physical health, emotional health, financial wellbeing, and relationships with family, compared to their counterparts (p<0.05 for all comparisons).

Conclusions: Individuals reporting LEHL or disadvantaged SEP identified unique and actionable contributing factors to diagnostic errors, including communication, care access and coordination, and out-of-date/inaccurate medical records. Interpreter access should be viewed as a diagnostic safety imperative, social determinants affecting care access/affordability should be routinely addressed as part of the diagnostic process, and patients/families should be encouraged to access and update their medical records. The frequent and disproportionate long-term impacts from self-reported diagnostic error among LEHL/disadvantaged SEP respondents raises urgency for greater prevention and longer-term supportive measures.

Performance of ICU Severity Scoring Systems Across Different Ethnicities in Open-source and Private ICU Datasets

E. Chou ¹, I. Banerjee¹, J. Gichoya¹, C. Josef¹

¹Emory University, Atlanta, GA

Background: Despite wide utilization of severity scoring systems for case-mix determination and benchmarking in intensive care units, the possibility of scoring bias across ethnicities has yet to be thoroughly examined. Recent guidelines on the use of illness severity scores to inform triage decisions for allocation of scarce resources such as mechanical ventilation during the current COVID-19 pandemic warrant examination for possible bias in these models. Previously, we investigated the performance of three severity scoring systems across ethnic groups in two large ICU databases and found statistical evidence that suggested that illness severity scores did not discriminate for severity of disease but were poorly calibrated for African Americans and Hispanics where these scores over predicted mortality. In this study, we continue to investigate the performance of two severity scoring systems (specifically OASIS, SOFA, and the first-day SOFA derivative) across ethnic groups in another vast ICU database and in a safety net hospital ICU dataset to further identify possible ethnicity-based bias.

Methods: Data from the Medical Information Mart for Intensive Care (MIMIC-IV) and from the Grady Memorial Hospital were analysed for score performance in African Americans, Asians, Hispanics and Whites after appropriate exclusions. Discrimination and calibration were determined for both scoring systems in all four groups.

Results: We have similar findings to prior studies, whereby the area under the receiver operating characteristic curve (AUROC), our measurement of discrimination, has few significant differences among the groups. The AUROCs did not display any discernible systematic patterns of bias across racial groups. In contrast, the standardised mortality ratios (SMR), our measurement of calibration, indicated consistent and significant patterns of difference between Hispanics versus the other ethnicities in both cohorts. While calibrations were imperfect for all groups, the scores consistently demonstrated a pattern of over-predicting mortality for Hispanics compared to other ethnicities.

Conclusion: The systematic differences in calibration across ethnic groups suggest that the illness severity scores reflect bias in their prediction of mortality. Further investigation of available data will attempt to identify major contributing comorbidity factors that may influence and explain the poor calibration of these ICU scoring systems for Hispanics.

Stop Dropping the Ball - An Academic - Community Partnership to Decrease Follow-Up System Failures in a Rural Health System

C. Peterson ¹, A. Olson², M. Lindow¹, H. Keyes³

¹Lakewood Health System, Baxter, MN

²University of Minnesota, Minneapolis, MN

³MAPS

Background: There are significant challenges with appropriate follow-up of diagnostic testing results after acute care encounters in the US. Successful follow-up by ED patients in the US is frequently poor and has been estimated to be 26-56% depending on the population. Follow-up system failures are a contributor to diagnostic error. However, the scope of the problem in rural health care systems is not well described and the effectiveness of described solutions in this setting is not known. Here we report the baseline results from an academic-community partnership aimed at decreasing follow-up system failures in a rural health system.

Methods: Lakewood Health System (LHS) is a non-profit, independent rural health care system that, in partnership with the University of Minnesota Medical School, the Minnesota Alliance for Patient Safety, and Stratis Health, has developed a team to address diagnostic error within their health care system due to follow-up system failures. Follow-up performance, including failures, was analyzed for two specific situations arising in emergency department (ED) visits: findings requiring followup on chest radiography and abnormal eGFR. Charts were reviewed and rates of appropriate follow-up determined, in addition to identification of factors associated with appropriate follow-up and system failures.

Results: There were several factors found to be associated with a higher risk of patients being lost to follow-up. Of those who did not follow-up regarding abnormal eGFR, 27% were non-Staples residents and had no primary care provider (PCP); this was also true for of 28% for the chest imaging group. Of the patients who did follow-up, 100% had a PCP. Both eGFR and chest imaging groups also had higher rates of no follow-up in those who had a PCP from an outside facility; 25% and 14%, respectively. Another factor associated with non-compliance with follow-up was being a resident of a long-term care facility; 27% did not follow-up in the eGFR group and 28% in the chest imaging group. In the eGFR group, 53% of the non-follow up group also had Medicare insurance.

Conclusions: Our data suggests that recognizing these factors and susceptible patients for loss to follow-up after an ED visit and putting focus on scheduling follow-up appointments along with encouraging patients to establish with a primary care provider may help increase follow-up compliance and decrease delays in diagnosis. The project is currently focusing on using community engagement to identify strategies to mitigate these failures and decrease diagnostic error.

Assessing Diagnostic Safety in Medical Offices: The AHRQ Surveys on Patient Safety Culture™ (SOPS^®) Diagnostic Safety Supplemental Item

T. Famolaro ¹, J. Sorra¹, Y. Naomi¹, K. Zebrak¹, C. Deborah¹, A. Caporaso¹, L. Gray¹, J. Kirchner¹

¹Westat, Rockville, MD

Background: Diagnostic errors are the most common patient safety error in the medical office setting (Gandhi et al., 2006). To assess this problem, AHRQ established a contract with Westat to develop and pilot test a supplemental item set on diagnostic safety that can be added to the Surveys on Patient Safety Culture^TM (SOPS^®) Medical Office Survey. The purpose of the SOPS Diagnostic Safety Supplemental items is to enable medical offices to assess provider and staff perceptions about how the organizational culture in medical offices supports the diagnostic process, accurate diagnoses, and communication around diagnoses.

Methods: The development process began with a literature review on diagnostic safety, and background interviews with experts and front-line staff to identify key dimensions. We developed survey items and cognitively tested the items in three rounds of cognitive interviews with a total of 32 medical office providers and staff. After each round of testing, we consulted with a Technical Expert Panel (TEP) on the survey iterations. A pilot study of the diagnostic safety supplemental items was conducted in 66 U.S. medical offices in late 2020. The supplemental items were administered toward the end of the SOPS Medical Office Survey, just before the background questions. We then analyzed the pilot data to examine the psychometric properties of the supplemental items (e.g., reliability and validity), with the goal of including only the best performing items. Results from these analyses and input from the TEP were used to determine which items to retain. The final supplemental item set includes 12 items that assess three composite measures made up of 3 to 5 survey items. • Testing and Referrals • Provider and Staff Communication Around Diagnosis • Time Availability

Results: The results are based on 812 respondents across 66 medical offices, with an overall response rate of 44%. The final supplemental items had good psychometric properties with Cronbach’s alpha reliability for each composite measure over 0.70. The composite measure Testing and Referrals was the most positive area of diagnostic safety (79% positive), followed by Provider Communication Around Diagnosis (65%) and Time Availability (56%).

Conclusion: The Diagnostic Safety Supplemental Items are psychometrically sound and measure important aspects of diagnostic safety in medical offices. Medical offices and health systems can use the survey items and results to identify ways to improve the diagnostic process.

Global Virtual Morning Report: Evaluating the Effects of a Global Diagnostic Reasoning Program

R. Medina dos Santos ¹, M. J. Alemán², K. Camacho³, G. Kaur⁴, V. Roldán Galarreta⁵, S. Banthiya⁶, G. Pucci⁷, K. Nandlalbhai Patolia⁸, Y. Kara Lau⁹, P. Vilas Athavale⁹, R. Geha⁹, A. Rodman¹⁰, S. Nematollahi¹¹ (senior author)

¹Universidade Estadual de Maringá, Maringá, Brazil

²Universidad Francisco Marroqui

³Universidad Cientifica del Sur, Lima, Peru

⁴Brigham and Women’s Hospital, Boston, MA

⁵Universidad Peruana Cayetano

⁶Ascension Providence Hospital

⁷São Paulo State University

⁸B.J. Medical College

⁹UCSF, San Francisco, CA

¹⁰Beth Israel Deaconess Medical Center, Boston, MA

¹¹Johns Hopkins, Baltimore, MD

Purpose/Problem: The last decade has seen an explosion of organic digital education (ODE), which includes virtual educational communities with digital educational products. Previous research has shown that despite high rates of internet and mobile phone use in low- and medium income countries (LMIC), ODE is mostly used in high-income countries.

Description of Program, Assessment, or Study: In 2020, The Clinical Problem Solvers created the first multi-institutional case-based diagnostic reasoning conference, Virtual Morning Report (VMR), soon followed by Global Virtual Morning Report (GVMR), designed specifically for medical trainees outside the US. The GVMR community intentionally includes international team members that provide translations and encourages participants to share their thoughts in the language they feel most comfortable. These interventions provide a natural experiment and control group to study if an ODE resource specifically targeted towards learners in LMIC can succeed. We collected usage and demographic information from four weeks of VMR and GVMR and calculated penetration in LMIC and high-income countries. Data was normalized per total number of morning report sessions and for the total population of individual countries. The chi-square test was used to compare 1) the non-normalized number of participants from LMIC in the VMR and GVMR groups and 2) the non-normalized number of participants from non-English-speaking countries (NESC) in the VMR and GVMR groups.

Outcomes: With normalized data, participation from LMIC was 56% during VMR and 79% during GVMR. NESC represented 73% of participants during VMR and 91% during GVMR. With non-normalized data, 46% of the VMR group were from LMIC, compared with 73% in the GVMR group (χ2=27.87, p<0.001). Additionally, 35% of VMR group were from NESC, compared with 56% in the GVMR group (χ2=17.08, p<0.001).

Discussion: GVMR has a greater penetration than VMR in LMIC. Similarly, non-native English-speakers were also significantly higher in GVMR. Having a dedicated initiative that targets an international audience increases usage in LMIC. Our findings have important implications for the digital education field. Educators who wish to reach learners in LMIC should recruit diverse team members, create relevant content to trainees in these regions, and provide translation services.

Significance of Findings: This is the first prospective study evaluating an ODE intervention targeting learners outside of high-income countries. By fostering a safe and welcoming learning environment, ODE can be effectively used to bridge digital educational disparities and improve access to resources for trainees worldwide, enhance diagnostic reasoning processes, and hopefully reduce global health disparities.

Evaluating the Pancreatic Cancer Diagnosis Pathway in U.S. Veterans

N. Khalaf ¹, K. Choi¹, T. Jamali¹*, F. Kanwal¹, H. Singh¹

*(Presenting Author)

¹Baylor College of Medicine, Houston, TX

Background: Pancreatic cancer is currently the number three cancer killer in the U.S. and is projected to rise to number two in the next decade. Nearly 85% of patients have incurable (locally advanced or metastatic) disease at diagnosis, suggesting need for early cancer detection efforts. We characterized the timeliness of pancreatic cancer diagnosis and treatment.

Methods: We reviewed EHRs of 150 consecutive patients diagnosed with pancreatic adenocarcinoma at a large VA medical center from 1/2015 to 12/2020 with at least 3 years of comprehensive medical data prior to cancer diagnosis. We evaluated clinician notes, referral patterns, diagnostic tests, including radiology and pathology files and determined key time intervals related to diagnosis and treatment. In a subset of 50 randomly selected patients, we further evaluated for presence of known pancreatic cancer risk factors and cancer-associated presenting signs and symptoms.

Results: Of 150 patients, 127 patients presented with symptomatic disease at diagnosis (i.e., not incidentally found on imaging). Average time between onset of first cancer-associated symptom as self-reported by patient and physician documentation of symptom was 68 days (range 1-365 days); onset of first symptom to histological diagnosis (diagnostic interval) was 122 days (range 3-586); and histological diagnosis to treatment initiation (treatment interval, n=46) was 36 days (range 7-92 days). Notably, 40% (61 out of 150) of patients had a diagnostic interval of ≥90 days - a delay known to be associated with lower odds of potentially curative surgical resection. In a subset of 50 records reviewed in detail, most patients (63.3%) had at least 1 known risk factor (e.g., diabetes, current tobacco use, obesity) in combination with a classic warning sign or symptom of pancreatic cancer at the time of diagnosis. Details of initial cancer presentation and setting of care are shown in Table 1.

Conclusion: The average diagnostic delay for pancreatic cancer was ≥90 days, suggesting the presence of opportunities for earlier diagnosis. Most patients had at least 1 known risk factor and a classic warning sign or symptom of pancreatic cancer. As a next step, we will evaluate types of diagnostic process breakdowns in a large national cohort.

Closing the Loop on Radiology Recommendations: Developing and Applying a Taxonomy to Track Performance of Recommended Actions

M. Mirica ¹, T. White², M. Aronson³, S. Sternberg³, U. Shafiq³, S. Berkowitz³, J. Benneyan⁴, R. Phillips³, G. Schiff¹

¹Brigham and Women’s Hospital, Boston, MA

²Harvard Medical School, Boston, MA

³Beth Israel Deaconess Medical, Boston, MA

⁴Northeastern University, Boston, MA

Background: The current lack of reliable systems to classify and track completion of radiology recommendations to ensure reliable diagnosis and follow-up of abnormal findings represents a serious diagnostic patient safety problem. In this study, we report an initiative to comprehensively characterize actionable radiology recommendations, to measure and understand rates and processes of loop closure on such recommendations from findings from radiology in a primary care setting.

Methods: We retrospectively reviewed 532 reports for commonly performed radiology studies ordered by PCPs in 2018 at a large academic center following redesign of radiology reports to include actionable recommendations in a dedicated field. Leveraging these innovative reports, we developed a taxonomy that conceptualized three recommendation domains: a) what is recommended (test, referral, etc.) b) specified timeframe to perform recommended action and c) contingency language qualifying the recommendation (e.g “if a smoker,” “if clinically worsens”). We then tracked whether these recommended diagnostic follow-up actions were completed vs. remained “open loops” in the timeframe radiology recommended. We used detailed chart review to reconcile and exclude open loops where PCP judgement or charted information suggested radiology recommended actions were either not needed or done elsewhere. All open loops were forwarded to PCPs where outstanding safety issues were found.

Results: Using our 3-element framework, we determined there were 462 recommendations for additional radiology studies (230 repeat same study in future; 231 order other studies), 196 non-radiology diagnostic actions (specialty referrals 119, biopsies/invasive diagnostic procedures 47). Of these, 228 had specific time frames in the radiology report, whereas 370 recommendations lacked specific time frames. Contingency language was present in 147 recommendations, with no contingencies in 451. Tracking recommendation outcomes, we found an overall diagnostic loop closure rate of 87.4%. Factors significantly positively associated with loop closure included 1) absence of any accompanying contingency 2) shorter recommended timeframes and 3) evidence of direct communication with PCP.

Conclusions: Failure to “close the loop” on abnormal radiology findings is an ongoing challenge in medical diagnosis whose measurement and improvement resides in characterizing, codifying, and tracking actionable radiology recommendations. Using a dedicated recommendation text field to code follow-up recommendations permits tracking of diagnostic loop closure rates and demonstrated that 1 in 7 of recommendations’ reports were failed to be followed up, even after detailed chart review and accounting for any clinical disagreements with the recommendations by the responsible PCPs.

Defining Diagnostic Error: A Scoping Review to Assess the Impact of the NASEM Report “Improving Diagnosis in Health Care”

T. Giardina ¹, H. Hunte², M. Hill², L. Heimlich², H. Singh¹, K. Smith²

¹Michael E. DeBakey VA Medical Center and Baylor College of Medicine, Houston, TX

²MedStar, Washington, DC

Background: The National Academies of Science, Engineering, and Medicine (NASEM) committee published, Improving Diagnosis in Health Care, and proposed a new definition of diagnostic error, “the failure to (a) establish an accurate and timely explanation of the patient’s health problem(s) or (b) communicate that explanation to the patient”. Almost six years later, it is unclear how the new definition has impacted the study of diagnostic errors. Our scoping review explored how researchers have operationalized the new definition in peer reviewed published literature and establish its impact on this growing field.

Methods: Using the York framework, we identified published literature from October 2015 to February 2021 using Medline and Google Scholar. We also conducted subject matter expert interviews with researchers to explore how they are operationalizing the NASEM definition and its impact on their work and the field in general.

Results: Our search strategy returned 1,076 unique articles; 52 mentioned the NASEM’s definition of diagnostic error. Most articles referencing the NASEM definition did so in the introduction (61.5%), discussion (19.2%), or methods (15.4%). Upon full text screening, 33 potentially eligible studies were identified and 15 were included in the final analysis to determine how diagnostic error was operationalized and measured (see Table 1). Studies were grouped by the theme most closely related to the study focus: Epidemiology, Patients, Measurement, and Clinicians. Seven studies operationalized diagnostic error using previous definitions proposed by experts, two studies used but did not operationalize the NASEM definition, and five studies operationalized the NASEM definition specifically for their study context. SMEs indicated that the NASEM definition functions as a foundation for researchers to conceptualize diagnostic error.

Conclusions: In recent peer reviewed literature, most researchers continue to use pre-NASEM report definitions to operationalize accuracy and timeliness. However, the report has catalyzed use of concepts related to patient-centeredness in the definition of diagnostic error, and studies focused specifically on communication of a diagnosis to patients are now emerging.

Examining Variation in Time-to-diagnosis and Symptomatic Presentation of Endometriosis

E. Whitfield ¹, M. E. Barclay¹, G. Lyratzopoulos¹

¹University College London, London, UK

Background: Endometriosis affects women of reproductive age. UK patient survey data suggests a mean time-to-diagnosis (from first symptomatic presentation) of 5.7 years. Symptoms can vary from abdominopelvic pain and heavy menstrual bleeding to infertility. Formal evaluation of diagnostic intervals in population-based data is limited. We examined variation in symptomatic presentation and time-to-diagnosis for endometriosis patients.

Methods: For 9,413 patients with a recorded diagnosis of endometriosis in Clinical Practice Research Datalink data, consultation records from up to 10 years prior to each patient’s diagnosis date were extracted. Codes indicating possible symptoms were identified and grouped into features, for example infertility or painful menstruation. Age, presenting symptom (from the features identified), and time-to-diagnosis at first presentation were recorded for each symptomatic patient. A random sample of 500 symptomatic women formed a ‘baseline group’ - their median time-to-diagnosis was 3.7 years. Differences in time-to-diagnosis between the baseline group and women of different ages at first presentation (under 25s, 26-30s, 31-35s, 36-40s and over 40s) were examined. Further, a repeated sampling approach was used to determine which presenting symptoms were associated with significantly different time-to-diagnosis compared to the baseline group.

Results: From the 29 symptoms we identified, 8,164 (86.7%) patients reported at least one symptom up to 10 years prior to diagnosis. Amongst the symptomatic cohort, the most frequently reported symptoms were abdominal pain (61%), mental health symptoms (34%) and irregular menstruation (32%). Symptomatic patients had medians of 7 distinct symptomatic visits (IQR [3, 13]) and 4 distinct symptoms (IQR [2, 7]) prior to diagnosis date. Median age at first symptomatic presentation was 33 (IQR [28, 38]). Time-to-diagnosis differed significantly from the baseline group for those who first presented either under the age of 25 (median 6.3 years, 95% CI 5.7 - 6.7) or over the age of 40 (median 2.5 years, 95% CI 2.2 - 2.7). Five symptoms had significantly different intervals in over 90% of the iterations: dysmenorrhea, infertility, pelvic pain, menorrhagia, and vomiting.

Conclusion: These findings substantiate the presence of long diagnostic intervals in endometriosis, signaling potential for improvements in the diagnostic process. Patients present with a wide range of possible symptoms, often multiple times before diagnosis. Time-to-diagnosis for endometriosis varies significantly with both age and presenting symptom, indicating that both patient-level and disease factors are responsible for variation in time-to-diagnosis.

Unsupervised and Semi-Supervised Cluster Analysis for Detection of Diagnostic Delays in Large Longitudinal Data Sources

A. Miller ¹, A. Arakkal¹, S. Koeneman¹, J. Cavanaugh¹, P. Polgreen¹

¹University of Iowa, Iowa City, IA

Background: Most methods to identify a potential diagnostic delay rely on a priori expert-specified criteria to define delay. These expert-derived criteria usually include symptoms or diagnoses that can be used to identify potential missed opportunities during healthcare visits prior to the actual disease diagnosis. However, experts may provide incomplete criteria to define potential diagnostic delays, leading to an under-identification of potential diagnostic opportunities. Machine-learning approaches may be used to recommend symptoms or diagnoses that match the expected pattern of diagnostic delays. The objective of this study was to pilot test unsupervised and semi-supervised clustering approaches to detect patterns in longitudinal healthcare visits suggestive of missed diagnostic opportunities.

Methods: We considered three diseases of interest - tuberculosis, endocarditis, and acute myocardial infarction. For each disease, we labelled a preliminary set of symptomatically-similar diagnoses (SSDs) commonly associated with a missed opportunity. For example, before tuberculosis, symptoms such as cough or dyspnea and alternative diagnoses such as pneumonia or asthma may occur. Using the IBM MarketScan databases from 2001-2018, we identified patients with an index diagnosis of each disease and all diagnoses in the year prior. For every ICD-9-CM/ICD-10-CM code we computed the number of patients with the diagnosis each day prior to the index disease. We fit time-series models to each diagnosis code and used the parameter estimates to cluster diagnosis codes into similar groups. For our unsupervised approach we considered k-means clustering. For our semi-supervised approach, we used a semi-supervised k-means algorithm seeded by our previously identified SSDs.

Results: In general, our unsupervised approach was able to separate diagnostic patterns into clinically meaningful clusters. In particular, clusters could be organized into one of three types: (1) a biologically plausible diagnostic opportunity, (2) a risk factor for the disease (e.g., diabetes prior to tuberculosis) or (3) unrelated conditions. Our semi-supervised approach consistently resulted in clusters where biologically plausible conditions were added to the previously identified SSD or risk factor clusters. As an example, for tuberculosis, when clusters were initially seeded with cough, dyspnea, asthma and pneumonia, the unlabeled diagnoses of fever and weight loss were consistently added to resulting clusters.

Discussion: Our cluster-based approach appears to provide a valid way to identify and recommend longitudinal healthcare patterns indicative of a diagnostic delay. This approach may be useful to identify clusters of symptoms or diagnoses suggestive of a diagnostic opportunity or to augment an established list of criteria defining a diagnostic opportunity.

Acute on Chronic Lithium Toxicity

D. Amin ¹

¹Cook County Stroger Hospital, Chicago, IL

Learning Objectives: • Describe three distinct patterns of lithium toxicity. • Analyze an acute on chronic presentation of lithium toxicity. • Appraise the utility of initial levels of lithium in a clinically symptomatic patient.

Case: 62-year-old male with bipolar disorder on lithium presented to the hospital confused, with tremors, and diarrhea and vomiting. He was in the hospital the day prior for worsening mania and discharged with prescription for lithium. VS: BP 115/50 HR 65 RR 20 Sa02 99%. On examination, the patient appeared tremulous, oriented only to person, and his speech was slurred. Laboratory results showed normal electrolytes and renal function. The serum lithium level was 2.2 mEq/L. Patient received IV fluids and emergent hemodialysis with no improvement and repeat serum lithium level of 3.6 mEq/L. Patient received five more rounds of hemodialysis before resolution of his neurological symptoms and therapeutic levels of serum lithium.

Discussion: Lithium has been used as the gold standard in the treatment of major depressive and bipolar disorders for decades. Though its narrow therapeutic index makes toxic side effects common, lithium remains one of the most widely used drugs. Toxic effects can be classified into acute, chronic, and acute-on-chronic. Acute toxicity is an overdose in a lithium naïve patient and blood levels can be high, but GI symptoms predominate due to delayed absorption into the brain. Chronic toxicity occurs when the lithium level rises despite a patient’s adherence to prescribed dosing. Lastly, acute-on-chronic toxicity is acute ingestion in a patient who is chronically on lithium. Brain lithium concentration rise to dangerous levels and there is rapid onset of neurological toxicity. As is evident, owing to its diverse effects, lithium toxicity can easily be missed if a proper history and exam are not conducted. Diagnosis is also complicated by the fact that serum levels correlate poorly with the severity of clinical toxicity.

A Case of Hyperbilirubinemia from E. Coli Meningitis Despite Negative PCR Testing

O. Bawmann ¹, M. Crenshaw¹, R. Levy¹, J. Barry¹

¹University of Colorado School of Medicine, Aurora, CO

Learning Objectives: 1) Highlight a potential diagnostic pitfall with polymerase chain reaction (PCR) testing of cerebrospinal fluid (CSF) for E. Coli meningitis. 2) Discuss importance of the K1 capsule in pathogenicity of E. Coli CNS infections.

Case Information: A former 30-week gestational age boy developed hyperbilirubinemia - initially considered physiologic jaundice - requiring phototherapy until day of life three. On day of life nine, he was noted to be jaundiced and had a total bilirubin of 14.8mg/dL, direct of 6.4mg/dL; seven hours later, they increased to 18.2mg/dL and 9.4mg/dL. Initial work up revealed a high procalcitonin (7.94ng/mL). When two sets of blood cultures were positive for E. Coli, a lumbar puncture found low CSF glucose (13mg/dL), high CSF protein (384mg/dL), and a neutrophilic pleocytosis (1025 10ˆ6/L, 92% neutrophils). However, the CSF meningitis-encephalitis panel (MEP) was negative, including for E. Coli. This patient remained well-appearing on exam without fontanelle fullness, apnea, bradycardia, or other vital sign abnormalities. He was treated with a 21-day course of ceftazidime and a 14-day course of ursodiol with resolution of his hyperbilirubinemia.

Discussion: This case highlights a clinical scenario in which CSF PCR testing may be falsely negative for E. Coli. The MEP targets the K1 capsular polysaccharide on E. Coli. While K1-positive E. Coli is the predominant pathogenic strain, K1-negative strains can also cause disease. Clinicians should not be falsely reassured by a negative MEP when there is other laboratory or clinical evidence of meningitis, as described in this case. Clinical studies have demonstrated that the K1-negative strain is less pathogenic, including less likely to cause CNS disease. This could explain why our patient was well-appearing without exam or vital sign abnormalities, instead of following the typical illness script of E. Coli meningitis in a neonate, which often includes vital sign instability, apnea, and feeding intolerance.

Evolving Low Back Pain: Delayed Diagnosis in New Era of Telemedicine

N. Breitnauer ¹, J. Venci¹

¹Denver Health, Denver, CO

Learning Objectives: 1. Highlight limitations of telemedicine visits 2. Identify potential biases that can occur during telemedicine visits 3. Recognize the importance of new knowledge and skills acquisition in a telemedicine era

Case Information: A 37 year-old woman with a past medical history of major depressive disorder and benzodiazepine substance use disorder in early remission had a telemedicine (phone only) appointment for new low back pain with symptoms consistent with unilateral sciatica. Conservative treatment of non-steroidal anti-inflammatory, ice, and exercises discussed. 6 weeks later patient had second telemedicine visit and described ongoing pain, now with bilateral sciatica symptoms. Denied symptoms consistent with cauda equina. Xray of lumbar spine performed and short course of tramadol given. Patient called 3 weeks later with 2-week history of progressive bowel and bladder incontinence. Advised to go to emergency room where she was found to have diffuse large B-cell lymphoma, stage IV with several levels of lumbar spine boney erosion.

Discussion: This case highlights a delayed diagnosis error early in the SARS-CoV-2 global pandemic when the majority of clinical visits abruptly shifted from in-person to telemedicine visits. Confirmation and anchoring biases are demonstrated due to limited information (i.e. no physical exam) and contributed to the delayed diagnosis. Data is equivocal on quality of care provided in the telemedicine domain prior to the SARS-CoV-2 global pandemic1,2. As we enter a post-vaccine phase of this pandemic, both office and telemedicine visits may be available to patients. Outpatient medical groups must decide how to effectively triage patients for in person office visits versus telemedicine visits. In addition, clinicians must recognize how cognitive biases during telemedicine visits can lead to a delayed diagnosis. To avoid harm, clinicians must acknowledge these risks and seek opportunities to advance knowledge and skills in this rapidly developing era of telemedicine.

1Whitten, PS et al. “Systemic review of cost effectiveness studies of telemedicine interventions,” British Medical Journal (2002); 324:1434-1437 2Ekeland, A. et al. “Effectiveness of telemedicine: A systemic review of reviews,” Int. Journal of Medical Informatics (2010); 79(11): 736-771.

Imaging for Right Upper Quadrant Pain

M. Byers ¹, M.D., J. Triozzi¹, S. Bujarski¹

¹Baylor College of Medicine, Houston, TX

Learning Objectives: 1. Diagnosis of acute cholecystitis vs. cholangitis 2. Sensitivity and specificity of common imaging

Case Information: A 72-year-old presented with fever, right upper quadrant abdominal pain. Studies revealed leukocytosis, hyperbilirubinemia and cholestatic pattern of transaminases. Computed tomography (CT) of the abdomen showed cholelithiasis without cholecystitis. Right upper quadrant ultrasound reported pericholecystic fluid but was inconclusive and recommended hepatobiliary iminodiacetic acid (HIDA) scan. After giving cholecystokinin the patient developed anaphylactic shock and later underwent cholecystostomy decompression. A magnetic resonance cholangiopancreatography (MRCP) performed did not demonstrate biliary duct dilation and he underwent cholecystectomy.

Discussion: Overuse of imaging is common. This patient with cholangitis experienced a rare, life-threatening reaction to a study that was unnecessary. The 2018 Tokyo Guidelines provide an approach to diagnosis of acute cholangitis using a triad of clinical, laboratory, and imaging findings.1, 2 Ultrasonography (US) is first-line for acute right upper quadrant pain with high sensitivity (0.88) and specificity (0.8) for acute cholecystitis but low sensitivity (0.42) for common bile duct (CBD) dilation. CT is second-line based on availability and improved statistical profile. HIDA scan is used when the diagnosis of acute cholecystitis remains uncertain after US given its higher sensitivity (0.97) and specificity (0.9) for gallbladder stone but is ineffective for evaluating the CBD3. MRCP is used if the diagnosis remains uncertain and initial workup suggests a diagnosis other than acute cholecystitis such as choledocolithasis with high sensitivity (0.95) and specificity (0.9) 4. Imaging should add to diagnosis and management. Because complications are rare, the needlessness of many studies is not palpable until we experience a teachable moment.

Does a Bottle of Calcium Carbonate a Day Keep the Doctor Away?

C. Carver ¹

¹University of Colorado Hospital - Anschutz Medical Campus, Denver, CO

Learning Objectives: To recognize the triad of laboratory abnormalities seen in milk alkali syndrome to aid in your diagnostic approach to a patient with a new diagnosis of hypercalcemia. To consider disparities that may contribute to this unique presentation.

Case: The patient is a 43-year-old man with a past medical history of Celiac disease and opioid use disorder (OUD) currently experiencing homelessness. He presented to the hospital due to nausea and vomiting in the setting of opioid withdrawal. On the basic metabolic panel, the patient had hypokalemia to 2.8, carbon dioxide of 33, BUN of 26, creatinine of 2.05, and calcium of 14.1 and was admitted to medicine for evaluation for hypercalcemia. Patient was initially non-participatory in the interview due to discomfort from withdrawal symptoms. As a surrogate for history-taking, a PTH, PTHrP, Vitamin D, TSH, SPEP/UPEP were all ordered. When re-interviewed after receiving methadone dose, patient endorses that he was taking a bottle of over-the-counter calcium carbonate daily. Due to his housing insecurity, he had significant difficulty with adherence to a gluten-free diet but combining food with calcium carbonate tablets allowed him to tolerate gluten-containing food without gastrointestinal side effects.

Discussion: The initial evaluation for hypercalcemia typically distinguishes between the two most common causes: primary hyperparathyroidism and humoral hypercalcemia of malignancy. However, milk-alkali syndrome is now considered to be the third most common cause, due to the wide availability of calcium carbonate for use in both osteoporosis as well as peptic ulcer disease. The triad of laboratory abnormalities consists of hypercalcemia, metabolic alkalosis, and kidney injury, which may be irreversible. Diagnosis only requires a carefully taken patient history and a basic metabolic panel. Medarov BI. Milk-alkali syndrome. Mayo Clin Proc. 2009 Mar;84(3):261-7. doi: 10.1016/S0025-6196(1161144-0). PMID: 19252114; PMCID: PMC2664604.

A Stroke of Hypoglycemia: Diagnosing an Insulinoma with Stroke-like Presentation

L. Chun ¹, J. Bu¹, L. Chun¹, E. Wang¹, A. Fuentes Baldarrago¹, B. Lovetro¹

¹University of California San Diego School of Medicine, San Diego, CA

Learning Objectives: (1) To consider atypical presentations of insulinoma. (2) To recognize the utility of a 72-hour fast as a diagnostic tool for hypoglycemia.

Case Information: A 58-year-old man with hypertension presented to the ED with acute onset dysarthria, blurred vision, and right-sided weakness. A code stroke was called, with a negative CT head and CTA head/neck. Blood glucose was 57 mg/dL on arrival and increased to 107 mg/dL without intervention. On admission to the floor, he was hypertensive to 151/88 mm Hg. Physical exam was remarkable for thin body habitus. Labs were notable for glucose 60 mg/dL and sodium 132 mMol/L. Although the patient had never experienced dysarthria or motor impairment, he had experienced episodes of nausea, flushing, and unsteadiness for 15 years. Additional stroke workup was unrevealing. A diagnostic 72-hour fast was conducted to assess his recurrent hypoglycemia. When the patient’s serum glucose was 30 mg/dL, his labs were remarkable for proinsulin 329.7 pmol/L, insulin 22.0 uIU/mL, C-peptide 4.77 ng/mL, and beta-hydroxybutyrate 0.12 mMol/L, suggesting a high endogenous insulin state. MRI of the abdomen confirmed a 1 x 0.8 cm hyperintensity at the pancreatic head, consistent with insulinoma.

Discussion: Insulinomas are functional neuroendocrine tumors arising from pancreatic beta cells. Although symptomatic insulinoma most often presents with palpitations, tremulousness, diaphoresis, and confusion, this patient presented with focal neurologic deficits and classic stroke symptoms. As a result, he underwent a full neurologic workup before the final diagnosis was made. Similarly, up to 20% of insulinoma patients are misdiagnosed with neurologic or psychologic conditions before insulinoma is diagnosed. This case demonstrates the importance of considering the wide spectrum of neuroglycopenic symptoms of insulinomas, ranging from psychiatric disturbances and seizures to dysarthria and ataxia.

Impact of Implicit Bias and Systemic Racism on the Delayed Diagnosis of Diabetes

B. Theodore ¹, S. Ferguson¹

¹University of Wisconsin SMPH, Madison, WI

Learning Objectives: 1. Implicit bias and systemic racism in ambulatory transitions impact health disparities 2. Patient portal use can widen health disparities if principles of health literacy are not embedded

Case Description: A 47-year-old black woman with obesity was referred to rheumatology for established diagnosis of lupus. She experienced polyuria, polydipsia, chronic nausea, and abdominal pain for months, and was diagnosed with bladder instability. She had normal upper and lower endoscopies. Over 11 months, 5 lab results met criteria for diabetes, but she was not diagnosed until rheumatology visit. After beginning diabetic treatment, symptoms resolved.

Discussion: Diabetes was missed by multiple providers despite suggestive symptoms. Patient instructions during ambulatory transitions failed to advise follow up for hyperglycemia. The patient noticed abnormal labs in the portal but was not concerned based on interpretation and sign off by providers. She referenced race concordance with the medical student and rheumatologist as helping her “be heard”. One provider disclosed belief that she was non-adherent to treatment, an example of implicit bias delaying diagnosis. As patient portal use grows, awareness of the potential unintended propagation of systemic racism and for worsened health disparities for black patients is critical. Further work is needed to implement health literacy principles in the interpretation of results, with opportunity for standardization of language as a potential lever for improving health disparities. In addition to implicit bias training and workforce diversification, developing standardized workflow to support abnormal results during ambulatory transitions is likely to improve health disparities.

The Crowned Dens Syndrome: A Rare Cause of Neck Pain and Fever

A. Johnson ¹, D. Ramadurai², C. Selph¹

¹University of Colorado, Denver, CO

²University of Pennsylvania, Philadelphia, PA

Learning Objectives: Crowned dens syndrome (CDS) is caused by calcium pyrophosphate deposition and presents with neck pain and elevated inflammatory markers. CDS is diagnosed by characteristic CT findings. Treatment with NSAIDs, colchicine, or steroids results in rapid clinical improvement.

Case Information: A 76-year-old woman with a past medical history of chronic kidney disease (CKD) and chronic hypomagnesemia presented with one week of neck pain and fevers. On exam, she was febrile and tachycardic with full active and passive range of motion of the neck, no nuchal rigidity or neurologic deficits. Labs were notable for elevated leukocytes, erythrocyte sedimentation rate (ESR) >80 mm/h and C-reactive protein (CRP) of 96.1 mg/L. Empiric antibiotics for meningitis were initiated. CT of the head was unremarkable and lumbar puncture revealed clear cerebrospinal fluid with no cells and normal protein. Closer review of a CT of the neck revealed calcium pyrophosphate deposition along the atlanto-axial joint, diagnosing crowned dens syndrome. Antibiotics were discontinued and prednisone was initiated with prompt resolution of her neck pain.

Discussion: Crowned dens syndrome (CDS) is an inflammatory condition due to calcium pyrophosphate deposition in ligaments surrounding the dens. It often presents with neck pain, fever, and elevated inflammatory markers. CDS has a higher predominance in older individuals; our patient also had other risk factors including CKD and disorders of magnesium. It is important to consider and recognize CDS, given the significantly different management compared to other entities that present similarly, such as meningitis and temporal arteritis. In the case of this patient, the diagnosis was missed on an admission CT of the neck which resulted in additional antibiotic exposure, a lumbar puncture, and delayed initiation of treatment. Treatment of CDS with steroids, NSAIDs, or colchicine should invoke rapid resolution of symptoms, as seen in our patient.

“Copy Paste” Diagnostics Causing Delays

D. Kirelik ¹, C. Duarte¹

¹University of Colorado, Aurora, CO

Learning Objectives: 1. Depression should be a diagnosis of exclusion for weight loss in the geriatric population 2. When a patient is admitted multiple times for the same diagnosis, we recommend a cognitive reset to critically evaluate for the presence of bias affecting medical decision making

Case Information: The patient was a 77-year-old African American woman with type II diabetes, and hypothyroidism who had four hospital admissions for failure to thrive over 6 months. She had a rapid decline in functional status with a documented weight loss of 40lbs. During her hospitalizations, multiple lab abnormalities were noted, including anemia, intermittent hypercalcemia, and a protein gap leading to a diagnosis of MGUS via SPEP. Additionally, her blood sugar became labile with hemoglobin A1C increasing from 7.3% to 9.6%. Despite her lab abnormalities and evidence of catabolic weight loss, a major focus of her admissions was “failure to thrive” largely attributed to depression with involvement of the psychiatry team. On a subsequent admission, the team took a cognitive pause and holistically re-evaluated her case. This led to a workup for other causes of her worsening diabetic control resulting in a diagnosis of latent autoimmune diabetes in adult (LADA) with positive GAD and anti-islet cell antibodies. Additionally, given her complaints of back pain with a MGUS diagnosis, the diagnosis of multiple myeloma was further considered by the team and was proven on subsequent bone marrow biopsy.

Discussion: This case demonstrates multiple diagnostic issues. There was implicit bias based on her age, race and gender delaying diagnostic workup. Providers attributed her symptoms to “failure to thrive” and depression despite ongoing workup for MGUS, multiple lab abnormalities, and worsening diabetic control. There was diagnostic momentum with providers who carried forward this diagnosis to subsequent admissions. It took a conscious shift by a new team to reevaluate of her symptoms and send appropriate workup. Depression should be a diagnosis of exclusion, especially in the context of objective lab abnormalities in elderly patients.

An Antidote to Cognitive Bias – Know Thy Patient

B. Krauss ¹

¹City University of New York School of Public Health, New York City, NY

Learning Objectives: Representativeness bias is well-documented for aging patients. Representativeness is often accompanied by other biases that “fit” an initial picture of the patient. Social psychology’s rich literature on addressing stereotypes will be applied, particularly work on social identity and cross-categorization of patient, provider, and family identities.

Case Information: A 76-year-old man, with a history of esophageal ulcers and low hemoglobin, was admitted to hospitals on three occasions, first by his primary care doctor “to do tests to find out what is going on,” and subsequently as his condition deteriorated. Many symptoms—weakness, confusion, fatigue, “crankiness”—were attributed to his age. There was premature closure on the initial diagnosis of bleeding ulcers. During three hospitalizations the patient was seen by nine different doctors on rotation. Family was avoided during rounds. Family protests that “something else is going on” went unheard. The primary care physician’s calls to the hospital doctor in charge of the case were not returned. A correct diagnosis was made by a hospice physician, days before the death of the patient, and confirmed by autopsy.

Discussion: How would the process of diagnosis have changed with acknowledgement of the following life details? The patient regularly hiked 2 miles, rowed on a rowing machine for 40 minutes, and swam 40 minutes per day. He consulted with faculty and students at University of Arizona every week on methodological issues. He had over 100 journal and book publications. His content and statistical consultation guided the data supporting the Americans with Disabilities Act. He was head of the psychology department at Pace University before retiring, and mounted community and responder bereavement interventions since Pace was within blocks of the World Trade Center Disaster. And one medical detail: on the second hospitalization his ulcers were found to be infected with an unusual virus.

Diabetes, Drinking, and Dirt: A Cautionary Tale of Disseminated Nocardiosis

C. Levandowski ¹, A. Buehler¹, J. Lessing¹, R. Hanna¹

¹University of Colorado, Boulder, CO

Learning Objectives: • Identify risk factors for nocardiosis noting that 1/3 of patients will not be overtly immunocompromised. • Recognize nocardiosis as a possible cause of SIADH. • Recognize nocardiosis constellation of symptoms.

Case Information: A 72-year-old male soil excavator with controlled type-2 diabetes and alcohol use disorder presented with two months of progressive dyspnea. Evaluation revealed acute hypoxic respiratory failure with multifocal pneumonia and sodium of 119 mmol/L determined secondary to syndrome of inappropriate anti-diuretic hormone (SIADH). He was discharged after treatment for pneumonia but returned with worsening symptoms. During hospitalization he developed seizures, which led to imaging that revealed numerous brain micro-abscesses, cavitary lung nodules, adrenal gland abscess, and later chorioretinitis. Biopsy and cultures diagnosed disseminated Nocardia beijingensis. Despite 10 weeks of appropriate antibiotics (including imipenem, minocycline, ceftriaxone, trimethoprim-sulfamethoxazole, linezolid), his condition deteriorated, and family opted to transition to hospice.

Discussion: Nocardiosis is a disease caused by the gram-positive bacterium Nocardia, found in soil. While typically considered in the differential for the immunocompromised, up to one-third of patients are non-immunocompromised. As in this case, it is important to note that many common conditions, such as diabetes, heavy alcohol use, and advanced age, lead to a sufficient relative-immunocompromise that should prompt inclusion of nocardiosis in the differential. In-depth history-taking often reveals risk factors including working near soil. Pulmonary involvement is most common (71% of cases) and symptoms are nonspecific - cough, dyspnea, and sputum production. Dissemination is reported in 11% of cases primarily involving the brain with neurologic symptoms. SIADH has diverse etiologies, and therefore often discounted. As in this case, SIADH has been reported as a presenting symptom in numerous case reports of both pulmonary and disseminated nocardiosis: Persistent SIADH in the context of non-resolving pneumonia should prompt consideration of nocardiosis. Early diagnosis and initiation of treatment is necessary to improve outcomes.

Hyperkalemia Succs! A Case of Succinylcholine Induced Hyperkalemia

S. Maltby ¹, M. Frank¹

¹University of Colorado, Aurora, CO

Learning Objectives: 1. Recognize importance of assessing medications as causative agents for our patient’s conditions 2. Review characteristics of succinylcholine induced Hyperkalemia

Case Information: A 70 year-old woman was admitted after elective atrial fibrillation ablation. She has history of compensated diastolic heart failure, coronary artery disease, well-controlled diabetes mellitus, morbid obesity, gastroesophageal reflux and hypertension. Her home medications included amlodipine, aspirin, atorvastatin, apixaban, lisinopril, insulin, metoprolol and omeprazole. Laboratories checked pre-procedure were non-contributory and included a basic metabolic panel [serum Potassium (K) of 4.3 meq/L], complete blood count and INR. Twenty-four hours post-procedure her K was 5.9 meq/L. Work-up showed no evidence of hemolysis or renal failure and normal Creatine-kinase. It was noted that patient received succinylcholine during the procedure; however, no K documented between pre-procedure and 24-hours post-procedure. Lisinopril and colchicine, which was initiated for post-procedure pericarditis, were discontinued as are known causes of hyperkalemia (hyperkalemia was identified before initiation of colchicine). Patient received kayexalate. Hyperkalemia resolved within 3 days and she was discharged home. Hyperkalemia has since not recurred.

Discussion: Hyperkalemia is common in hospitalized patients and can represent a medical emergency. Determining the cause of hyperkalemia is key to prevent reoccurrence. This case showcases a patient with an uncommon cause of hyperkalemia, succinylcholine administration. Succinylcholine-related-hyperkalemia usually presents immediately after administration and can persist for up to 2 weeks. It is associated to depolarization of acetylcholine receptors in the neuromuscular junction, resulting in an efflux of intracellular potassium. This case highlights the importance of a thorough history including an extensive review of the medications, as many medications can result in hyperkalemia through a diverse array of mechanisms. It also emphasizes the importance of inter-professional collaboration and a team approach to patient care. Undergoing a comprehensive medication review should be considered an essential part of our history taking and including in our critical decision making as it can be key in discovering medication -related conditions.

Typhoid Fever: When Prior Antibiotic Use Contributes to a Delayed Diagnosis

S. Mucheli ¹

¹Tan Tock Seng Hospital, Singapore

Learning Objectives: In a returning traveler with fever, it is important to consider differential diagnoses based on the travel itinerary. Typhoidal salmonella may partially respond to certain antibiotics used empirically in treating common bacterial infections. This may further delay the time to diagnosis.

Case Information: A 30-year-old lady presented to the Emergency department (of a tertiary hospital in Singapore) with a 10-day history of fever with chills, epigastric pain, anorexia and episodes of vomiting. She had returned from Yangon, Myanmar 4 weeks prior to the onset of symptoms. She was diagnosed with Urinary Tract infection (UTI) based on positive leucocyte esterase and nitrites on Urine dipstick in spite of no urinary symptoms and given empiric oral amoxicillin-clavulanate. She was hospitalized 2 days later in view of persistent symptoms. A computerized tomography scan of the Abdomen Pelvis did not show any abnormality. Her blood and urine cultures returned negative. She was discharged shortly to complete a 7-day course of oral amoxicillin-clavulanate as she had shown partial response. She presented again 13days later with fever and chills of 3days duration. Repeat blood cultures were performed, which grew Salmonella typhi. She was treated with IV ceftriaxone and responded well. It is important to note that more than 90 per cent of Enteric fever cases in Singapore are imported.

Discussion: This case highlights a delayed diagnosis of Typhoid fever in a returning traveler, possibly due to prior treatment with oral antibiotics. This patient’s demographics and a positive urine dipstick anchored the initial team to favour UTI, instead of Typhoid fever. Lack of knowledge with regards to the pretest probability for potential diagnoses in a returning traveler may have contributed to the bias of premature closure. Typhoid fever may respond partially to commonly prescribed oral antibiotics in the outpatient setting (e.g. Ampicillin, Trimethoprim-sulfamethoxazole, Fluoroquinolones, Azithromycin depending on the antibiotic susceptibility of the pathogen). This might potentially delay diagnosis as the antibiotic doses and duration may be suboptimal to achieve cure.

Tuberculous Meningitis in an Immunocompetent Patient

l. Paykargar ¹, M. Sulieman¹, R. Ahmed¹

¹Emory Saint Joseph Hospital, Atlanta, GA

Learning Objectives: To highlight the importance of making a presumptive diagnosis of tuberous meningitis in low-risk individuals

Case: A 69-year-old healthy female, born in Greece and lived most of her life in the US. Presented with fever, and confusion for one week. CT chest revealed Evidence of prior granulomatous disease. Initially her CSF was most concerning for Bacterial Meningitis, treatment with antibacterial, antiviral, and steroids for one week, however patient did not improve evidence by worsening mental status and SCF parameters. MRI brain showed obstructive hydrocephalus that required CSF diversion. CSF cultures, extensive viral and fungal studies, HIV and PRC Mycobacterium tuberculosis were negative. She continued to decline and died 13 days after presentation. Mycobacterium tuberculosis was isolated in CSF culture 4 weeks postmortem. Neuropathology of brain autopsy is currently pending.

Discussion: Unrecognized Tuberculous meningitis (TBM) is fatal. 7,163 Tuberculosis (TB) cases were reported in the US in 2020, 71% among individuals born in endemic areas [1]. TBM is rarely reported in immunocompetent patients. TBM is more indolent and presents with headache and confusion preceded by a prodrome of fever and malaise [2]. CSF findings in TBM include pleocytosis with lymphocytic predominance, high protein, and low glucose levels. AFB stains are infrequently positive and TB culture takes weeks to result. nucleic acid amplification has high sensitivity and is the preferred diagnostic test [3]. Evidence of pulmonary lesions is often found in chest images [2]. In our patient, granulomatous changes in images hint at possible prior infection. Obstructive hydrocephalus demonstrated in this case is well documented as a sequel of TBM [4].

Break for Bias

Al. Rezigh ¹, Au. Rezigh²

¹Baylor College of Medicine, Houston, TX

²UT Southwestern Medical School, Dallas, TX

Objectives: 1. Recognize fat embolism syndrome. 2. Note how the COVID-19 pandemic can contribute to cognitive bias.

Case: A 20 year-old man presented with right leg pain following a motor vehicle collision. On arrival, his vital signs were unremarkable. His physical examination was notable for tenderness of the right hip and gross deformity of the right ankle. Initial laboratory evaluation revealed a WBC count of 19,800, but no other abnormalities. CT scans demonstrated a right acetabular fracture. X-rays showed a pilon (distal tibia) fracture. Screening SARS-CoV-2 testing returned positive. He was admitted and underwent external fixation of the right ankle. Two days later, a rapid response was called. SpO2 was 70% on room air. On exam, he was in mild distress with clear lungs and post-surgical changes of the right ankle. Over 30 minutes his oxygen requirement escalated from 2L to 15L. Chest x-ray revealed new bilateral multifocal opacities. While his decompensation was initially attributed to SARS-CoV-2, the team reassessed and noted the complete absence of viral symptoms and lack of any findings on initial CT chest to be atypical. A CT PE study was obtained revealing bilateral dense consolidations and subpleural ground glass opacities felt to be most consistent with fat embolism syndrome. He was managed conservatively and recovered.

Discussion: Fat embolism syndrome (FES) is defined as those at risk for fat emboli presenting with respiratory distress, petechial rash of non-dependent areas, or neurologic symptoms. The pathophysiology is unknown, though there are two proposed mechanisms. The mechanical theory, where fat globules released from damaged tissue lodge in a distal vascular bed, and the biochemical theory, where end organ damage occurs secondary to the inflammatory and prothrombotic response induced by the presence and breakdown of circulating fat. FES typically manifests 24 to 72 hours after the initial insult. The classic “triad” need not all be present. Pulmonary symptoms typically occur earliest and can range from insidious dyspnea to ARDS. Neurologic manifestations can range from transient confusion to seizures and focal deficits. Other manifestations include anemia, thrombocytopenia, fever, DIC, and myocardial dysfunction. Diagnosis is clinical, as imaging findings can be highly variable. Treatment is supportive. Prognosis is generally good, with most recovering spontaneously. In the current pandemic, SARS-CoV-2 infection should be considered in patients with dramatic respiratory decline. However, we must remain vigilant to avoid premature closure when approaching patients with possible overlapping symptoms, even in those who have tested positive.

How We Define Diagnostic Error in Comfort-Focused Settings: A Case of Missed Coccidioides

E. Stein ¹, L. Foster², J. Lessing², T. Anstett²

¹University of Colorado Internal Medicine Residency Program, Aurora, CO

²University of Colorado School of Medicine, Aurora, CO

Learning Objectives: 1) Appropriately frame missed diagnoses leading to death in the context of comfort-focused care plans. 2) Recognize Coccidioides infections occur beyond the Southwest United States.

Case Information: An 82-year-old woman from Mexico with a renal transplant presented to a Colorado hospital with chills, dyspnea and urinary frequency. Prior to admission, there were signs of transplant rejection and suspicion for malignancy based on pulmonary nodules. She had expressed preference for a comfort focused care plan, though had not discussed specific goals regarding hospitalization. During hospitalization, she developed worsening hypoxia and encephalopathy. Chest imaging revealed new multifocal nodular opacities. Her condition deteriorated despite broad-spectrum antibiotics and stopping immunosuppression. The patient transitioned to comfort-focused care and died within hours. After death, cultures and serology returned positive for Coccidioides immitis/posadasii, consistent with disseminated disease and a missed diagnosis.

Discussion: The area endemic for Coccidioides is classically described as the Southwest United States. However, the distribution area is evolving (PMID: 32159021), and coccidiomycosis should be considered in any immunocompromised patient with pulmonary symptoms, nodules, or unexplained decline. This case of missed Coccidioides raises important questions about how we define diagnostic error. In a patient whose wishes align with a comfort-focused approach, we posit that death itself should not necessarily be viewed as an adverse outcome. The idea of a “good death” is well accepted, often incorporating themes of symptom control, dignity, and self-determination. If death is not an adverse outcome, perhaps a missed diagnosis leading to that death should not be framed as harm. Patients who have adopted a comfort-focused care plan have varying views on whether an error that shortens life constitutes harm (PMID: 27515997). Several variables should be explored in understanding missed diagnoses at the end of life, primarily the patient’s values and wishes. Taking these into account allows for appropriate framing of diagnostic error in this challenging context.

Legionella Pneumonia with Complicated Parapneumonic Effusion

J. Tram ¹, A. Kumar¹, M. Jamshidian¹, T. Moinizandi¹

¹University of California, San Diego School of Medicine, San Diego, CA

Learning Objectives: Recognize an atypical presentation of Legionella infection and risk factors.

Case Information: A 55 year-old man presented with several episodes of “liquid black” stools and nearly two weeks of subjective weight loss and headaches. Esophagogastroduodenoscopy (EGD) demonstrated several pyloric, duodenal and antrum ulcers that were cauterized. Helicobacter pylori antibodies were negative. Routine thoracic CT in the emergency department was remarkable for an incidental finding of right upper lung lobe consolidation and loculated pleural effusion. Infectious workup including Aspergillus, Legionella, Cryptococcus and Tuberculosis was completed with only urine Legionella antigen positive. Thoracentesis demonstrated an exudative pleural effusion with neutrophil predominance. The patient was treated with Azithromycin and three doses of intrapleural Dornase alfa and tPA. At time of discharge, the patient’s cough had improved, and repeat chest CT revealed near-complete resolution of the pleural effusion.

Discussion: This is an atypical presentation of Legionella pneumonia that began with a chief complaint of melena. Legionella infection has been associated with stress ulcers and is the most likely cause of this patient’s ulcers given his negative H.pylori workup. Our patient also had the following risk factors: age >50, smoking history, HIV+ status (albeit well-controlled with undetectable viral loads), and likely long-term respiratory exposure at the workplace. The patient reported presence of a poorly maintained ventilation system directly above his workstation that blew cold mist on him constantly. The mortality rate may be as high as 27% with inadequate antibiotic therapy, so Legionella should be included in the differential if there is possible exposure. It is relatively uncommon for Legionella pneumonia to be associated with a complicated parapneumonic effusion, and it is also rare for the chief complaint to be gastrointestinal in nature. To the best of our knowledge, this is the first reported case of both occurring in the same patient.

Diagnostic Uncertainty: A Software Tool for Calculating the Uncertainty of Diagnostic Accuracy Measures

T. Chatzimichail ¹

¹Hellenic Complex Systems Laboratory (HCSL), Dorchester, UK

Purpose/Problem: Screening and diagnostic tests are applied for the classification of people into diseased and non-diseased populations. Although diagnostic accuracy measures are used to evaluate the correctness of classification in clinical research and practice, there has been limited research on their uncertainty.

Description of Program: For this reason, the freely available interactive program “Diagnostic Uncertainty” was developed for calculating the uncertainty of diagnostic accuracy measures of diagnostic or screening tests, measuring a normally distributed measurand, applied at a single point in time in samples of non-diseased and diseased populations.

Outcomes: The program “Diagnostic Uncertainty” calculates and plots the standard combined, measurement and sampling uncertainty and the resultant confidence intervals of various diagnostic accuracy measures of screening or diagnostic tests, for differing sample sizes, mean and standard deviation of the measurand, diagnostic threshold and standard measurement uncertainty of the test.

Discussion: This program calculates the standard and expanded combined, measurement and sampling uncertainty and the resultant confidence intervals of diagnostic accuracy measures of diagnostic tests, applied to samples of populations, providing 99 different types of plots and three different types of comprehensive tables, many of which are novel. To the best of our knowledge, no software, including all major general or medical statistical and uncertainty related software packages, provides this range of plots and tables without advanced programming.

Significance of Findings: The program “Diagnostic Uncertainty” can be used as a flexible, user-friendly, interactive educational or research tool in medical decision-making, to calculate the uncertainty of diagnostic accuracy measures.

A Metacognitive Calibration Tool to Help Medical Students Scaffold Diagnostic Reasoning during High Fidelity Simulations

L. Garbayo ¹, D. Harris¹, S. Fiore¹, M. Robinson¹, J. Kibble¹

¹University of Central Florida, Orlando, FL

Background: Metacognition is a necessary skill for successful self-monitoring and refining action with the avoidance of unnecessary and/or harmful medical interventions (Papa et al, 1999; Wachter et al, 2004). Within the breadth of metacognition, critical reasoning, as the ability to reflect and apply normative standards of reasoning to problem solving, is a fundamental skill for avoiding many biases. In particular, the Dunning-Kruger effect (1999) - the bias of overconfidence - is of interest in the context of a practitioner’s self-assessment. This bias prevails when the subject disregards the gaps in her knowledge base and acts confidently nonetheless, displaying a higher level of psychological confidence - against prudence and safety. This study had the following objectives: 1) Help novice students scaffold problem-solving, engage safely in the deliberate practice of diagnostic reasoning and medical decision-making in real-time, 2) Assess how accurately students gather and apply data in medical reasoning and treatment during High Fidelity Patient Simulations (HFPS), 3) Identify students scientific misconceptions related to the HFPS case, 4) Provide a stimulus to promote student metacognitive processing, self-assessment and self-efficacy, 5) Facilitate the explicit calibration of student confidence in deliberate reasoning with patient outcomes and team medical decision-making.

Method: Mixed method study; metacognitive calibration self-assessing (MCC) survey tool applied to high fidelity patient simulation (HFPS); N=80, 20 teams of 6 medical students; Semi-structured interviews with faculty N=5 Results: When compared against the faculty rubric, the mean correct student response rate ranged from 23-74 %, whereas the self-assessment of confidence ranged from 71-86 %. This result revealed overconfidence bias in novice students regarding the correctness of wrong responses. The most common misconception identified was inverting cause and effect: that metabolic acidosis was pointed as the cause of the patient’s problems, rather than a consequence of untreated diabetes mellitus. The most common treatment error was overtreatment, with unnecessary added medication. Semi-structured interviews with faculty facilitators suggested the MCC tool improved team process by slowing students down, required to think through their answers and that overall, the tool improved their critical thinking.

Conclusion: This study demonstrated the feasibility of using a metacognitive confidence calibration tool to assist novice students in learning safely to make deliberate diagnostic reasoning and medical decisions on patient care in real time during complex simulations, while observing objectively their levels of psychological confidence against patient outcomes. We suggest several uses for this new self-assessment instrument.

Implementation of a Novel Clinical Reasoning Workshop for Pre-clinical Medical Students

J. Inglis ¹, N. Wood², K. Chan³, A. Chrispal⁴

¹Central Adelaide Local Health Network, Adelaide, Australia

²College of Medicine and Public Health Flinders University, Bedford Park, Australia

³Southern Adelaide Local Health Network, Adelaide, Australia

⁴Central Adelaide Local Health Network, Adelaide, Australia

Purpose: Clinical reasoning involves integrating and applying different types of knowledge and reflecting upon the process used to arrive at a diagnosis (1). Clinical reasoning is a key competency of graduating medical professionals making it a focus of contemporary medical education endeavors (2). We aimed to develop a novel workshop to improve understanding of the key concepts of clinical reasoning and to further develop clinical reasoning skills.

Description of Program: A 120-minute clinical reasoning workshop was delivered to 154 second-year pre-clinical Doctor of Medicine students at Flinders University by a consultant physician in conjunction with trainee medical officers. The workshop began with a short didactic introduction, followed by a clinician role modelling clinical reasoning by thinking out loud and ending with students working through clinical cases in small groups with clinician facilitators. Students were asked to complete a pre- and post-session self-evaluation of competence in clinical reasoning on a 4-point Likert scale.

Outcomes: The self-evaluation was completed by 136 students (88% response rate). A statistically significant improvement in self-assessed competence was demonstrated for all evaluation questions; explaining how a clinician reasons through a case, comparing System 1 and System 2 thinking, describing how diseases are stored as illness scripts, generating a problem representation, using a diagnostic schema to generate differential diagnoses, ranking the likelihood of different diagnoses, and using feedback to improve clinical reasoning (p <0.0001 for all measures).

Discussion: There were perceived learning gains in understanding of clinical reasoning concepts and clinical problem solving after targeted workshops. The flipped classroom model with incorporation of team-based learning allowed students to engage in active learning through the lens of a clinical case (3). This prompted students to authentically construct their schemas and illness scripts consistent with adult learning theory.

Significance of Findings: Students perceived that these novel workshops were an effective means to better understand and apply clinical reasoning concepts. A follow-up study will assess to what extent students have been able to apply these concepts to cases encountered whilst on clinical placements and within case-based classroom teaching.

TOO PRICEY: A Conversational Tool to Promote Diagnostic Stewardship

R. Kowalski ¹, C. Moore¹, G. Moulder¹, A. Parsons¹

¹University of Virginia, Charlottesville, VA

Purpose: Unnecessary tests and treatments are a major driver of low-value care. Some of these interventions derive from patient request. Almost half of physicians report at least one patient request per week for an unnecessary test or treatment. Lack of training in how to navigate these requests leads clinicians to defer to patient wishes, increasing costs and potentially causing patient harm. We describe a formalized educational initiative for clinicians in training using a novel communication tool.

Description: The TOO PRICEY tool (Figure 1) was developed using a student-led, iterative approach with expert faculty following review of the available literature on patient-provider value conversations. TOO PRICEY has two temporal components: TOO outlines the clinician’s pre-encounter preparation, and PRICEY outlines of the conversation itself. The preparation elements include Test/Treatment, Options (consider less invasive, lower risk options), and Open (be openminded about patient’s concerns and requests). The conversational elements include Perspective (seek patient’s perspective), Reassure (convey confidence that concerning conditions are effectively ruled out), Incidental/Iatrogenic (discuss risk of incidental findings / adverse effects), Cost, Empathize, You (return conversation to patient). The tool was taught during a required, dedicated session within the pre-clerkship clinical skills course, in which a didactic presentation was followed by a hands-on workshop that included two practice scenarios: requests for MRI for headache and antibiotics for viral infection.

Outcomes: The workshop included 153 student participants. Pre- and post-session survey response rates were 98% and 95%, respectively. Prior to the workshop, 43% of students felt somewhat or extremely comfortable having conversations with patients about their requests for nonindicated tests or treatments; following the workshop, this percentage doubled to 84%. On the whole, 75% of students felt the quality of the workshop was “Good” or “Very Good.”

Discussion and Significance: Here we introduce the TOO PRICEY communication tool. Coupled with a hands-on workshop, this tool increased medical students’ confidence in navigating conversations with patients following requests for non-indicated tests and treatments. As a framework for eliciting and addressing patient concerns and promoting shared decision-making for diagnostic and therapeutic management, the TOO PRICEY tool is a step towards reducing unnecessary tests and treatments.

Increasing Feedback on Residents’ Clinical Reasoning Documentation Using Machine Learning

V. Schaye ¹, D. Kudlowitz¹, D. Sartori¹, J. Burk Rafel¹, M. Marin¹, B. Guzman¹, J. Feldman¹, I. Reinstein¹, Y. Aphinyanaphongs¹

¹NYU Grossman School of Medicine, New York City, NY

Purpose: Residents receive little feedback on their CR documentation due to time constraints of supervisors and lack of a shared mental model. We developed an innovative workplace-based assessment tool using machine learning (ML) to provide feedback on CR documentation. Here, we describe the impact of feedback using this tool.

Description of Study: In earlier phases, we developed a ML model that categorizes medicine resident admission notes as documenting high-quality vs. low-quality CR. All H&Ps written by residents are automatically analyzed by the ML model and displayed on a dashboard accessed via EPIC. In this study, we assessed the impact of implementation of feedback via this dashboard on residents’ CR documentation. The majority of H&Ps are written on night shifts which occur in two-week blocks. Half-way through the two-week block, the residents received an email including a PowerPoint overviewing the importance of high-quality documentation, a shared mental model for CR documentation, and an overview of the dashboard. Residents then had the opportunity to review their dashboard data and complete a post-survey documenting their action plans for improvement and provide feedback on the dashboard.

Outcomes: 27 residents received the intervention, 19 of which accessed the dashboard. 7 of those residents were excluded from data analysis as they did not have both pre- and post-intervention data. There was not a significant difference pre- compared to post-intervention in CR note quality. 46% (SD 26%) notes were high-quality pre- compared to 44% (SD 25%) post-intervention. However, there was a significant correlation between number of dashboard views and improvement in note quality with a pearsons correlation coefficient of 0.70. 7 residents completed the post-survey documenting specific action plans for improvement: “I will try to always include at least one alternative diagnosis with at least two data points explaining my choice.” Feedback for improvement on the dashboard included the desire for more specific feedback and suggestions to use this tool with faculty to review notes.

Discussion/Significance of Findings: While we did not see an improvement in note writing quality, there was a trend in improvement post-intervention in those residents who accessed the dashboard more than once. Adding this feedback modality to a program of assessment can lead to increased frequency of feedback and improve documentation quality with sufficient feedback. Next steps are to continue to collect data on impact, to iterate on the model to provide more specific feedback, and to train faculty to use this tool.

The Management Morning Report: An Educational Conference to Teach Management Reasoning

A. Zimmet ¹, R. Haug¹, S. David¹, A. Jepsen¹, B. Uthlaut¹, A. Parsons¹

¹University of Virginia, Charlottesville, VA

Purpose: The vast majority of residency education in clinical reasoning emphasizes diagnostic reasoning (DR). Recently, Cook et al. highlighted the need for greater focus on management reasoning (MR), defined as “the process of making decisions about patient management, including choices about treatment, follow-up visits, further testing, and allocation of limited resources.” Though how best to teach MR has not been well described, Parsons and colleagues recently proposed a conceptual model demonstrating the relationship between DR and MR and the key role of problem representation (PR) leading to a prioritized differential diagnosis.

Description of Program: We instituted a new curriculum via the daily teaching conference for internal medicine residents. In contrast to traditional conferences, these sessions began at the point of diagnosis and then used a case-based approach to emphasize core MR topics: testing and treatment thresholds, risk, high-value care, tolerating uncertainty, framing, overconfidence, monitoring, shared decision-making, and the management script. Topics were chosen based on literature review and group consensus. Sessions also focused on management as a means to diagnosis and employed a “PR time-out,” a pause in discussions where each resident was asked to generate an accurate, concise PR in order to refine and prioritize the differential diagnosis.

Outcomes: Pre- (N = 28) and post-intervention (N = 31) surveys revealed a statistically significant increase in residents’ confidence using clinical reasoning skills to obtain a diagnosis (mean: pre 3.68, post 4.10, p = 0.009) and to make management decisions (mean: pre 3.50, post 4.45, p = 0.0001) during clinical care, as well as an improvement in their understanding of the concept of PR (mean: pre 3.71, post 4.39, p = 0.001). There was also improvement in residents’ understanding of the concept of scripts and ability to verbally describe their reasoning, however these were not statistically significant.

Discussion: A case-based educational conference series on MR increased resident confidence in diagnostic and management decision-making during patient care. Future directions include further investigation of key components and methods for effective MR instruction.

Significance of Findings: We describe a novel approach to teaching MR to residents using a case-based educational conference series.

The Effect of a Pandemic on the Diagnosis of Common Illnesses with Similar Clinical Presentation

M. Gusack ¹

¹MANX Enterprises, Ltd., Huntington, WV

Statement of Problem: When faced with a pandemic, other illnesses may go undiagnosed, at least for a period of time. This is due to the redirection of societal resources, tendency of clinicians to become biased towards the pandemic diagnosis, as well as to the delay in patients seeking attention for other illnesses. In addition, infectious illnesses often overlap in their presentation, clinical course, and medical supportive requirements. This can lead to a rise in misdiagnosis and risk of harm to the patient.

Description of the Intervention or Program: Presented is a comparative overview of the clinical presentation, as well as acute physical, radiologic, and laboratory findings for Covid-19 against selected infectious and respiratory illnesses. A model based on prior epidemiologic data for these other illnesses is used to determine the degree of risk for diagnostic error as well as how this might affect epidemiologic data for the other, underdiagnosed illnesses that occur coincidently. The models compare several variables including but not limited to: • Transmissibility • Infectivity • Virulence • Physical findings, radiology, and laboratory test results

Findings to Date: Although simplistic, the model shows that, in the absence of a reliable laboratory test, the more an alternative illness overlaps in terms of its clinical presentation and course, and the higher its prevalence relative to the Covid-19 infection, the more likely it is that a misdiagnosis will occur confounding epidemiologic data used to track the pandemic and to determine the effectiveness of medical intervention including vaccination.

Lessons Learned: Pandemics overlapping other common illnesses may: • Lead to its under diagnosis of the cause of the pandemic early before it’s recognized as a new infection • As the pandemic becomes apparent, additional diagnostic error may occur in the absence of a confirmatory test • Overdiagnosis of the new infection after its initial growth in prevalence diverts attention away from other illnesses • Confound epidemiologic data upon which critical decision are made in fighting a pandemic

Leveraging the Clinical Laboratory to Advance Diagnostic Excellence

I. Lubin ¹, J. R. Astles¹, J. Bunn¹, N. Cornish¹, M. Gagnon¹, P. Sandhu¹, M. Scott¹, V. De Jesús¹

¹Centers of Disease Control and Prevention, Atlanta, GA

Statement of Problem: Clinical laboratory services, include testing and expert consultation, are not optimally used by clinical care providers to inform accurate and timely clinical diagnoses communicated to healthcare providers and the patient.

Description of the Intervention or Program: The Division of Laboratory Systems at the Centers for Disease Control and Prevention established the Diagnostic Excellence Team in 2021 tasked to leverage clinical laboratory capabilities to measurably reduce diagnostic errors and promote diagnostic excellence. A strategy was developed that builds upon the Division’s 2020 publication, Bringing the Laboratory into the Strategy for Diagnostic Excellence that describes an integrated quality management approach across laboratory and patient care settings to advance diagnostic excellence. (1)

Findings to Date: Criteria were defined to prioritize diagnostic excellence initiatives These criteria emphasize steps needed to identify elements of the laboratory testing process associated with preventable diagnostic error to inform efforts to improve diagnoses. To map laboratory to patient-care processes associated with making diagnoses, the traditional total testing process model was revised to emphasize patient-centered care, communication channels among patients, clinicians, laboratory professionals, other medical professionals, and the use of information technology, data management, and analytics. The criteria and revised total testing process model were used to identify and prioritize candidate projects designed to measurably reduce wrong, missed, and delayed diagnoses. Potential projects identified include 1) Identifying challenges and opportunities for the laboratory during the course of the pandemic to inform interventions for future events, 2) identifying challenges and opportunities associated with the use of next generation sequencing across medical disciplines, 3) assessing compliance with existing guidelines for prenatal screening, and 4) improving identification and reduction of blood culture contamination. Stratifying findings across demographics may identify health disparities and inform interventions. A partner engagement strategy to support this work was developed to engage healthcare providers, laboratory professionals, managers, leadership, and patient advocates.

Lessons Learned: Current models for study and practice are limited in associating elements of the total testing process to informing accurate and timely clinical diagnoses communicated to healthcare providers and the patient. Adaptation of clinical and laboratory practice models, metrics, and partner engagement is expected to be a viable strategy for leveraging the laboratory to reduce diagnostic errors and promoting diagnostic excellence.

Citation: (1) Lubin IM, Astles JR, Shahangian S. Bringing the laboratory into the strategy for diagnostic excellence. Diagnosis (Berl). 2021. Online ahead of print.

Difficult Airway Diagnosis Identification and Response Team (DART): A Five-year Health System Experience

S. McManus ¹, P. Mathur¹, B. Abdelmalak¹, R. Khatib¹, A. Hamilton¹, A. Gill¹, E. Lamarre¹, S. Mummadi¹

¹Cleveland Clinic, Cleveland, OH

Statement of Problem: Difficult airway is encountered in 1-12 % of the adult population, as commonly seen in patients undergoing general anesthesia. The wide range of incidence reflects the lack of a consensus definition and includes a composite of difficult laryngoscopy, difficult mask ventilation, and difficult intubation. A 3-5 minute delay in successful management of an airway can result in severe and permanent harm, which continues to occur despite advances in airway management devices and consensus management guidelines. Ready availability of information related to difficult airway across healthcare encounters is critical in implementing quality improvement measures and preventing future airway management difficulties.

Description of the Program: The unique process of difficult airway identification was developed and implemented in 2014 at the Cleveland Clinic Health System. Our process includes use of multimodal communication tools triggered by addition of difficult airway diagnosis to the problem list. This operationalizes electronic health record (EHR) tools, highlighting the diagnosis to key stakeholders including a difficult airway icon, wrist band, SBAR information, and a difficult airway letter for the patient). We have also successfully paired our initiative with a multidisciplinary difficult airway response team to help manage challenging airways.

Findings to Date: We have more than 4,000 patients with the diagnosis of difficult airway since 2014. In 2020, we identified 551 patients with a difficult airway diagnosis among 136,766 inpatient admissions, translating to an inpatient incident rate of ∼0.4%. Since 2014, we have seen an increasing number of patients diagnosed with a difficult airway (Figure 1).

Lessons Learned: We successfully implemented a standardized process for identifying patients with difficult airway across a large, multi-center health system. While capturing difficult airway data is a crucial step in executing actionable change, the challenge of defining and predicting patients with difficult airways remains. We plan to further characterize and predict which patients may have a difficult airway through a natural language processing and machine learning approach that measures accuracy and validates key features of the airway management procedure note. This will allow clinicians to proactively assemble resources to decrease the possibility of adverse events.

Information Overload: Keeping Teams Up to Date in the COVID Crisis

S. Peterson ¹, M. Saheed¹, S. Figueroa¹, P. Johnson¹

¹Johns Hopkins University, Baltimore, MD

Background: The March11th declaration by the WHO of a pandemic solidified what many leaders in Emergency Medicine in the United States had both prepared for and feared. While many institutions had developed workflows and strategies for surge and disaster management, getting that information into the hands of the end user in a useful way was not always clear. Integrated pathways in the electronic medical record have been utilized in Emergency Departments across the country to aid in standardization of practice and to guide appropriate recourse utilization. We repurposed this tool to disseminate clinical information in pathway format to deliver just in time COVID-19 information for 1) workflow changes, 2) PPE recommendations, 3) testing and discharge information, 4) and management recommendations.

Methods: Laid out in a pathway format, we had the ability to make daily changes based on the most up to date national information and institutional hospital infection control resources. As stages of the surge plan were activated including Alternate care sites (tents) and negative flow pods, workflows could be progressively updated. PPE recommendations vary in certain areas of the department and just in time information including PPE Videos that are embedded into the pathways allowed staff to have immediately available information on how to keep themselves safe. Testing algorithms were initially rapidly changing, including near daily geographic changes and later being risk profile based. Testing recommendations are updated daily as needed in the pathway. In addition, advisements from the CDC related to self-isolation as well as local health department resources and requirements are regularly updated and available with one click printing for discharge instructions and prescriptions as needed. While there was initially a paucity of strong evidence basis of medical management for COVID patients, data from experiences in China and Europe was steadily becoming available and integrated into pathways. Information around local and national resource and medication shortages, and recommendations for highest risk patients were also integrated into the pathways. Airway management strategies including methods to reduce aerosolizing, and early intubation considerations were included in addition to Acute Respiratory Distress Syndrome (ARDS) vent management recommendations.

Results: As volumes of covid rule out patients have increased; we have seen utilization of the COVID resources steadily increase. Given the ease of use and rapid ability to update site specific information, COVID guidelines were activated at each of our five health system sites with rapid adoption.

Study of Association of UACR with Endothelial Dysfunction in HIV Patients on HAART

A. Rajanna ¹, S. Basava Hosakare²

¹Rajive Ghandhi University of Health and Science, Bangalore, India

²BMCRI

In HIV (Human immunodeficiency virus) positive patients several factors have been proposed to explain the cardiovascular (CV) risk, which include pro-athero-thrombotic viral effect, infection-mediated immune dysfunction and possible effects of HAART (Highly active antiretroviral therapy). Presence of endothelial dysfunction, an early CV risk marker, in HIV-infected patients on long-term HAART. Endothelial dysfunction can be measured by bFMD (Brachial artery flow mediated vasodilatation). Urine albumin to creatinine ratio (UACR)is frequent in HIV patients and is predictor of cardiovascular risk. UACR and endothelial dysfunction are positively associated in HIV affected patients thereby increased UACR might help to identify endothelial dysfunction in HIV patients on HAART.

Objectives: • To measure urine albumin to creatinine ratio in HIV patients on HAART • To find out correlation between UACR and bFMD

Methodology: Study was conducted on HIV Patients admitted to hospital, considering the inclusion and exclusion criteria. Detailed clinical history, examination and Blood investigations were done. Endothelial function was assessed by brachial artery flow mediated vasodilatation(bFMD).

Results: Among 100 HIV patients,14% had UACR <30 and 86% had UACR >30 and mean UACR is 90 ±06. Mean BFMD was 8.44 ± 7.65. We identified a significant Negative correlations were found between bFMD and UACR(r=- 0.832, P=<0.001).

Conclusion: Use of urine albumin to creatinine ratio (UACR) as a routine screening tests in those who are HIV positive on HAART can be used to find out those cases of HIV who are progressing towards cardiovascular impairment so that newer approaches can be used in them to prevent further cardiovascular involvement.

The Impact of Diagnostic Delays on Household Transmission of Pertussis

A. Arakkal ¹, P. Polgreen¹, S. Koeneman¹, J. Cavanaugh¹, AaronMiller¹

¹University of Iowa, Iowa City, IA

Background: Household transmission of pertussis is common. Because pertussis is extremely contagious early in the course of the disease, delayed diagnosis of pertussis may increase household transmission. The purpose of this study is to assess the association between delays in diagnosis of pertussis and the risk of secondary household transmission of pertussis.

Methods: We conducted a retrospective cohort study using the Truven Health Analytics Commercial Claims and Encounters and Medicare Supplemental databases from 2001 through 2017. Our analysis was restricted to households with two or more family members enrolled in the same insurance plan in a given month and in which at least one pertussis case occurred. We then identified the initial (index) case of pertussis in each household and determined if a secondary case occurred in another family member within 30 days after the index case. Next, we determined if the index case experienced a diagnostic delay based on having one or more symptomatically-similar diagnoses (SSDs) (e.g., fever, cough, asthma, croup, etc.) within 21 days prior to their index diagnosis. Finally, we collapsed the data at the household level and used logistic regression to estimate the likelihood of secondary transmission of pertussis associated with exposure to a family member who experienced a diagnostic delay.

Results: A total of 45,472 cases of pertussis were identified among multiple-member households, corresponding to 41,020 households included in the analysis. Across all households, 1,184 (2.9%) individuals experienced a secondary transmission of pertussis within 30 days of the index pertussis case in the household. Furthermore, the index cases in 19,688 (48.0%) households were observed to experience a diagnostic delay for pertussis. Results of the logistic regression model show an increased odds of secondary transmission among households in which the index case experienced a diagnostic delay. The odds of secondary transmission in a household were 5.09 times (95%CI: 4.38, 5.91; p<0.001) greater in households in which the index case experienced a diagnostic delay.

Conclusion: In this study we show that diagnostic delays for pertussis is associated with an increased likelihood of household-pertussis transmission. This study adds to the growing body of evidence highlighting the public-health implications of diagnostic delays.

Applying the DEER Taxonomy to Patient-Perceived Breakdowns in Diagnosis, Early During COVID-19: An Exploratory Study

K. Baker ¹, M. Brahier², M. Penne², S. Davis³, M. Hill¹, K. Smith¹ *

*(Presenting Author)

¹MedStar Health, Washington, D.C.

²Georgetown School of Medicine, Washington, D.C.

³Howard University, Washington, D.C.

Background: Diagnostic error is a leading cause of patient harm, impacting an estimated one in twenty primary care patients in the U.S. Identifying breakdowns in diagnosis is challenging, and patients are an underused source of information. In this study, we aimed to explore patient-perceived diagnostic breakdowns during primary care encounters by examining routinely collected patient satisfaction surveys.

Methods: We used a mixed methods study design to explore patient-perceived breakdowns in diagnosis during primary care encounters in April-May 2020. The data consisted of free text comments from a patient experience survey, completed by patients after their primary care visit. Comments sentiment-coded as negative were systematically reviewed using content analysis to identify patient-perceived breakdowns in diagnostic accuracy, timeliness, and communication. Breakdowns were also coded using the Diagnostic Error Evaluation in Research (DEER) taxonomy. Descriptive statistics and bivariate analysis were used to summarize the data using Stata 16.

Results: A total of 626 surveys were confirmed to be sentiment-coded as negative. The negative commenters were majority women (n=395; 63.1%), White (n=392; 63.0%) or Black (n=136; 21.9%), with an average age of 62.0±17.2 years. Of the negative commenters, 160 patients (25.6%) identified a breakdown in diagnosis. There were no significant differences in the gender, race, or age of the patients who perceived diagnostic breakdowns compared to those who did not. Diagnostic breakdowns were significantly more likely to be reported on paper surveys (n=87) than web-based surveys (n=73) (p = 0.000). Breakdowns were categorized by type (accuracy, timeliness, communication), with communication errors being most common (n=84, 52.5%). Breakdowns were also classified using the DEER taxonomy, with breakdowns in access (n=49, 30.6%) and testing (n=40, 25%) most commonly reported. Although treatment is not included in the DEER taxonomy, breakdowns in treatment were frequently reported by patients (n=40, 25%).

Conclusion: The application of the DEER taxonomy served as an anchor for examining patient-perceived diagnostic breakdowns. Modifications to the DEER taxonomy are needed to better reflect the patient perspective. Our future research aims at developing and validating a patient-focused tool for identifying breakdowns in diagnosis to be used for systems-based learning.

Pilot Usability Assessment of AHRQ Common Formats for Event Reporting - Diagnostic Safety

A. Bradford ¹, U. Shahid¹, A. Marinez¹, G. Schiff², M. Graber³, P. DiStabile⁴, J. Brady⁴, H. Jalal⁴, H. Singh⁵

¹Baylor College of Medicine, Houston, TX

²Harvard Medical School, Boston, MA

³Society to Improve Diagnosis in Medicine

⁴Agency for Healthcare Research and Quality, North Bethesda, MD

⁵Baylor College of Medicine and Michael E. DeBakey VA Medical Center, Houston, TX

Background: A lack of consensus around definitions and reporting standards for diagnostic errors limits the extent to which healthcare organizations can share and learn from these events. Common definitions and reporting formats are also needed to facilitate the contribution, aggregation, and analysis of data to the Agency for Healthcare Research and Quality (AHRQ) Network of Patient Safety Databases for shared national learning about how to improve diagnostic safety. In response, in 2019 AHRQ began development of Common Formats for Event Reporting - Diagnostic Safety (CFER-DS). We conducted a pilot usability assessment of the draft CFER-DS to inform future revision and implementation. Our aim was to assess whether users found the CFER-DS items valid, appropriate in scope, and adequate to the task of capturing details of diagnostic safety events. We also explored users’ perceptions of feasibility and potential for future adoption by healthcare organizations.

Methods: We recruited a purposive sample of quality and safety personnel working in 8 US healthcare organizations. Participants were invited to use the CFER-DS to simulate reporting for a minimum of 5 cases of diagnostic safety events of their choice. We collected and analyzed written and verbal qualitative feedback about the CFER-DS using a written questionnaire and semi-structured interview guide. Because terminology and concepts related to diagnostic safety are relatively novel in the context of safety event reporting, we were primarily concerned with perceptions of content validity and ease of use.

Results: Estimated average completion time for all CFER-DS items was 30-90 minutes per case. Participants shared generally positive feedback about content coverage and item clarity but identified reporter burden as a potential concern and provided suggestions to streamline data capture. Participants also identified opportunities to clarify conceptual definitions, ensure applicability to both inpatient and outpatient encounters, and develop guidance to operationalize use. Findings led to refinement and restructuring of CFER-DS content.

Conclusion: Our evaluation provided early evidence that standardized definitions of diagnostic safety events, contextual information, and contributing factors can be used by clinicians and safety professionals to capture diagnostic safety events. AHRQ is now seeking additional feedback from the general public to ensure that the CFER-DS captures the information needed to make future reporting efforts relevant and useful to a broad range of users. We will present findings from our pilot study as well as a summary of feedback from the public comment period and later revisions to the CFER-DS.

Self-Explanation Use in Nurse Practitioner Student Diagnostic Reasoning

L. Burt ¹, S. Corbridge¹

¹University of Illinois Chicago, Chicago, IL

Background: Accurately diagnosing patients is a vital skill that nurse practitioners (NPs) must frequently and fluently execute in order to facilitate positive patient outcomes. However, there is currently a major gap in educational best practices regarding how to foster diagnostic accuracy. The ramifications of this educational gap are demonstrated by the fact that misdiagnoses, originating from all provider types, remain the most common cause of medical error in the US. This research explores self-explanation, a promising learning strategy that has the potential to foster diagnostic accuracy. Self-explanation is defined as the purposeful technique of generating self-directed explanations to process novel information while problem-solving. Self-explanation facilitates problem-solving accuracy and learning of didactic information. High-quality ways that learners self-explain have yet to be explored among NP students.

Methods: This mixed-methodology content analysis occurred in two phases. Phase one explored the research question: How do NP students self-explain during diagnostic reasoning? Phase two further analyzed data using associational and comparative quantitative analysis techniques. The research aims of phase two were twofold: (a) explore relationships between ways of self-explaining and diagnostic accuracy, and (b) compare differences between student expertise levels in terms of diagnostic accuracy and ways of self-explaining.

Results: Qualitative analysis revealed 17 meaningful categories of NP diagnostician self-explanation. Categories were classified and described as either inferences or non-inferences. Expert students self-explain in qualitatively different ways than novice students, using significantly more clinical and biological inference self-explanations (expert: M = 0.31 vs. novice: M = 0.17, t(35) = -4.13, p = 0.0002), and less non-inference self-explanations (expert: M = 0.69 vs. novice M = 0.83, t(35) = 4.25, p = 0.0002), compared to novice students. In terms of diagnostic accuracy, the most strongly predictive way of self-explaining is inferring biologically (r(35) = 0.49, p < 0.002). The association between clinical inference frequencies and diagnostic accuracy scores did reach levels of significance (r(35) = 0.14, p < 0.3).

Conclusions: Findings of this study should guide the development of diagnostic accuracy interventions leveraging biologically focused self-explanations. Educational curricula should integrate anatomy, physiology, and pathophysiology into disease-state courses, supporting students to view patient presentations in terms of underlying biology. Future research should explore the effectiveness of these interventions and curricular changes on diagnostic accuracy within actual clinical settings. Fostering the momentum created by this study will improve the ways we teach NP diagnostic accuracy in hopes of ultimately promoting patient well-being.

EHR Re-access by ED Clinicians after Critically Ill Children are Admitted to the Pediatric ICU

C. Cifra ¹, E. Westlund¹, C. Tigges¹, S. Miller¹, R. Gomez¹, A. R. Pauley¹, P. TenEyck¹, P. R. Pennathur¹, D. F. Sittig², H. Singh³

¹University of Iowa, Iowa City, IA

²University of Texas Health Science Center, Houston, TX

³Michael E. DeBakey Veterans Affairs Medical Center and Baylor College of Medicine, Houston, TX

Background: High-quality emergency care for children depends on emergency department (ED) clinicians’ ability to improve their practice by learning from past performance. Clinicians may obtain information on patients’ outcomes by re-accessing their electronic health records (EHRs) after their ED visit. We characterize how ED clinicians re-access critically ill children’s EHRs after pediatric intensive care unit (PICU) transfer.

Methods: We performed a retrospective cohort study of 100 randomly selected patients admitted to a single PICU from the same institution’s ED over a year. Using EHR access logs, we determined the frequency and characteristics of EHR re-access by ED attending and resident physicians up to 30 days from the patient’s ED visit. We only included EHR access episodes which occurred after clinicians signed their ED note. Note-signing signaled that documentation was complete, thus subsequent EHR re-access likely reflected clinicians’ intent to follow-up on patients. We also compared the characteristics of patients whose EHRs were re-accessed vs. not re-accessed.

Results: Sixty-two of 100 (62%) patients’ EHRs were re-accessed up to 13.4 days (maximum) from their ED visit. EHRs were re-accessed a median of 2 (IQR 1-4) times by both ED attendings (n=18) and residents (n=41) who spent a median of 3.1 (IQR 0.6-16.8) minutes reviewing the EHR. The most frequently accessed sections were clinician notes (47%), miscellaneous reports with patient data (36%) and laboratory/imaging results (29%). Patients whose EHRs were re-accessed had more ED diagnostic tests performed (median 11.5 vs. 8 tests, p=0.016), had a shorter ED stay (median 157 vs. 207 minutes, p=0.036), and were more often seen by a pediatric-trained ED attending physician (74% vs. 42%, p=0.003). Diagnosis discrepancies and unplanned transfers to the PICU after initial floor admission were similar in both groups.

Conclusions: ED clinicians commonly re-access critically ill children’s EHRs after PICU transfer likely to learn about patients’ subsequent clinical outcomes. Differential EHR re-access may result in ED clinicians missing information on diagnosis discrepancies and unplanned PICU admissions. These findings support and inform the development of formal systems to deliver consistent patient outcome feedback which may improve the practice of frontline clinicians who care for critically ill children.

Virtual Multi-Disciplinary Conferencing Compared to In-Person During the COVID-19 Pandemic: Better, Worse or No Difference?

T. Pan ¹, M. Escudero-Fung¹*, B. Jonik¹, N. Cumbo¹

*(Presenting Author)

¹Penn State College of Medicine, Hershey, PA

Background: The Multidisciplinary Team (MDT) is a collection of professionals from various clinical specialties who come together to make decisions regarding the treatment of individual patients, specifically those with complex medical conditions or oncological malignancies. During unprecedented times, throughout the COVID-19 pandemic, adaptations to traditional clinical practice must be made in order to minimize the risk of contact and spread of infection, while simultaneously providing quality healthcare. Consequently, physicians have turned their attention and focus to the virtual space. This study explores the impact and value of virtual conferencing on MDTs compared to the in-person counterpart during the current pandemic.

Methods: A 10-question survey was generated and sent out electronically to various specialties and providers across Penn State Health Milton S. Hershey Medical Center. The goal of the survey was to gauge how the virtual MDTs compared with the in-person counterparts through both multiple choice and open-ended questions. The focus was on the standard of patient care, potential challenges faced, decision making and organization, viewing of pathology slides or images, interpersonal communication, rapport and continuity of care.

Results: 31 responses were generated. Specialties included radiology (68%), pathology (21%), oncology (3%), rheumatology (3%), surgery (3%) and other (9%). The three greatest challenges of virtual MDTs were communication, organization and viewing of images or pathology slides. A majority responded “no difference” in the virtual space compared to the in-person counterpart with regards to standard of care (70%), decision making (65%) and continuity of care (58%). Interpersonal communication was split between “worse” or “no difference” in the virtual space (42%). Organization of thought and in reducing the number of simultaneous speakers was split between “better” or “worse” (35%). A majority supported the continued use of virtual MDTs following the end of the pandemic (58%). An area of improvement of virtual MDTs is the education of users on the program interface, such as breakout rooms and various annotating tools.

Conclusion: Virtual MDTs should be seen as an effective and pragmatic tool that can minimize interpersonal contact and risk of viral transmission while providing comparable quality and continuity of care during the COVID-19 pandemic. As modern medicine continues to adapt to the technological age, the virtual space may serve as a viable alternative in the diagnosis, planning and treatment of patients.

How Does a Diagnosis Described in the Patient’s Own Words Compare to a Clinician-documented Diagnosis?

K. Gleason ¹, M. Dahm²

¹Johns Hopkins School of Nursing, Baltimore, MD

²Australian National University, Canberra, Australian

Background: It is critical to the diagnostic process that the patient both understands and agrees with a new diagnosis. Failure to convey an explanation of a health problem to a patient constitutes diagnostic error. However, little is known about what patient may retain of any diagnosis they receive from clinicians. We investigated how patients describe a new diagnosis following Emergency Department (ED) discharge, and how a patient-reported diagnosis compared to the diagnosis documented in the electronic medical record (EMR).

Methods: To compare patient-reported diagnoses to EMR-documented diagnoses, we analyzed from a cohort study of patients discharged from three EDs in an academic health system in the Mid-Atlantic in which patients completed questionnaires regarding their understanding of their diagnosis and/or follow-up steps. Inclusion criteria: adult ED patients aged 18 and older seen within the last 7 days. Two independent researchers compared patient-reported new diagnosis following discharge to EMR- documented diagnosis using a categorical 4-point scale for diagnostic content (identical, insignificantly different, different, not enough detail) and a 3-point scale to rate the technicality level of language of diagnostic description (medical term, semi-technical, lay). Statistical analyses explored the association between race, sex, education, and technicality and how the patient-described diagnosis compared to the documented diagnosis using ordinal logistic regression.

Results: Of the total sample recruited (n=137), the majority (n=95) reported receiving a diagnosis and stated the given diagnosis. Of those who reported their diagnosis, the majority were women (66%, n=62), the average age was 43 (SD 16), and a fourth (n=24) were Black and 66% (n=63) were white. The majority (84%) described diagnosis identical or with insignificant differences, 11% gave a different diagnosis than the one documented, and in 5% of cases not enough detail was available for comparison. More than half of patients used semi-technical (34%) or technical (25%) language to describe their diagnosis, with over a third (40%) using lay language to give their diagnosis. Only older age was minimally associated with a higher likelihood of the patient-reported diagnosis differing from the documented diagnosis (OR 1.03, 95% CI: 1.001, 1.06).

Conclusion: Patient reports of their new diagnosis following ED discharge had moderate agreement with the diagnoses recorded in the EMR. Previous research reported that patients’ and clinicians’ understanding and knowledge of semi-technical and technical term diverge widely. Thus, patients might reproduce verbatim semi-technical or technical diagnoses they received from clinicians, but not fully understood what the diagnosis means for them.

A Qualitative Evaluation of a Toolkit to Engage Patients and Families to Improve Diagnostic Safety

M. Hill ¹, K. Baker¹, H. Haskell², T. Giardina³, D. Wesley⁴, E. Hochberger⁴, K. Smith¹

¹MedStar Health Institute for Quality and Safety, Columbia, MD

²Mothers Against Medical Error

³Baylor College of Medicine, Houston, TX

⁴MedStar Health Research Institute, Hyattsville, MD

Background: Patients and families have unique contributions to creating an accurate and timely diagnosis, yet patient and family engagement (PFE) in diagnostic safety is rare. The Engaging Patients and Families to Improve Diagnostic Safety toolkit includes an agenda setting tool (Share Your Story note sheet) to prepare patients to share their diagnostic story, coupled with an active listening and reflective practice exercise for providers (60 Seconds for Diagnostic Safety). The present study examined the feasibility of implementing this multi-strategy toolkit within primary care settings.

Methods: Twenty-three primary care practices from across the U.S. participated in a field test from September 2020 to March 2021. After onboarding to the program, implementation champions and practice staff engaged in biweekly community of practice sessions using the tele-ECHO model. Post-implementation, semi-structured interviews and focus groups were conducted with practice stakeholders to assess strategy feasibility, receptivity, and usability. In addition, practices collected feedback from patients on receptivity and perceived impact using surveys and/or review of completed note sheets.

Results: Nineteen practices (86.4%) completed the field test. The primary reason for study withdrawal was the COVID-19 pandemic. Interviews with providers (n=23), practice staff (n=14) and administrators/leaders (n=11) from the nineteen practices were completed. Interviews produced 580 individual comments and recommendations from providers (n=372; 64.1%) and staff/administrators (n=136; 23.4%) responsible for implementation or decision making around adoption. Overall, the practices found the strategy feasible and beneficial. Most practices (n=18, 95%) reported that the agenda-setting note sheet tool helped patients organize their thoughts and concerns related to their diagnosis. One provider said, “It’s like the patient is writing their own note.” Most patients also thought the note sheet improved communication with their providers (n=102; 87.6%). Practices utilizing the 60 Second active listening strategy and reflective practice exercise (n=17, 89.5%) perceived improved diagnostic reasoning, while 98.3% of patients surveyed (n=115) reported that their provider listened carefully to them after strategy implementation. Providers reported the 60 Seconds, coupled with the note sheet, together yielded information that may not have otherwise been revealed by the patient. Improvements to the strategy focused on simplifying the materials, enhancing health literacy considerations, and improving “getting started” directions for busy practices.

Conclusion: Practices and patients found the Engaging Patients and Families to Improve Diagnostic Safety strategy feasible to implement. Feedback to improve the strategy focused on reducing complexity and enhancing usability.

Mitigating the Exacerbation of Inequalities During COVID-19 in Colorectal Cancer; Patient and Healthcare Professional Perspectives

A. Ip ¹, G. Black², C. Vindrola², C. Taylor¹, N. Fulop², A. Bhuiya³, R. Oliphant⁴, C. Taylor⁵, M. Machesney⁶, A. Wong⁶, J. Green⁶, S. Otter⁷, M. Hewish⁷, J. Callin¹, B. Bartlett¹, K. Whitaker¹

¹University of Surrey, Guildford, United Kingdom

²University College London, London, United Kingdom

³North Central London Cancer Alliance, London, United Kingdom

⁴NHS Highland, Inverness, Scotland

⁵London North West University Healthcare NHS Trust, London, United Kingdom

⁶Bart’s Health NHS Trust, London, United Kingdom

⁷Royal Surrey NHS Trust, Guildford, United Kingdom

Background: COVID-19 has led to rapid changes in health care delivery, raising concern that these changes may exacerbate existing inequalities in patient outcomes and impact on diagnostic delays in cancer. The aim of this study was to provide an in depth understanding of patient and healthcare professional experiences of the colorectal cancer pathway during the COVID-19 pandemic.

Methods: We carried out interviews with 39 participants (20 higher socioeconomic (SES); 19 lower SES, indexed by education) across the UK who contacted primary care about symptoms of colorectal cancer during the pandemic. We used framework analysis (informed by the candidacy framework) with patient interview data to understand how people decided to seek medical help and how they navigated services. Comparative thematic analysis explored differences between higher and lower SES groups. We are currently interviewing healthcare professionals across the care pathway (n∼30) to understand changes in healthcare delivery and the potential impact on inequalities in care. Healthcare professional interviews will be completed in July 2021 and included in our presentation.

Results: COVID has introduced more ambiguity into diagnostic processes in primary care. Our findings suggest that this ambiguity has exacerbated disparities between higher and lower SES participants in terms of understanding their symptoms, deciding to seek help, accessing the GP and knowing what to do next. People in the lower SES group conflated COVID and colorectal symptoms, causing diagnostic delay in help-seeking. In contrast higher SES participants had more confidence appraising symptoms and reported heightened symptom vigilance during the pandemic due to lifestyle changes as a result of COVID. There was potential for increased disparity in diagnostic process errors, with new safety-netting methods (e.g., apps/email to re-contact) taken up more readily by higher SES patients. It will be important to gather the views of healthcare professionals to understand more about these changes and consider changes further along the care pathway. We will present further detailed comparative analyses of challenges in help-seeking in the context of the pandemic, as well as findings from the healthcare professional interviews.

Conclusions: Changes in the diagnostic interval post-COVID may risk deepening disparities in colorectal cancer outcomes. Our study will identify the main changes in the colorectal pathway in the UK and the impact on patient outcomes and inequalities in care. We will generate practical recommendations, tailored to different groups/needs, to amplify the benefit of positive changes and to avoid exacerbating inequalities.

Incidence, Impact and Risk Factors of Misdiagnoses Among Patients Hospitalized with Suspected Bacteremia

A. Kaal ¹, E. Speelman², S. Meziyerh², E. Steyerberg², K. Nieuwkoop¹, N. Kolfschoten¹, M. van Aken¹, E. Schippers¹

¹Haga Teaching Hospital, The Hague, Netherlands

²LUMC

Background: Misdiagnosis of infection are among the top of diagnostic errors that are associated with increased morbidity and mortality. The objective of this study is to evaluate the rate of misdiagnosis in patients presented to the emergency department (ED) with suspected bacteremia, the impact of misdiagnosis on clinical outcomes and to identify risk factors associated with misdiagnosis.

Methods: This is a single-center observational cohort study that took place at Haga Teaching Hospital, the Netherlands. All adult patients who were admitted after presentation at the ED and from whom a blood culture was taken, were enrolled from April 2019 through May 2020. Misdiagnosis was defined as a discrepancy between the suspected site of infection at hospital admission and the final diagnosis at discharge. Clinical outcome measures were 30-day mortality, need for admission to the intensive care unit (ICU), length of hospital stay and use of broad-spectrum antibiotics. Risk factors for misdiagnosis were determined using a multivariable logistic regression model.

Results: Overall rate of misdiagnosis was 11.6% (171/1477), with relative high rates for the following site of infections: intravascular (46.7%, 14/30), central nervous system (CNS) (38.1%, 8/21), bone and joint (31.6%, 12/38), abdominal (14.8%, 30/203) and urogenital (12.8%, 43/336). Regarding the impact of misdiagnosis, no relation was found between misdiagnosis and 30-day mortality, ICU admission and duration of admission (adjusted odds ratio (aOR) 0.8, 95% CI 0.3-1.9, p=0.60; aOR 1.3, 95% CI 0.6-3.0, p=0.54; adjusted effect size 1.1, 95% CI, 0.9-1.3, p=0.41 respectively). Misdiagnosis was associated with higher odds of receiving broad-spectrum antibiotic therapy (aOR 4.0, 95% CI 1.8-8.8, p<0.001). Adjustment was made for age, Charlson comorbidity index, quick Sepsis Related Organ Failure Assessment score and site of infection at discharge. Older age, dementia, a positive urine sediment without urinary symptoms and the suspected site of infection (intravascular, CNS and bone and joint indicated highest risk) were the most prominent risk factors associated with misdiagnosis.

Conclusion: Misdiagnoses among patients admitted to the hospital with suspected bacteremia occurs in every one out of nine patients, but it is not associated with worse clinical outcomes. Older age, dementia and a positive urine sediment without urinary symptoms are risk factors for misdiagnosed site of infection. Measures to reduce the occurrence of misdiagnosis are important since it is associated with antibiotic overtreatment. Trial registration: www.trialregister.nl: NL7852

Diagnostic Error in Surgery: Analysis of a Comprehensive National Database of Closed Medicolegal Cases

J. Kwan ¹, L. Calder², C. Bowman², A. MacIntyre², L. Honey², R. Mimeault², C. Dunn², H. Singh³

¹University of Toronto, Toronto, Canada

²Canadian Medical Protective Association, Ottawa, Ontario, Canada

³Baylor College of Medicine and Michael E. DeBakey Veterans Affairs Medical Center, Houston, TX

Background: Surgical safety literature to date has largely focused on improvements in treatment, rather than diagnosis. Conversely, the vast majority of research in diagnostic error has been conducted in medical disciplines. We sought to describe the characteristics of diagnostic error and its impact on patient harm in surgical settings.

Methods: We performed a retrospective analysis of closed medicolegal cases from 2014-2018 at the Canadian Medical Protective Association (CMPA), a national not-for-profit mutual defense organization representing >95% of all Canadian physicians. The CMPA maintains a repository of medicolegal data about civil legal actions and complaints to regulatory authorities and hospitals. Each case represents a matter brought to the CMPA by a physician seeking medicolegal advice.

Results: We identified 387 surgical cases that involved diagnostic error. The majority of cases (61.2%) occurred in the inpatient setting. The most common surgical specialties involved were general surgery (37.8% of cases), obstetrics and gynecology (16.2%), orthopedic surgery (11.9%), urology (9.7%), and plastic surgery (7.0%). Errors were classified as diagnoses that were missed (54.4% of cases), delayed (34.1%), and wrong (17.6%). One-third of cases occurred in the pre-operative phase, with errors most commonly involving neoplasms. Cases occurring in the intra-operative phase made up 31.0% of cases, with errors most commonly involving lack of injury recognition during surgery. Almost 45% of cases occurred in the post-operative phase, with errors most commonly involving delayed identification of complications related to surgical injury, including failure to recognize subsequent clinical deterioration. Just over 54% of patients experienced at least moderate harm, with 14.4% of cases resulting in death. Errors were attributed to the provider (82.1% of cases), health care team (50.3%), and system (11.9%). At the provider level, deficiencies in clinical decision making, failure to follow-up on complications, and loss of situational awareness were the most common contributing factors. At the health care team level, communication breakdown with patients and physicians, and documentation issues were the most common contributing factors. At the system level, resource issues, protocol, policy, and procedure issues, and office issues were the most common contributing factors.

Conclusion: Diagnostic error involves most surgical disciplines and occurs across all phases of surgical care, resulting in significant patient harm. Given that a substantial proportion of these errors were unique to the care of surgical patients, additional research is needed to characterize epidemiology and explore potential solutions specific to surgical disciplines.

A National Program for Measurement of Timeliness of Test Result Communication to Patients: Communication Performance and Next Steps

A. Meyer ¹, T. Scott¹*, H. Singh¹

*(Presenting Author)

¹Michael E. DeBakey VA Medical Center and Baylor College of Medicine, Houston, TX

Background: Failure to follow-up on abnormal test results remains a persistent problem leading to diagnosis and treatment delays. The Department of Veterans Affairs (VA) recently implemented a national program for measurement of timeliness of test result communication to patients. We measured communication performance to establish next steps for practice and policy.

Methods: A national VA policy guided performance measurement. Measures included timeliness of communication to patients of abnormal (within <7 days if action required; within <14 days if no action required), normal (within <14 days), and all test results (within 30 days). Measurement guidance for manual data collection was implemented through the External Peer Review Program (EPRP). Computerized algorithms identified random patients with pre-specified outpatient laboratory/radiology/pathology tests (Table 1) every quarter. Then, charts were abstracted to evaluate documentation of timely patient notification of test results and feedback of data to local VA facilities for quality improvement/benchmarking. We examined these data and one item from the Survey of Healthcare Experiences of Patients (SHEP; “in the last 6 months, when this provider ordered a blood test, x-ray, or other test for you, how often did someone from this provider’s office follow up to give you those results?”) for January-December 2019.

Results: EPRP measures show that communication of test results occurred in a timely manner for 5,925 of 8,372 abnormal tests (70.8%); 9,472 of 11,784 normal tests (80.4%); and 16,479 of 20,156 all tests combined (81.8%). Performance varied by facility, with ranges between 45.5-93.9% for abnormal results, 52.9-96.8% for normal results, and 58.5-94.8% for all combined tests. SHEP data show that of 213,464 patients surveyed, 8.2% reported results were “never” communicated, whereas 6.3% said “sometimes”, 15.8% said “usually”, and 69.8% said “always”. This also varied by facility, with facilities ranging between 2.6-23.5% for “never”, 1.8-10.9% for “sometimes”, 10.6-22.5% for “usually”, and 50.7-83.5% for “always”.

Conclusion: A national VA performance measurement system reveals substantial gaps in communicating test results to patients in a timely manner, also supported by patient surveys. These measures could be useful for benchmarking and accountability and testing the impact of feedback and improvement interventions in both VA and non-VA settings.

Associations Between Individual Physician Characteristics, Diagnostic Accuracy, and Calibration

A. Meyer ¹, E. Patterson², D. Sittig³, Z. Sargsyan⁴, H. Singh¹

¹Michael E. DeBakey VA Medical Center and Baylor College of Medicine, Houston, TX

²The Ohio State University, Columbus, OH

³UTHealth, Houston, TX

⁴Baylor College of Medicine, Houston, TX

Background: Individual physician characteristics have been proposed as important, yet understudied factors contributing to both diagnostic performance and calibration (alignment between diagnostic accuracy and confidence). We examine the relationships between certain physician characteristics, diagnostic accuracy, and calibration to better understand diagnostic reasoning performance.

Methods: We conducted an online experimental study and recruited general internal medicine physicians to diagnose three randomly ordered patient vignettes of varying difficulty. For each case, physicians provided up to 10 possible diagnoses at two steps (1-after chief complaint and history were provided; 2-after all additional diagnostic information was given). Physicians included judgements to indicate their confidence (on a scale of 1-10) that the diagnoses they gave would be the correct, eventual diagnoses. They also answered a series of questions to measure certain individual characteristics: risk-aversion, stress from uncertainty (using subscales for anxiety due to uncertainty and concern about bad outcomes), humility (using subscales for openness to revising one’s viewpoint and lack of intellectual overconfidence), gender, type of work institution (academic versus non-academic), and experience (years since residency). Diagnostic accuracy was scored from 0-10 based on similarity between diagnoses provided by participants and correct diagnoses. Calibration was examined using two measures: calibration alignment (using the “C-statistic”) and miscalibration direction (using the “O/U index”). Case difficulty, step, and all characteristics above were regressed onto diagnostic accuracy and both measures of calibration using linear or logistic regression with bootstrapping (n=1000) and Bonferroni corrections for multiple comparisons.

Results: Analyses of data from 38 physicians (36.8% female) show that diagnostic accuracy was higher for lower difficulty cases (p<.001), in the later step (p=.003), for males (p=.002), and for individuals working mostly in academic institutions (p=<.001). Calibration alignment (C-statistic) did not differ significantly for any factor measured. While miscalibration direction (O/U index) was mostly in the underconfident direction (physicians showed under confidence in 69.4% of cases, perfect calibration in 5.4% of cases, and overconfidence in 25.2% of cases), overconfidence occurred more often in difficult cases (p<.001) and with females (p=.003).

Conclusion: Some individual physician characteristics may be associated with better diagnostic accuracy and calibration. For example, diagnostic accuracy appears to be higher for physicians working mostly in academic institutions, possibly indicating a need to increase access to knowledge resources and expertise for non-academic clinicians, who often do not directly benefit from academic networks and associated activities. Further research should validate these findings and examine additional factors associated with accurate and well-calibrated diagnostic reasoning.

The Quest for a High-Quality Diagnosis: Measuring and Comparing Acute Clinical Encounters and Clinical Documentation

M. Khazen ¹, E. Sullivan², M. Mirica³*, J. Ramos⁴, M. Linzer⁵, A. Olson⁶, G. Schiff³

*(Presenting Author)

¹University of Tel Aviv, Tel Aviv, Israel

²Suffolk University, Boston, MA

³Brigham and Women’s Hospital, Boston, MA

⁴Emory Medical School, Atlanta, GA

⁵Hennepin Healthcare, Minneapolis, MN

⁶University of Minnesota Medical School, Minneapolis, MN

Background: The quest to craft standardized diagnostic safety measures and uncover diagnostic improvement opportunities has been challenging, and efforts to evaluate and improve diagnosis quality have struggled to accurately capture and measure diagnosis processes and outcomes, particularly for assessment of acute symptoms. Researchers also struggle with the question of how closely written documentation mirrors the actual clinical encounter.

Methods: We developed and tested a tool for measuring diagnostic process quality (DPQ), addressing the following elements: chief complaint, differential diagnosis, psychosocial/contextual information, uncertainty, etiologies, contingencies, testing, time frames, red flags, don’t miss diagnoses, and articulation of any diagnostic pitfalls. The DPQ tool was iteratively developed and refined by evaluating a series of EMR notes’ assessments as well as subsequently analyzing transcribed audiotapes of clinical encounters with detailed review of the clinical notes for the corresponding visits. Transcripts were entered into NVivo12 qualitative software and analyzed inductively and deductively, using content and discourse analysis. Two social science PhD researchers independently coded each transcript, and then reconciled coding decisions. The larger team of researchers and clinicians revised the coding scheme, ensured consistent coding application, and reached consensus on coding disagreements.

Results: Comparison between the transcribed audiotaped encounters and corresponding clinical note found concordance between note and encounter in addressing chief complaint (96%), red flags (92%), etiologies (88%), uncertainty (71%), contingencies (71%), time frames (60%), don’t miss diagnoses (58%), discussion of diagnostic testing (57%), differential diagnosis (54%), psychosocial/contextual (35%), pitfalls (eg, failure to address chief complaint) (7%). Of the 28 encounters and corresponding notes, percentages of elements present in the encounter while absent from the note were: differential diagnosis (25%), psychosocial (11%), etiologies (4%), uncertainty (25%), don’t miss diagnosis (21%), time frames (11%), pitfalls (7%). The element most often present in the note but absent from the encounter was “contingencies” (22%; 6 of 28 notes included a generic sentence to return if worse). The element most absent from both encounter and note was mention/addressing of “pitfalls” (86%).

Conclusions: Developing a tool to assess the diagnostic process afforded an opportunity to conceptualize and measure key elements of diagnostic assessment of clinical encounters for acute problems, as well as concordance between actual encounter transcripts and written documentation. Our findings suggest reassuringly overall good concordance as well as highlight selected key areas of significant differences which likely fruitful areas for future work in understanding and improving diagnosis.

Validation of Skin Surface Thermal Imaging to Differentiate Cellulitis from Pseudocellulitis in the Emergency Department

M. Pulia ¹, R. Schwei¹, E. Harwick¹, A. Haleem¹, J. Hess¹, R. Glinert¹, T. Keenan¹, J. McBride¹, R. Redwood¹

¹University of Wisconsin-Madison, Madison, WI

Background: Cellulitis is misdiagnosed in up to 30% of cases due to mimic conditions termed pseudocellulitis. The resulting overuse of antibiotics is a threat to patient safety and public health. Surface thermal imaging has been proposed as a tool to help differentiate cellulitis from pseudocellulitis due to an expectation of increased tissue temperature cellulitis. The study objective was to compare skin surface temperature measurements between patients with cellulitis and pseudocellulitis.

Methods: We prospectively enrolled patients presenting to the emergency department (ED) with acute dermatologic lower extremity complaints that involved visible erythema. Using a thermal imaging camera, the maximum temperature value (Tmax) for the affected area of skin and corresponding area on an unaffected limb were captured. The Tmax gradient between the affected and unaffected limb was also calculated. Gold standard diagnosis (cellulitis versus pseudocellulitis) was determined by consensus of a blinded, multidisciplinary physician review panel (two infectious disease, two dermatologists and two emergency medicine). Differences in temperature variables (Tmax and Tmax gradient) between cellulitis and pseudocellulitis were compared using t-tests.

Results: The sample included 204 participants, 59% male with an average age of 57 years. Based on expert panel consensus diagnosis, 92 (45%) of the participants had cellulitis. The cellulitis group had an average Tmax of 33.2°C and 30.2°C for affected and unaffected skin respectively, which was a significant difference of 2.9°C (CI: 2.5 to 3.6; p<0.001). The difference in the Tmax gradients between patients with cellulitis and pseudocellulitis was 2.08°C (CI: 1.46-2.70; p<0.001).

Conclusion: This represents the largest validation study of skin surface temperature differences between cellulitis and pseudocellulitis. Significant difference in temperature gradients between cases of cellulitis and pseudocellulitis suggests thermal imaging could be a useful diagnostic adjunct that can help differentiate these conditions. Such a modality could be particularly helpful in the ED setting where providers must balance diagnostic uncertainty with antimicrobial stewardship principles. Future work will identify the best performing temperature variables and determine optimal cutoff values for use in diagnostic algorithms.

Evaluation of Diagnostic Apps for Myocardial Infarction and Other Causes of Chest Pain: Informing Patient Use

S. Raman ¹, H. Fraser²

¹Warren Alpert Medical School of Brown University, Providence, RI

²Brown University, Providence, RI

Background: Symptom checker (SC) apps use symptoms entered by patients to output a list of possible diagnoses. Although these apps may influence how patients seek urgent care, they remain under-evaluated. This study used a database of patients’ actual symptoms and final diagnoses to approximate home use. Ada, WebMD, and Isabel SC apps were evaluated for accuracy in diagnosing both urgent cardiac and low risk noncardiac cases presenting with chest pain.

Methods: Data from patients presenting with chest pain and other symptoms to emergency departments (Kennedy et al, 1996, European Heart Journal, V 17; 8: 1181-1191) was used to create a cardiac symptom dataset (D1) and a noncardiac symptom dataset (D2). 50 cases (S1) randomly selected from D1 and 29 low risk, noncardiac cases (S2) purposely selected from D2, were entered into WebMD, Isabel, and Ada. Matches between the final diagnosis and the top 5 app suggested diagnoses were identified.

Results: Results for the top 5 suggestions per app were as follows. The S1 diagnostic categories included 11 acute myocardial infarctions (AMIs), 9 unstable anginas (UA), 11 stable anginas, and 19 categorically noncardiac cases. The noncardiac low risk sample had no cardiac diagnoses. WebMD: Acute cardiac ischemia (UA and AMI)- sensitivity 100%, specificity 13.3%, PPV-43.5%, NPV-100%. Identified 100% of AMIs, 100% of UAs, 0% of stable anginas. For noncardiac diagnoses in S2, Sensitivity was 24.1%, with AMI as the first suggestion for 34.5% of cases. Isabel: Acute cardiac ischemia - sensitivity 75%, specificity 83.3%, PPV-75%, NPV-83.3%. Identified 100% of AMIs, 44.4% of UAs, 90.9% of stable anginas. For noncardiac diagnoses in S2, Sensitivity was 24.1%, with AMI as the first suggestion for 17.2% of cases. Ada: Acute cardiac ischemia - sensitivity 95%, specificity 56.7%, PPV-59.4%, NPV-94.4%. Identified 100% of AMIs, 88.9% of UAs, 18.2% of stable anginas. For noncardiac diagnoses in S2, Sensitivity was 48.3%, with AMI as the first suggestion for 34.5% of cases.

Conclusions: All three SCs identified 100% of AMIs and were highly sensitive to acute cardiac ischemia. While WebMD identified all UAs and cases of gastro-intestinal disease, Isabel identified the most stable anginas, and Ada identified the most noncardiac diagnoses. This pilot study found all 3 apps provided generally safe advice for patients with symptoms of AMI but were risk adverse and had limited ability to identify noncardiac causes even in younger, low risk patients.

Investigating the Impact of Interruptions Within a Unique Set of Medical Cases

J. Sloane ¹, N. Chan², B. Newell¹, C. Donkin¹

¹University of New South Wales, Sydney, Australia

²Vident Medical Center

Background: Previous research by Meyer, Payne, Meeks, Rao, and Singh (2013) explored the relationship between physician accuracy and confidence and found confidence was much higher than expected for harder medical vignettes. We look to extend this research in two ways. First, we interrupt people while they receive case information, as we know interruptions are a common occurrence, with emergency physicians and interns being interrupted 7.1 times per hour on average (Westbrook et al., 2010). Second, we include a unique type of cases that we refer to trick cases, which are designed to encourage a common, but incorrect diagnosis.

Methods: 26 year 5 and 6 UNSW medical students participated in this study. There was a total of 6 medical cases, including 2 easy, 2 hard, and 2 trick cases. Participants completed all 6 cases. After each case, participants were asked to provide up to three possible final diagnoses. Then, they were asked to rate their confidence in their #1 diagnosis and finally to rate how difficult they found the case to be. Each participant completed 3 cases in the baseline condition and 3 cases in the interruption condition, which were randomly assigned. Cases in the baseline condition were presented on a single page, while cases in the interruption condition were presented across 4 pages, during which there was a 5-second interruption task that asked participants to search a visual scene for a particular item.

Results: Unsurprisingly, accuracy was highest for the easy cases (M=89%) compared to hard (M=4%) and trick (M=19%) cases. Despite the large difference in accuracy between the case types, confidence did not differ as much, with average confidence ranging from 45 to 57 out of a scale that ranged from 0 to 100. Furthermore, the interruptions affected each of the case types in different ways. Those who were interrupted on easy and hard cases were less confident in their diagnoses compared to those who were not interrupted. Although the evidence is less conclusive, trends in the data suggest the reverse was true for trick cases, where those who were interrupted were more confident compared to those who were not.

Conclusions: These results give us insight into the diagnosis process within unique types of medical cases and how interruptions affect the processing of information and the type and incidence of errors in decision making.

Systematic Review and Meta-analysis of the Affect of Cognitive Reasoning Tools on Diagnostic Accuracy

J. Staal ¹, J. Hooftman², S. Mamede³, J. Alsma¹, M. A. Frens¹, W. W. van den Broek³, L. Zwaan³

¹Erasmus Medical Center, Rotterdam, Netherlands

²Amsterdam UMC, iMERR, EMC

³iMERR, Erasmus Medical Center, Rotterdam, Netherlands

Background: Preventable diagnostic errors are a large burden on healthcare, and many interventions have been proposed to reduce these errors. Cognitive reasoning tools (defined as tools that foster additional or different thinking about a diagnosis, such as checklists) have been successful in preventing medical errors, especially in surgical procedures. However, it is unclear whether they can also meaningfully reduce diagnostic errors in practice because rigorous testing of the effectiveness of such tools is sparse. In this systematic review and meta-analysis, we aim to estimate the effect of cognitive reasoning tools on improving diagnostic performance among healthcare professionals and students, and to identify factors that were associated with larger improvements.

Methods: Controlled studies that assessed whether cognitive reasoning tools improved the diagnostic accuracy of individual healthcare professionals or students in a clinical setting were eligible. In April 2020 we searched Embase (1971-2020), Medline All (1946-2020), Web of Science Core Collection (1975-2020), Cochrane Central Register of Controlled Trials (1992-2020), and Google Scholar (1992-2020), supplemented with hand searching. Meta-analysis was performed using a random-effects model.

Results: Of 2155 identified studies, 21 studies with data from 1858 participants were included for meta-analysis. Cognitive reasoning tools led to a small improvement in diagnostic accuracy (g = .31, 95% confidence interval: .18-.44) (Figure 1). This pooled estimate showed considerable heterogeneity (I2 = 99%). Two study characteristics were identified that were associated with larger effects: (1) tool type: content-specific tools were more effective than general tools and (2) moment of intervention: tools were more effective for verification after initial diagnosis than when used during diagnosis. Sensitivity analyses separating studies based on these characteristics did not reduce heterogeneity.

Conclusion: Cognitive reasoning tools resulted in small improvements in diagnostic accuracy for healthcare professionals and students. Larger effects were associated with tools focusing on task content rather than reasoning processes. Additionally, verifying an initial diagnosis with a tool was more effective than using the tool during initial diagnosis. However, even when taking these factors into account, the large heterogeneity between study effects remains unexplained. The factors predicting tool effectiveness should be better understood to meaningfully reduce diagnostic error.

The Impact of System and Diagnostic Errors for Medical Litigation Outcomes in Orthopedic Surgery

N. Yamamoto ¹, T. Watari²*, A.Shibata³, T. Noda⁴, T. Ozaki⁵

*(Presenting Author)

¹Miyamoto Orthopedic Hospital

²Shimane University

³Yodogawa Christian Hospital

⁴Kawasaki Medical School

⁵Okayama University

Background: Medical litigation resulting from medical errors and malpractice has a negative impact on medical safety and health economics for both patients and medical practitioners. In medical litigation involving orthopedic surgeons, we aimed to identify factors contributing to plaintiff victory (orthopedic surgeon loss) through a comprehensive assessment.

Methods: This retrospective study included 166 litigation claims against orthopedic surgeons using a litigation database in Japan. We evaluated the sex and age of the patient (plaintiff), initial diagnosis, diagnostic error, system error, the time and place of each claim that led to malpractice litigation, the institution’s size, and clinical outcomes. The main outcome was the litigation outcome (acceptance or rejection) in the final judgment. Acceptance meant that the orthopedic surgeon lost the malpractice lawsuit, and rejection meant the plaintiff lost. We conducted multivariable logistic regression analyses to examine the association of factors with an accepted claim (orthopedic surgeon loss).

Results: The median age of the patients was 42 years (interquartile range [IQR] = 23-58), and 65.7% of the patients were male. The litigation outcome of 85 (51.2%) claims was acceptance (orthopedic surgeon loss). The adjusted median indemnity paid was $151,818 (IQR = 32,865-331,818). The multivariable analysis showed that diagnostic error, system error, sequelae, inadequate medical procedure, and follow-up observation were significantly associated with the orthopedic surgeon losing the lawsuit. In particular, claims involving diagnostic errors were more likely to be acceptance claims, in which the orthopedic surgeon lost (adjusted odds ratio 16.7, 95% CI: 4.7 to 58.0, p < 0.001). All of the claims in which the orthopedic surgeon lost were associated with a diagnostic or system error, with the most common one being system error.

Conclusions: Regarding litigation outcomes, system errors and diagnostic errors were significantly associated with acceptance claims (orthopedic surgeon losses). Since these are modifiable factors, it is necessary to take measures not only for individual physicians but also for the overall medical management system to enhance patient safety and reduce the litigation risk of orthopedic surgeons.

Development of an Electronic Quality Measure of Late-stage Cancer Diagnosis

A. Zimolzak ¹, D. Murphy¹, D. Upadhyay², L. Wei³, U. Mushtaq¹, P. Jolly³, S. Korukonda², G. Lyratzopoulos⁴, G. Abel⁵, A. Offner¹, H. Singh⁶

¹Baylor College of Medicine, Houston, TX

²Geisinger Health System, Danville, PA

³Michael E. Debakey VA Medical Center, Houston, TX

⁴University College London, London, United Kingdom

⁵University of Exeter, Exeter, United Kingdom

⁶Michael E. Debakey VA Medical Center & Baylor College of Medicine, Houston, TX

Background: Advanced stage at cancer diagnosis is associated with worse outcomes. Several factors including race, access to care, and insurance type, are associated with later stage at diagnosis, suggesting care disparities. The percentage of cancer patients diagnosed at early stage is already reported publicly in England. Building on this work, we developed and tested an electronic outcome measure of late-stage cancer diagnosis in two large healthcare systems in the US.

Methods: We queried EHR and cancer registry data from 2019 for lung cancer and colorectal cancer (CRC) in 130 Veterans Affairs (VA) facilities, as well as Geisinger, a health system in rural Pennsylvania. We retrieved stage from the registry and defined the measure as the percentage of late-stage (stage III and IV) patients (i.e. numerator) among total cancer patients (denominator). Two clinicians reviewed randomly sampled records and determined whether the query correctly identified the stage. We classified chart review discrepancies as critical (which affected the measure’s numerical value) or minor (no effect on measure).

Results: At VA, 22,873 of 50,686 (45.1%) lung cancer patients had late-stage diagnosis (11,780 unknown/unstaged cases were excluded from numerator and denominator). Stratified by VA medical center, the 25th percentile of this measure was 42.3%, and the 75th percentile was 51.2%. For CRC, 8,687 of 23,905 (36.3%) patients had late-stage diagnosis (4,496 without staging were excluded). Stratified by VA medical center, the 25th percentile was 32.0%, and the 75th percentile was 40.6%. At Geisinger, 97 of 268 (36.2%) patients had late-stage CRC diagnosis (41 without staging were excluded). At VA, of 20 CRC cases reviewed, 1 had a critical discrepancy (stage unknown in registry but stage IV per clinical notes), and 5 had a minor discrepancy. Of 19 lung cases reviewed, none had a critical discrepancy, and 7 had a minor discrepancy. At Geisinger, of 20 CRC records reviewed, 2 had critical discrepancies (1 with stage 0 per registry but stage IV per notes, and 1 with unknown stage per registry but stage III per notes), and 5 had minor discrepancies.

Conclusions: An electronic measure of late-stage cancer diagnosis can be applied to integrated EHR and cancer registry data. Preliminary review shows few critical discrepancies but a sizable proportion with missing registry stage data. We will now determine the e-measure’s positive predictive value for potential missed opportunities and identify modifiable contributory factors. We will also analyze the measure’s statistical reliability across hospital location and time.

REACT: Rapid Evaluation Assessment of Clinical Reasoning Tool

B. Peterson ¹, C. Magee¹, J. Dreicer¹, K. Mutter¹, L. Parsons¹, J. Martindale², K. Warburton¹, A. Parsons¹ (Senior Author)

¹University of Virginia Health System, Charlottesville, VA

²UVA School of Medicine, Charlottesville, VA

Purpose: Clinical reasoning encompasses the process of data collection, synthesis, and interpretation to generate a working diagnosis, facilitating management decisions. Situated Cognition Theory suggests that knowledge is relative to contextual factors, and clinical reasoning in urgent situations adds pressure of consequential, time sensitive decision-making with respect to diagnosis and management. These unique aspects of urgent clinical care may limit the effectiveness of traditional efforts to assess, teach, and remediate clinical reasoning utilizing deliberate and analytical processes.

Description: Using two validated frameworks: 1) SIDM’s Assessment of Reasoning Tool (ART) to assess clinical reasoning through oral presentations and 2) the Association of American Medical College’s (AAMC) Entrustable Professional Activity 10 (EPA10) designed to formatively assess a learner’s recognition of patients requiring urgent care, a multidisciplinary group of clinicians trained to remediate clinical reasoning and with experience in urgent clinical care encounters iteratively designed the novel Rapid Evaluation Assessment of Clinical Reasoning Tool (REACT). REACT is a behaviorally anchored assessment tool scoring five domains used to provide formative feedback to learners evaluating patients in urgent clinical situations. A pilot study was performed during the 2021 Intern Readiness Course (IRC) for fourth year medical students. Simulated patient interactions across a range of urgent clinical scenarios were recorded. Learners were scored using REACT by a separate multidisciplinary group of clinician educators with no additional training in the clinical reasoning process. REACT scores were analyzed for internal consistency across raters and observations.

Outcomes: Overall internal consistency for the 41 patient simulations as measured by Cronbach’s Alpha was .84. A mean weighted Kappa statistic was used to assess overall score interrater reliability. Substantial reliability was observed at .72.

Discussion: REACT is a novel, reliable and valid tool for clinician educators from diverse specialties to formatively assess a learner’s clinical reasoning during urgent clinical encounters. Future directions include deploying REACT in non-simulated patient care environments.

Significance of Findings: To our knowledge, this is the first validated tool designed for assessment and formative feedback of a learner’s clinical reasoning during urgent clinical situations. This tool facilitates teaching and remediation of clinical reasoning with the goal of reducing diagnostic and management errors to limit patient harm.

Natural Language Processing for Detection and Reporting of Findings Requiring Follow-Up in Ra

J. Domingo ³, P. Soni², G. Galal², V. Mukhin², J. Huang², S. Caron¹, S. Xinos¹, R. Slavicek¹, P. Creamer¹, C. Altman¹, M. Massa¹, K. Kadiyam¹, B. Ware¹, M. Suna¹, H. Gwardys¹, T. Bayer¹, P. Salamone¹, J. Gilstrap¹, T. Brooks¹, T. Byrd², L. Yuen², M. Etemadi²

¹Northwestern Medicine, Chicago, IL

²Northwestern University, Evanston, IL

³Northwestern Memorial, Chicago, IL

Problem Statement: Medical diagnostic imaging studies frequently detect findings which require further evaluation. Whether these are expected or unrelated to the original indication for imaging, appropriate follow-up of these findings can prevent undetected progression of pathology. However, follow-up is often inconsistent due to lack of standard guidelines and workflows for reporting of findings, leading to suboptimal patient outcomes and increased healthcare costs.

Intervention: A deep learning natural language processing framework for detection and reporting of findings requiring follow-up was developed and implemented across an academic health system. The framework includes separate deep learning models trained to identify findings of lung and adrenal nodules from radiology reports based on retrospectively collected radiology reports manually annotated by clinical experts. Automated screening of radiology reports is accomplished via integration with the electronic health record (EHR). Upon detection of a finding, alerts are generated to notify ordering physicians and present a workflow for ordering and tracking completion of follow-up studies as needed.

Findings to Date: As of May 2021, over 11,000 imaging studies have been flagged as containing lung nodules since implementation of the framework in December 2020, representing a 4.98% rate of lung finding occurrence on relevant imaging studies and an average of 66 findings flagged per day. Evaluation of adrenal nodule detection was more limited, taking place over three months and yielding over 300 positive imaging studies but with a higher false positive rate than lung nodules. On manual prospective evaluation on a subset of clinical data, the model had an overall accuracy of 93% (specificity: 87%, sensitivity: 99%) for detection of lung and adrenal nodules, which was comparable to performance on retrospectively collected data.

Lessons Learned: We demonstrate that a deep learning natural language processing approach is able to detect findings on radiology reports with high accuracy, and additionally show the accuracy and scalability of such a system upon implementation in the EHR. We highlight difficulties pertinent to detection of less common findings, such as adrenal nodules, and the importance of robust collection of training data for deep learning models. While we report primarily on the detection of lung and adrenal nodules, other findings, including ovarian lesions and hepatic steatosis, are currently being implemented in this system. Further evaluation of the impact on physician workflow and patient outcomes is needed to demonstrate the clinical impact of this work.

Construction of the Illness Script of COVID-19

J. Sader ¹, M. Nendaz², M. Coen², N. Junod Perron³, M. Audétat³

¹Unit of Development and Research in Medical Education (UDREM), Geneva, Switzerland

²Department of medicine HUG

³Faculty of Medicine UNIGE

Background: Illness script are organized mental representations (summary) of a disease. They are built on the triad of enabling (predisposing) conditions (e.g., age, medical history), fault (pathophysiology), and consequences (i.e., how the disease manifests itself). Illness scripts acquisition and enrichment relies on clinical practice, and less on education. Moreover, they are organized for action. Pertaining to COVID-19, health professionals and scientists cannot rely on known scripts of this disease. Instead, their scripts develop and modify along with the growth of new knowledge and the progress of their experience in regard to this new disease. Our research aimed at understanding how the script of COVID-19 developed amongst health-care professionals, at a time when information and guidelines and were in continuous development.

Methods: We conducted focus groups and interviews with physicians with different levels of experience and involvement in clinical settings. The data were analyzed by a qualitative methodology based on grounded theory. This analysis allowed us to contrast the “usual” construction of clinical scripts versus that of COVID-19.

Results: Our preliminary results show that the illness script of COVID-19 varies greatly amongst physicians, and notably in terms of richness (i.e., the capacity to generate relevant information on all three illness script components). Interns, for examples, had more direct (“frontline”) exposition to patients, enabling them to create script that were particularly rich in the “Consequences” (i.e., sign and symptoms) component. Attending physicians, instead, a more global view of the situation including the effects of the pandemic on the institutional landscape; their illness scripts were richer in the “Fault” (i.e. pathophysiological mechanisms) and the “Enabling Conditions” (patient features and contextual factors) components. ICU physicians, for examples, mainly dealt with intubated patients, and communication with families was mainly reliant on video- phone-calls, and devoid of nonverbal-cues since families were not allowed in the hospital.

Conclusion: The illness script of COVID-19 was -and still is- changing, as the situation is constantly evolving. New knowledge and new guidelines are arising constantly, so that the script is rooted in uncertainty and instability. Further studies should assess how this illness script will stabilize in the future and how clinical reasoning may be influenced by this evolution.

Connecting the Dots: From Assessment of Implicit or Explicit Bias to Application in Diagnosis

B. Krauss ¹, N. Sivin-Kachala², J. Pell², D. Iskhakov², R. Pandya²

¹City University of New York School of Public Health, New York City, NY

²Hunter College, New York City, NY

Background: A scoping review of literature on prevalent cognitive biases in medical diagnosis and strategies that inhibit such biases discovered numerous articles assessing and documenting implicit and explicit biases against multiple groups. Groups included those who present as over or underweight, ethnic and sexual minorities, age groups, genders, those with low-income, those with multiple comorbidities, and those with mental health problems.

Methods: Given the evolution of research on cognitive biases, the review was limited to US and Canadian medical diagnosticians, students through advanced career, and their patients. The authors used a search string of provider-, bias-, and subject matter-related terms to identify and download over 4000 non-duplicated abstracts, published 2005 through 2020 from MEDLINE/PubMed and APA PsycINFO via the EBSCO interface. The three co-authors read and screened abstracts independently for inclusion in a focused scoping review on the prevalence of cognitive bias in medical diagnoses and policies/guidelines/interventions to address such bias, followed by review of included full texts. The team met weekly to resolve conflicts by consensus.

Results: Nearly half of all articles about bias toward groups concerned effects on treatment only and did not address diagnosis. Among the articles focused on diagnosis, most often theoretical or policy recommendations were made, followed by documentation of bias prevalence, and studies of medical education interventions. Medical students were most frequently assessed, with changes in attitude and promise of practice change, rather than actual practice change, the most common outcomes. Infrequently, were patient biases (e.g., in reporting of symptoms or conditions) or patient experiences with bias (e.g., former experiences leading to delay of healthcare access and diagnosis) documented. A very few research articles had specific recommendations resulting from their findings. Examples are specific instruction on how to sensitively conduct a physical examination of a patient with apparent over or underweight; changing the “shadow culture” of medical education which can model less-than-optimal practice; and broadening and deepening diagnostic knowledge and procedures to override potential bias. More frequently, possible moderators of diagnostic practice, e.g., ratio of provider/patient talk time, were identified.

Conclusion: The prevalence of implicit and explicit bias is well-documented. The field of diagnosis would benefit from well-designed studies of interventions to override such biases that have improvement in diagnostic practice and accuracy as outcomes. Research participation beyond student and early career diagnosticians, and inclusive of patient perspectives, would be informative.

Development of an Electronic Diagnostic Quality Measure Based on Emergency Cancer Presentations in the US

P. Kapadia ¹, A. Zimolzak², D. Murphy², D. Upadhyay³, U. Mushtaq², P. Jolly², S. Korukonda³, L. Wei², G. Lyratzopoulos⁴, G. Abel⁵, A. Offner², H. Singh²

¹Baylor College of Medicine, Houston, TX

²Michael E. DeBakey VAMC, Houston, TX

³Geisinger, Danville, PA

⁴University College London, London, United Kingdom

⁵University of Exeter Medical School, Exeter, United Kingdom

Background: Compared to routine cancer diagnosis, cancer diagnosis via an emergency presentation (diagnosis following unplanned hospitalization or emergency department [ED] visit) has been associated with worse clinical and patient-reported outcomes in international literature and may be linked to missed opportunities for earlier diagnosis (MODs). Emergency cancer diagnosis remains under-studied in the US. Increasingly available large-scale integrated encounter-level data can now outline diagnostic pathways across outpatient, ED, and inpatient settings and facilitate the development of electronic measures (e-measures) of diagnostic quality. We developed two outcome e-measures (lung cancer, colorectal cancer [CRC]) to detect emergency cancer presentations in two large US health systems.

Methods: We developed an algorithm to identify patients with new cancer diagnoses (measure denominator) using electronic health record data from >130 Department of Veterans Affairs (VA) facilities and Geisinger, a health system in rural Pennsylvania. The algorithm then applied encounter-level ICD codes and cancer registry data to identify patients diagnosed with cancer for the first time within 30 days of an unplanned hospitalization or ED visit (measure numerator). We excluded patients without recorded primary care encounters 2-years prior to presentation. Two clinicians at both sites reviewed randomly sampled records from the numerator to determine whether the algorithm correctly identified emergency cancer presentations. We classified anomalies in record identification as critical discrepancies (affecting the measure’s numerical value) or minor discrepancies (i.e., slightly different stage not affecting numerical value).

Results: Measure numerators and denominators for 2019 were 562/2332 (24.1%) for CRC and 1278/6525 (19.6%) for lung cancer at the VA and 26/309 (8.4%) for CRC at Geisinger (lung cancer pending). For CRC, all 30 reviewed records (20 VA, 10 Geisinger) were true emergency presentations. For lung cancer, 15 of 20 reviewed records (all VA) were true emergency presentations; 2 of 20 were true incidental diagnoses (patients diagnosed after presenting for unrelated causes such as vehicle accident and acute pancreatitis); and 3 of 20 were critical discrepancies (2 with prior diagnosis of lung cancer and 1 without mention of malignancy upon review). Preliminary review of the 10 Geisinger CRC records indicated potential MODs in most (6 of 10) cases.

Conclusions: Preliminary work suggests that a quality measure based on emergency cancer presentation is feasible and appears to represent MODs. Further work includes algorithm refinement to reduce critical discrepancies, reviews to validate measure accuracy and calculate positive predictive value for MODs and analyzing emergency presentations to identify areas of improvement in cancer diagnosis.

Inequalities in the Impact of COVID-19 on Cancer Referrals: A Review of Service Data

S. Wallace ¹, A. Adebamowo¹, E. Sherwood¹, I. Basnett¹, M. Warren¹, A. Wong¹

¹Barts Health NHS Trust, London, United Kingdom

Background: The impact of COVID-19 on healthcare usage was large and led to significant reductions in cancer referrals. As COVID-19 has not affected all demographic groups equally, we undertook this service data review to identify inequalities in impact of COVID-19 on referrals.

Methods: The Trust cancer database was used to identify referrals into Trust cancer services and obtain demographic/service data, and Trust electronic records were used to increase completeness of ethnicity recording. When comparing 2019 and 2020, we used data from March-November. Data were extracted in early January; therefore, December was excluded so that delays in data entry did not affect results.

Results: Comparing March-November 2019 with the same period in 2020, we observed a significantly greater reduction in referrals among men (26% (95%CI 25%,28%)) compared with women (22%(21%,23%)). Patients of Asian (26%(24%,28%)) and ‘Other’ ethnic groups (27%(24%,30%)) have seen the greatest overall reduction in referrals. In April 2020 both sexes and all ethnic groups dropped to a nadir of approximately 45% referrals compared with 2019, coinciding with the COVID peak. When examining the percentage recovery of referrals by month (Figure 1), Black/Black British men and Asian/Asian British women had a slower recovery in referral numbers compared with other Ethnic groups. For breast cancer referrals from the national screening program (N=464 in 2019) there has been a greater drop among Asian (79%) and Black women (63%) compared to White women (52%) in March to November. Conversely cervical cancer screening referrals have dropped by 12% among White women but increased by 36% among Black and Asian women (N=428). Adults have shown a consistent pattern of both greater reductions and poorer recovery among older age groups which persisted to December 2020 (Figure 1).

Conclusions: The impact of COVID-19 on cancer referrals has not been equal across demographic groups; the patterns observed are important to direct our community-based work and with primary care partners to encourage presentation with symptoms and screening participation during the recovery phase. However, any future COVID-19 peaks may cause similar inequalities and careful monthly monitoring of referral patterns will enable early action.

Needle in a Haystack: Diagnosis of Subarachnoid Hemorrhage

D. Amin ¹

¹Cook County Stroger Hospital, Chicago, IL

Learning Objectives: • Discuss diagnostic error in the diagnosis of subarachnoid hemorrhage (SAH) • Describe the pathophysiology of SAH

Case: 67-year-old male with history of hypertension presented 5 hours after onset of acute occipital and neck pain after trying to cough up some phlegm. He described a sudden pulling sensation and pain to his posterior neck. VS: 36.9C, HR 62, BP 178/89, RR 18, SaO2 98%. He had no neurologic deficits on exam. He was given acetaminophen and his home blood pressure medication and discharged home. He returned 24 hours later to the emergency department for worsening pain. BP 190/89 and on neurological exam was unable to tandem gait. Computed tomography (CT) head demonstrated diffuse SAH without ventriculomegaly, and CTA brain demonstrates an ACoA-A1-A2 junction aneurysm as the likely etiology of the hemorrhage. Patient was started on a nicardipine drip with goal systolic blood pressure of 140 and admitted to the neurosurgical intensive care unit. Patient underwent a cerebral angiogram with endovascular embolization of aneurysm the next day.

Discussion: SAH affects nearly 30 000 individuals annually in North America and results in serious impairment or death in 40% to 60% of cases. The pathophysiology of SAH includes rupture a vessel in the brain, resulting in blood rapidly entering the subarachnoid space. There is a subsequent increase in intracranial pressure with symptoms of an intense headache. Migraine or tension headache was the most common incorrect diagnosis, and failure to obtain a CT scan was the most common diagnostic error. The misdiagnosed SAH represents one of the largest sources of emergency department litigation claims and malpractice settlement payments in the United States. Outcome is highly dependent on early diagnosis and aggressive intervention. SAH can be devastating, yet its initial presentation may be limited to common symptoms and subtle signs, potentially leading to misdiagnosis.

When The Case Goes Cold: Working Up Hemolytic Anemia

B. Bernardin ¹, A. Viqar¹, A. Linn¹, P. Shah¹

¹University of Pennsylvania, Philadelphia, PA

Learning Objectives: • Understand clinical differentiators when assessing autoimmune hemolytic anemias. • Recognize the importance of clinical suspicion in guiding treatment in the face of non-confirmatory labs.

Case Information: A 39-year-old male presented with fever, black urine, and abdominal pain. His exam was notable for tachycardia, jaundice, and livedo reticularis on the knee. Workup revealed hemoglobinuria, elevated LDH, and undetectable haptoglobin, consistent with intravascular hemolysis. Hemoglobin on admission was 8.8 g/dL but required multiple transfusions to remain > 7 g/dL. Direct Coombs test was positive for complement and negative for IgG, consistent with cold agglutinin disease (CAD). Evaluation for etiologies of hemolytic anemia was positive only for a non-SARS-CoV-2 coronavirus. Flow cytometry for paroxysmal nocturnal hemoglobinuria (PNH), Donath-Landsteiner (DL) antibody for paroxysmal cold hemoglobinuria (PCH), and cold agglutinin titers were all negative. However, the blood bank noted that blood samples clotted and “behaved” like cold agglutinin. With a convincing clinical picture for CAD and negative labs for other causes of hemolytic anemia, he was ultimately diagnosed with titer-negative CAD. Cold avoidance and empiric prednisone followed by Rituximab improved the patient’s symptoms and labs. On discharge, he was no longer transfusion dependent.

Discussion: This case highlighted important differentiators among PCH, PNH, and CAD as unique causes of hemolytic anemia. Both PCH and CAD are exacerbated by cold and are Coombs-positive for complement. However, PCH antibodies fix complement which leads to intravascular hemolysis, while CAD involves agglutination of RBCs, is associated with extravascular hemolysis, and often lacks hemoglobinuria. While this patient did have signs of intravascular hemolysis, the DL antibody test was negative; ruling out PCH was important, as these patients are not treated with glucocorticoids as other etiologies are. PNH also demonstrates signs of intravascular hemolysis but is Coombs-negative, flow cytometry positive for PNH cells, and not exacerbated by cold.

The Elusive Diagnosis of Pulmonary Aspergillosis

C. Brickson ¹, R. Murphy¹, W. Turbyfill²

¹University of Colorado, Denver, CO

²University of Colorado CU SOM, Dept of Vet Affairs, Aurora, CO

Learning Objectives: 1. Pulmonary Aspergillosis is a spectrum of disease from Aspergillus nodule to invasive aspergillosis 2. Available testing lacks sensitivity. BAL galactomannan can aid in diagnosis. 3. Patients with chronic respiratory disease have higher incidence of aspergillus colonization, which can be difficult to differentiate from clinical disease 4. A combination of testing, clinical suspicion and monitoring response to antifungal therapy should be used in approaching patients with presumed pulmonary aspergillosis

Case Information: A 66 year-old man with bullous emphysema and relapsing polychondritis on chronic low-dose glucocorticoids presented with two months of cough, fevers, and weight loss after failing to improve with treatment for community-acquired pneumonia. A CT chest revealed a left upper lung cavitary lesion and left-sided pleural effusion. After a negative infectious work up, he was discharged on a prolonged course of antibiotics to treat presumed aspiration pneumonia. A BAL culture later grew Aspergillus fumigatus, which was thought to be non-pathogenic in the setting of his chronic respiratory disease. He was readmitted one month later for worsening hypoxia. A repeat CT chest demonstrated progression of the left-sided cavitary lesion and pleural effusion. The patient underwent a second bronchoscopy that yielded BAL and biopsy cultures positive for Aspergillus fumigatus and galactomannan. Ultimately, he improved with voriconazole to treat subacute invasive aspergillosis (SAIA).

Discussion: Subacute invasive aspergillosis lies within the pulmonary aspergillosis spectrum of disease. The definitive diagnosis requires mycologic evidence of infection in a patient with mild immunosuppression or chronic lung disease with subacute symptoms. This diagnosis can prove elusive as biopsy, sputum and BAL cultures, serum aspergillus IgG, and serum galactomannan all lack sensitivity. Plus, aspergillus colonization is common in patients with chronic respiratory disease. BAL galactomannan can improve testing sensitivity. Clinicians should consider repeat testing and monitoring response to antifungal therapy in patients whom the diagnosis is suspected.

An Unfortunate Case of Acute Blast Crisis and the Impact of Leukostasis on Morbidity

N. Ajayi ¹, M. Canizares-Otero¹*

*(Presenting Author)

¹Aventura Hospital and Medical Center, Aventura, FL

Learning Objectives: Identify and treat life-threatening blast crisis

Case Description: 78 year-old female, PMH significant for CAD, HTN, and T2DM presented to the ER with complaints of progressively worsening malaise for the preceding two days. Of note, patient spoke a specific dialect of Mandarin that was unknown to the translators and ancillary history was obtained from family at bedside. The patient had recently been seen in a homeopathic clinic wherein she was prescribed an unspecified medication. Although this medication temporarily improved her malaise, she was later noted to have difficulty with ambulation, excessive weakness, diaphoresis and dyspnea. On arrival to the ER, the patient underwent V/Q scan, which was found to be low probability for PE. The patient then underwent Duplex ultrasound of the bilateral lower extremities, with the leading diagnosis at that time being VTE. While in the ultrasound suite, the patient went into PEA arrest. Despite transfer to the ICU and further aggressive resuscitation measures, the patient experienced additional instances of cardiac arrest and eventually expired later that evening.

Discussion: Blast crisis and the associated hyperleukocytosis/leukostasis syndromes denote a medical emergency and require both rapid assessment, accurate diagnosis and immediate initiation of treatment. Any delays to commencement of therapy are directly related to increased mortality in an already highly fatal condition. In the case presented, the pathological elevation in leukocytes (above 200) was overlooked, and the differential diagnosis of VTE was pursued extensively despite negative V/Q scan results. Although the prognosis is poor for advanced chronic myelogenous leukemia, the diagnostic error which this patient suffered ultimately lead to her demise within hours of arrival to the hospital. Furthermore, it is important to identify the health disparities associated with patients from minority cultures and the associated language barrier which can further complicate medical care.

Median Arcuate Ligament Syndrome: A Forgotten Cause of Abdominal Pain - A Case Report

H. Chopra ¹, M. Bain¹, M. Schury¹

¹McLaren Oakland Hospital, Pontiac, MI

Median Arcuate Ligament (MAL) Syndrome is a rare diagnosis that is often overlooked by physicians and ranks low on the differential diagnosis for abdominal pain in most cases. Many times, it is a diagnosis of exclusion, often reached months later after initial presentation. It should not be confused with median arcuate ligament compression, which is typically an incidental finding without any symptoms. Symptomatic patients with MAL present with vague gastrointestinal symptoms like abdominal pain, nausea, vomiting, bloating, and weight loss. We present a patient who had similar symptoms in absence of weight loss, one of the classic symptoms of Median Arcuate Ligament Syndrome.

Epidural Abscess in a Patient with Alcohol use Disorder: A Delayed Diagnosis

H. Farajpour Bakhtiari ¹, C. King¹, M. Patel¹

¹University of Colorado, Aurora, CO

Learning Objectives: Describe an atypical presentation of epidural abscesses. Recognizing common biases encountered, particularly when treating patients with alcohol use disorder.

Case Information: A 59-year-old female with a history of treated hepatitis C and alcohol use disorder was admitted after an unwitnessed fall with acute encephalopathy, oriented to self only. Initial workup remarkable for leukocytosis to 13.7, platelets: 27, anion gap metabolic acidosis, AST: 61, total bilirubin: 1.5, blood alcohol: 72, normal TSH, and urine toxicology with opiates and ethanol. CT brain and CT cervical spine revealed no acute abnormalities. She was treated for encephalopathy secondary to presumed alcohol withdrawal and hepatic encephalopathy. On hospital day 1, the patient became febrile and tachycardic with worsening encephalopathy and higher alcohol withdrawal assessment scores. Treatment of withdrawal was escalated. Physical examination, although limited due to encephalopathy, did not reveal any localizing signs of infection. Infectious workup including urinalysis and chest x-ray was unremarkable. Lumbar puncture revealed WBC: 74 (neutrophil dominant), RBC:1775, glucose: 51 and protein: 177 in the cerebrospinal fluid. She was started on treatment for presumed bacterial and viral meningitis. Despite adequate antimicrobials, she had worsening mental status and was transferred to the Intensive Care Unit. There, she was noted to have weakness in her bilateral lower extremities. MRI showed a lumbar epidural abscess from L4 to S1 with cord compression. She underwent an excisional debridement and L5-S1 laminectomy. Her antibiotic regimen was eventually narrowed to nafcillin.

Discussion: This case demonstrated an atypical presentation of an epidural abscess with altered mental status and cerebrospinal fluid pleocytosis. It also displayed many cognitive biases which lead to a delayed diagnosis, including anchoring bias, premature closure, bandwagon effect, and attribution error. Inability to recognize these biases when encountering patients with substance or alcohol use disorder can lead to an inaccurate diagnosis or unnecessary treatment.

Diagnosing an Esophageal Mass; A Closer Look Required

A. Johnson ¹

¹University of Colorado, Denver, CO

Learning Objectives: While uncommon, candida infections can present with localized masses that may resemble malignancies. Esophageal candidiasis is definitively diagnosed by histology.

Case Information: An 87-year-old male with a past medical history that includes polymyalgia rheumatica on chronic steroids, hiatal hernia with GERD presented with a two-day history of dysphagia to solids and liquids. An EGD demonstrated a large, ulcerative friable mass at the gastroesophageal junction that was highly concerning for malignancy. A CT scan was recommended for staging purposes which showed no evidence of metastatic disease but did demonstrate an incidental side-branch intraductal papillary mucin-producing neoplasm (IPMN). Fortunately, the esophageal mass was ultimately found to stain positive for hyphal forms with no signs of dysplasia, and he was treated for candida esophagitis with resolution of symptoms.

Discussion: This is a case of an atypical presentation of candida esophagitis leading to an unnecessary imaging study. Candida species are the most common cause of infectious esophagitis, comprising up to 88% of cases, and typical symptoms include dysphagia, odynophagia, and sternal pain. Diagnosis is made via upper endoscopy that classically demonstrates white plaques; biopsy showing organisms and mucosal inflammation confirms the diagnosis. In rare cases, candida has been reported to present as a localized infectious mass that resembles malignancy, as is the case in our patient. Our patient’s presenting symptoms were not atypical, and he did have risk factors for candida esophagitis including esophageal dysmotility, use of acid suppressive medications, and chronic steroid use. However, the appearance of his lesion led to an incorrect preliminary diagnosis, and in the interest of expediting work up and treatment plan, an imaging study was ordered prematurely. As a result, the patient was found to have an IPMN of low malignant potential that will now be serially monitored.

Incidental ST Segment Elevation: Were We too Sensitive?

M. Indaram ¹, D. Feihel¹, K. Killian¹*

*(Presenting Author)

¹Lenox Hill Hospital, New York City, NY

Learning Objectives: 1. Estimate pre-test probability of myocardial infarction (MI) in an asymptomatic. 2. Use Baye’s theorem to determine likelihood of MI in case of incidental ST elevation. 3. Discuss how fear of embarrassment affects team decision-making.

Case Information: A 65-year-old African American male with hyperlipidemia and hypertension presented to primary care clinic for pre-employment physical and was noted to have an irregular heart rate. ECG showed premature ventricular complexes, subtle convex precordial ST-segment elevations and T-wave inversions. No prior ECGs were available. The patient had no chest pain or other symptoms. Repeat ECG showed progression of ST elevation in the precordial leads to 0.1mV in V1, V2, V4 and 0.2mV in V3 [figure], meeting ischemic criteria according to the third universal definition of myocardial infarction. The patient received aspirin and was transported to the emergency department (ED) where cardiac enzymes were normal and coronary computerized tomography demonstrated non-obstructive coronary disease.

Discussion: Despite a good outcome, high-risk ECG changes in this asymptomatic patient caused the medical team to hesitate, resulting in a 60-minute delay between initial ECG and ED transfer. The team reported that fear of embarrassment from overdiagnosis contributed to the delay. A Bayesian approach could have potentially reduced this social barrier. The pre-test probability of MI can be estimated using the patient’s 9% 10-year atherosclerotic cardiovascular disease risk (tools.acc.org) or the prevalence of silent MI: 14% in a cardiac MRI study of the ICELAND MI cohort. Published likelihood ratios for new ST-segment elevation for MI are high, ranging 5.7-53.9. Applying Baye’s theorem, presumed new ST elevations increase MI likelihood in this patient to 48-90%. Further study of fear of embarrassment contributing to diagnostic error or delay is needed. Bayesian reasoning during group decision-making could reduce fear of embarrassment.

Bedside Ultrasound Diagnostic Utility in the Setting of Undifferentiated Shock: A Case Study

S. Kolke ¹, J. Rutherford¹

¹Cleveland Clinic, Cleveland, OH

Learning Objectives: To demonstrate the utility of point of care ultrasound (POCUS) to narrow the diagnosis in the setting of undifferentiated shock.

Case Information: An adult African American female with chronic kidney disease, hypertension, hyperlipidemia and diabetes mellitus presented to the emergency department (ED) with altered mental status and hyperglycemia. Initial evaluation in the ED demonstrated an anion gap metabolic acidosis as well as a lactic acidosis, and there was concern for both diabetic ketoacidosis and severe sepsis due to pneumonia. The patient was subsequently fluid resuscitated, antibiotics were started, and an insulin infusion was initiated. Upon arrival to the intensive care unit, the patient developed respiratory decompensation requiring non-invasive positive pressure ventilation. POCUS demonstrated hyperdynamic left ventricular systolic function but right ventricular enlargement with intraventricular septal bowing concerning for a pulmonary embolism. Further diagnostics confirmed pulmonary embolism as the etiology for the undifferentiated shock. The patient was subsequently transferred to a tertiary facility for pulmonary embolism response team (PERT) evaluation and management.

Discussion: In the setting of an elevated lactate with potential contributory factors it is easy to commit to the sepsis pathway when managing undifferentiated shock. However, without clearly defined data points suggesting septic shock, the treatment for septic shock can potentially be detrimental towards other forms of shock. The use of POCUS in this case study demonstrated its utility in correcting the diagnostic error of septic shock and instead correctly treating obstructive shock. Thus, ultimately leading to the patient receiving the necessary treatment and transfer to a specialized medical institution.

Unexpected Worry with Increasing Survivorship: A Patient’s Story of MPNST Pseudoprogression Mistaken as Recurrence

A. Lessing ¹, J. N. Lessing¹

¹University of Colorado School of Medicine, Aurora, CO

Learning Objectives: 1) Recall that Malignant Peripheral Nerve Sheath Tumors (MPNST) are aggressive sarcomas and are more common in people with Neurofibromatosis-type1 (NF1). 2) Define pseudoprogression—the radiographic false appearance of cancer progression—and recognize pseudoprogression as a diagnostic possibility as survivorship increases. 3) Appreciate an example of the patient’s experience and perspective through the health care journey.

Case Information: The first author (AL), a then 39-year-old man with Neurofibromatosis-type1 (NF1), presented in 2018 with an asymptomatic right axilla lesion seen on MRI. A decade earlier, he had been diagnosed with a right brachial plexus malignant tritan tumor, a subtype of a highly aggressive sarcoma called malignant peripheral nerve sheath tumor (MPNST). After several rounds of surgery, chemotherapy, radiation, and brachytherapy for the tumor and two recurrences, in 2009 he was considered cancer-free. In 2018 he volunteered for a study whole body MRI incidentally revealing a 4.1x2.0x2.4 cm T2 hyperintense infiltrative enhancing mass. Given the location and his cancer history, his radiologists and oncologists suspected recurrent MPNST. He underwent surgical removal, including resections of latissimus dorsi flap, teres major, and scapula. Pathology showed skeletal muscle with atrophy, fibroadipose tissue, and fat necrosis, but fortunately no malignant cells. The mass was determined to be pseudoprogression, not recurrent or new cancer. He has since remained without evidence of local or distant tumor.

Discussion: Pseudoprogression is the radiographic false appearance of tumor progression in the absence of true tumor progression. Pseudoprogression has been described in soft tissue sarcomas after preoperative radiation and chemotherapy where the lesion enlarges as it undergoes cystic necrosis (PMID:12829150). Pseudoprogression also complicates response assessment after immunotherapy, where some responding lesions initially enlarge from infiltration of immune effector cells prior to subsequent shrinkage (PMID:30231380). Recognition of the possibility of pseudoprogression is necessary to providing best care. Here we describe the first shared case of pseudoprogression of a MPNST. While this may be a unique case, the experience—misdiagnosis due to reliance on technology—is not. As people survive cancers for longer, pseudoprogression is bound to become more common. It is the author’s hope that by sharing his story, doctors will consider pseudoprogression in their differential diagnosis.

“Intention to Treat”: Judicious Antibiotic Use in Asymptomatic Bacteriuria

N. Mistry ¹, J. Lessing¹

¹University of Colorado, Denver, CO

Learning Objectives: 1. Learn indications for treatment of asymptomatic bacteriuria. 2. Identify a less common presentation of Stevens-Johnson-syndrome/toxic epidermal necrosis (SJS/TEN). 3. Recognize 2 common antibiotic side effects and the importance of judicious antibiotic use.

Case Information: A 36-year-old woman hospitalized for alcoholic hepatitis was started on ceftriaxone and levofloxacin for bacteriuria. She denied urinary symptoms. One week later, she developed a painful, pustular, erythematous rash on her limbs with mucosal ulcerations in oropharynx. The patient was started on acyclovir for presumed HSV which did not help. Skin biopsy showed epidermal necrosis suggestive of SJS/TEN. Despite stopping antimicrobials and starting prednisone, her SJS/TEN resulted in uncontrolled pain, scarring, and anasarca from steroids and IV fluids, requiring prolonged hospitalization.

Discussion: Antibiotic overuse or misuse can lead to detrimental side effects. Prevalence of asymptomatic bacteriuria varies with age, gender, and comorbidities but ranges from 2% to 15% in the general population (1). Only in certain cases of asymptomatic bacteriuria is treatment recommended, such as pregnancy, post urologic procedures, and renal transplant. Unfortunately, antibiotic misuse in this patient severe adverse effects. Approximately 20% of patients experience adverse effects from antimicrobial use, such as C difficile infection and end-organ damage (2). Clinical decision tools and a local antibiogram can help decrease antibiotic misuse and prevent adverse effects. Given the multitude of cutaneous drug reaction manifestations (3), diagnosis and treatment can be delayed if other diseases are suspected. Our patient presented with a pustular rash instead of the typical bullous lesions seen with TEN and was started on additional antimicrobials, delaying steroid initiation for supportive treatment of TEN. We have all been tempted to treat an abnormal urinalysis. This unfortunate case is a reminder to be intentional about and judicious with antibiotic prescribing for all patients, and in particular for those with asymptomatic bacteriuria.

Delayed Diagnosis of Spondyloarthropathy in the Primary Care Setting

S. Pahlavan ¹, A. Wethington¹, M. Melton¹, J. Stichman²

¹University of Colorado, Aurora, CO

²Denver Health Medical Center, Denver, CO

Learning Objectives: Recognize increased risk for delayed diagnosis of spondyloarthropathy in women. Demonstrate limitations of serologic and radiographic imaging in diagnosis of sacroiliitis. Evaluate the impact of cognitive biases on the workup of back pain.

Case Information: A 34 year-old woman was seen in primary care clinic with a complaint of two weeks of photophobia and blurry vision. She has a history of chronic low back pain, depression, obesity, gastric ulcers, STIs, and interpersonal violence. Medications include cyclobenzaprine, tramadol, topical diclofenac, topical lidocaine and omeprazole. Chart review revealed at least two prior similar ocular episodes, both diagnosed as acute anterior uveitis, raising concern for recurrence. Back pain was diagnosed at various points over the 5 years prior as piriformis syndrome, acute lumbar strain and sciatica. MRI of the lumbar spine 3 years prior revealed a “small protrusion at L4-L5 [that] contacts the exiting nerve without compression” and “marrow edema across the sacroiliac joints is abnormal but incompletely evaluated”. HLA-B27 testing at that time was negative, and she was referred to neurosurgery. Subsequent ED and neurosurgery visits addressed only radiculopathy. She continued to receive medication refills without re-evaluation of her back pain during the pandemic. Sacroiliac radiographs obtained during our visit revealed bilateral sacroiliitis.

Discussion: We have presented the case of a young woman with a delayed diagnosis of ankylosing spondylitis (AS). Historically, male-female prevalence ratios estimated the prevalence of AS at 10:1. Though updated studies have found this to be closer to 3:1, diagnosis is still delayed in women compared to men. In this case, implicit bias with respect to obesity and nonadherence, as well as anchoring bias, delayed advanced imaging and diagnosis. The sacroiliac abnormality was recognized on MRI, but HLA-B27 testing was inappropriately ordered rather than pelvic MRI to further evaluate the sacroiliac joints.

Secondary Siadh Due to Physician/Patient Communication Error

S. Reddy ¹, S. Menakuru²

¹Gandhi Medical College

²Indiana University Health

Case Information: A 19-year-old woman came to the hospital complaining of left flank pain and was found to have a small renal calculus. The woman was instructed by her attending physician to “drink a lot of water” and that it would relieve her symptoms. She drank large quantities of water in only a few hours and was taken to a local medical clinic by her boyfriend after “she started to vomit and shake vigorously”. Later, she was brought to the hospital with symptoms of headache, seizures, confusion, vomiting, and weakness which progressed to a coma. She was found to have severe hyponatremia (sodium level of 123 mEq/L) and further laboratory testing showed hypertonic urine (320 mosm/kg). This led to the conclusion that secondary SIADH had developed in this patient, confirmed by the detection of elevated plasma ADH concentration.

Learning Objectives: This case is an example of why instructions to a patient must be clear and concise. Patients may have their own preconceived notions about what “a lot of water” means, as in a case of hyponatremia in a patient who believed drinking water would resolve her UTI (1). This case also raises the issue that doctors need to be more approachable in allowing patients to ask questions. If the patient had been comfortable enough with her physician to ask how much water she should drink, she would not have ended up in the hospital.

Discussion: In most instances water intoxication is seen due to psychotic illness. The physician/patient communication error seen here led to a unique case of water intoxication without psychiatric symptoms. Most other cases of water intoxication without psychiatric symptoms in literature only cause hyponatremia, secondary SIADH is present in only one of these cases (2). This case highlights the importance of understandable patient instructions as well as an approachable physician demeanor.

References 1. Lee LC, Noronha M When plenty is too much: water intoxication in a patient with a simple urinary tract infectionCase Reports 2016;2016:bcr2016216882. 2. Yamashiro M, Hasegawa H, Matsuda A, et al. A case of water intoxication with prolonged hyponatremia caused by excessive water drinking and secondary SIADH. Case Rep Nephrol Urol. 2013;3(2):147-152. Published 2013 Dec 21. doi:10.1159/000357667

Acromegaly - A Diagnosis Hiding in Plain Sight

D. Sato ¹, J. Branch¹, S. Nishiguchi¹, E. Tanaka¹

¹Shonan Kamakura General Hospital, Kamakura City, Japan

Learning Objectives: The diagnostic error due to confirmation bias could have been prevented by not neglecting a detailed medical history and careful physical examination.

Case Information: A 73-year-old female nursing home resident had been repeatedly admitted to and discharged from our hospital due to frequent urinary tract infections (UTIs). She was paralyzed from a cerebral hemorrhage 17 years ago. She had developed a fever, chills and back discomfort prior to admission (PTA) and was diagnosed with pyelonephritis. She improved well with antibiotic therapy. However, on examination she had some characteristic physical features including rhinomegaly, hoarseness and chiromegaly. Further history provided by her family revealed a diagnosis of a pituitary tumor four years PTA, but without further follow-up. Radiographs showed heel pad hypertrophy and enlarged skull frontal sinuses, respectively. A 75-gram oral glucose tolerance test failed to suppress growth hormone levels. Insulin-like growth factor and prolactin levels were also elevated. A pituitary macroadenoma (17 mm x 24 mm) was identified by head magnetic resonance imaging, leading to a diagnosis of acromegaly.

Discussion: In this case, because the patient was paralyzed, she may have been unaware of the change in her physical features. With repeated hospitalizations, all medical staff knew the patient well, including her physical characteristics, but no one questioned her odd appearance. There was anchoring bias due to the patient’s repeated admissions for UTIs, and few complications. There was no relation to the chief complaint, and so her acute presentation was not be directly linked to the endocrinological diagnosis. This case highlights that every admission should be considered a new admission, rather than a continuation of the prior one. A thorough history and physical examination may unveil a diagnosis that is otherwise hiding in plain sight.

A Snapshot in Time: The Challenge to Meet Diagnostic Criteria in a Patient with Hemophagocytic Lymphohistiocytosis

D. Scudder ¹, A. Friefeld¹, M. Weinberg¹, J. Lessing¹

¹University of Colorado School of Medicine, Aurora, CO

Learning Objectives: 1. Define and challenge diagnostic criteria in rare conditions. 2. Recognize that malignancy can be an underlying cause of HLH, even in an adult with rheumatologic disease, and that treatment varies with etiology.

Case Information: A 72-year-old woman with well-controlled rheumatoid arthritis (RA) presented with 4 months of fevers, night sweats, and malaise. Work-up revealed hepatosplenomegaly, pancytopenia, and liver injury. Bone marrow and liver biopsies were unrevealing. She was transferred to a tertiary care center. Admission labs showed pancytopenia, elevated ferritin, liver failure and—in contrast to prior normal studies—elevated triglycerides and decreased fibrinogen. After now meeting diagnostic criteria for hemophagocytic lymphohistiocytosis (HLH), she was started on steroids. Outside pathology was re-reviewed and showed hemophagocytosis. Soluble IL-2 and CXCL9 returned elevated. Anakinra then etoposide were initiated. Unfortunately, her disease progressed and she transitioned to comfort-care. Underlying HLH cause was eventually determined on repeat bone marrow biopsy to be T-cell lymphoma.

Discussion: HLH is a hyper-inflammatory disease that arises from genetic immune system defects or, in adults, more commonly triggered as a complication of malignancy, infection, or rheumatologic disease. Like many complex conditions, diagnosis is defined by meeting set criteria primarily created to have consistent patient populations for clinical study. This presents a real-life dilemma, as early treatment is essential but three HLH diagnostic criteria have long send-out times and a fourth requires biopsy. By the time our patient met criteria, her status had already deteriorated. It is important to recognize that any evaluation represents a single moment—a snapshot in time—and that conditions are liable to change. Additionally, the diagnosis was missed by “gold-standard” pathology the first time around. If HLH was recognized originally, she may have received treatment earlier. Treatment was also delayed by initial suspicion due to her history of RA for macrophage activation syndrome (MAS), which is treated with anakinra rather than etoposide; however, MAS is more commonly associated with juvenile idiopathic arthritis. This case exemplifies the challenge of reliance on diagnostic criteria; the value of a second opinion on subjective results; and the importance of determining the underlying cause to guide therapy.

Carotid Rupture in a Transplant Patient with CMV

M. Strom ¹, K. Sanford¹

¹University of Colorado, Aurora, CO

Introduction/Objectives: This case will discuss how immunosuppressed patient are at risk for infection and how infection can be a trigger for carotid artery rupture. 1) Discuss CMV infection in a patient that is immunosuppressed 2) Discuss signs and symptoms of carotid rupture 3) Review risk factors for carotid rupture including infection.

Case Description: 37-year-old male immunosuppressed after kidney and pancreas transplant from diabetes mellitus and end stage renal disease, was hospitalized for candida esophagitis, cytomegalovirus (CMV) viremia, BK viruria, and C-diff. Work-up was negative for EBV, HSV, and ZVZ. He was started on Ganciclovir, Vancomycin and Fluconazole. His Azathioprine was held, and the tacrolimus dose was decreased. During his hospital course, he developed left neck pain and swelling. Exam revealed swollen left anterior neck with mild tracheal deviation to the right, neck tender to palpation and firm, right bruit present and hypertensive to 190/90. Stat CT neck revealed arterial extravasation from the left common carotid artery with left neck hematoma that extended into retropharyngeal space and upper mediastinum. Patient was emergently taken for a left common carotid interposition graft.

Discussion: Immunosuppressed patients are at higher risk for opportunistic infections. The carotid rupture was an unexpected complication and although rare, can be caused by CMV infection. Etiology of carotid ruptures include atherosclerosis, connective tissue disorders, trauma, inflammatory disorders, and primary infection. 1 Only 100 case reports of carotid rupture are secondary to primary infection. 1 The thought is that inflammatory changes weaken the arterial wall and cause dilation. 1 The carotid artery pathology was negative for CMV infection. Etiology is unclear, but still thought to be secondary to infection.

Beware of the Thief! A Delayed Diagnosis of Subclavian Steal Syndrome

M. Yokose ¹, E. Maeno², T. Adachi³, H. Sugimura², T. Shimizu¹

¹Dokkyo Medical University, Shimotsuga, Japan

²Nikko Medical Center

³Tochigi Medical Center

Learning Objectives: Understand the risk of availability and anchoring biases in seemingly straightforward scenarios.

Case Information: A 73-year-old man with end-stage kidney disease on hemodialysis was hospitalized for syncope right after defecation. Just before this episode, he had a hemodialysis session. His dry weight had been gradually adjusted upward recently due to hypotension during hemodialysis and post-dialysis fatigue. He had type 2 diabetes mellitus, hypertension, dyslipidemia, asymptomatic lower extremity peripheral artery disease, and a history of lacunar stroke. He underwent left upper extremity arteriovenous fistula ligation for dialysis access steal syndrome and carotid endarterectomy for left internal carotid artery stenosis. On examination, his blood pressure in the left arm was 88/53 mmHg. Cardiopulmonary and neurological examinations were normal. There was normal thrill and murmur on the arteriovenous fistula in his right forearm. Electrocardiogram, head non-contrast computed tomography, brain magnetic resonance imaging, and magnetic resonance angiography results were normal. Under the diagnosis with situational syncope on top of hypovolemia, his dry weight was further increased. He was discharged several weeks later; however, he had recurrent syncopal episodes. Additional contrast-enhanced computed tomography of the neck and chest revealed severe stenosis of the prevertebral left subclavian artery, and duplex ultrasound showed retrograde flow in the left vertebral artery. He was diagnosed with subclavian steal syndrome. Scheduled endovascular stenting was performed and his symptoms eventually resolved.

Discussion: This case highlights a diagnostic delay due to availability and anchoring bias. Just like syncope after defecation seen in this case, seemingly typical scenarios could easily make our thinking inflexible. Clinicians should recognize the vulnerability of intuitive diagnostic thinking under such circumstances and evaluate the likelihood of diseases other than the most convincing one consciously. In other words, this “contemplate complement strategy” of considering the complement of an intuitive diagnosis will prevent diagnostic anchoring and improve diagnosis.

Relation: A Software Tool for Exploring the Relation between Diagnostic Accuracy and Measurement Uncertainty

T. Chatzimichail ¹, A. Hatjimihail¹

¹Hellenic Complex Systems Laboratory (HCSL)

Purpose/Problem: Although diagnostic accuracy is fundamental to clinical decision making and measurement uncertainty is critical to quality and risk management in laboratory medicine, there has been very limited research on their relation.

Description of Program: For this reason, the freely available interactive program “Relation” was developed for calculating, optimizing, plotting and comparing various diagnostic accuracy measures and the corresponding risk of diagnostic or screening tests measuring a normally distributed measurand, applied at a single point in time in non-diseased and diseased populations.

Outcomes: The program demonstrates the relation between the diagnostic accuracy measures and the measurement uncertainty for a screening or diagnostic test measuring a single measurand, for differing prevalence of the disease, mean and standard deviation of the measurand, diagnostic threshold, standard measurement uncertainty of the tests and expected loss. Furthermore, it provides calculators for the calculation of the effects of measurement uncertainty on the diagnostic accuracy measures and corresponding risk and for calculating the diagnostic threshold optimizing objective and loss functions.

Discussion: The program “Relation” provides 269 different types of plots of diagnostic accuracy measures, many of which are novel. To the best of our knowledge, this is the only program to provide this range of plots without advanced statistical programming.

Significance of Findings: The program “Relation” is user-friendly and can be used as an educational and research tool in medical decision-making, to explore the relation between measurement uncertainty and diagnostic accuracy.

Longitudinal Diagnostic Reasoning Curriculum (DoctoR) for Undergraduate Medical Students

H. Hofmann ¹, A. Wray²

¹Cleveland Clinic Foundation, Cleveland, OH

²University of California Irvine, Irvine, CA

Purpose/Problem: Teaching diagnostic reasoning (DR) in medical training is important. Approaches are often passive and learner-, teacher-, and environment-dependent. We developed and implemented a longitudinal, four-year DR curriculum (“DoctoR”) for medical students to standardize teaching at our institution.

Description of Program, Assessment or Study: First-year students have a video podcast on DR and dual-process theory (Y1A). Second-year students review and practice DR theory. A flipped-classroom video podcast presents frameworks for differential diagnosis and assessment (Y2A). Its case-based large group session with small group breakouts uses an evidence-based worksheet (Y2A). Each standardized patient case incorporates a DR objective and the worksheet (Y2B). Students also use the worksheet in clinic (Y2C). Students present a real case for the Clinicopathological Conference (Y2D). During the internal medicine clerkship, third-year students participate in a case-based session (Y3A). It uses structured reflection and serial cue methods. Students mid-year have a large-group “Diagnostic Error” didactic (Y3B). Fourth-year students have a session to recognize the importance of patient communication, including its role in avoiding diagnostic error (Y4A).

Outcomes: Students consistently reported faculty as committed to teaching (Likert scale, Y1A mean 4.09/5, SD 0.79, n=45; Y2A mean 3.84/4, SD 0.37, n=51; Y3B mean 4/5, n=34), and explained content at proper level of complexity (Y1A mean 4.07/5, SD 0.78; Y2A 3.82/4, SD 0.39; Y3B mean 3.9/5, n=34). One student noted for Y1A, “The podcast was great. To the point, and insightful.” A student in Y2A said, “Engaging session with appropriately sized break-out rooms. I felt the small group interactions kept us accountable and active.” Students on average found Y4A valuable (mean 4.6/5, SD 0.71, n=17) and rated the session 4.6/5 overall (SD 0.63). Students particularly enjoyed its small group format, session timing post-clinical rotations, and the self-reflection exercise.

Discussion: We successfully developed and implemented a longitudinal DR curriculum for medical students. We integrated content into established courses. Learners report sessions are engaging and at the appropriate level. The format promotes student engagement. A limitation is that evaluations are only based on student perspective. Future evaluation aims to assess knowledge retention and application of DR principles. Next steps for DoctoR include expanding to other clinical rotations. The curriculum can be readily adapted to other institutions.

Significance of Findings: A longitudinal DR curriculum for undergraduate medical training was established, well-received by learners, and presents the opportunity for expansion both internally and externally.

A Multidimensional Approach to Assessing Diagnostic Reasoning Skills in the Medicine Clerkship

A. Jagannath ¹, M. Sebasky², B. Kwan¹

¹VA San Diego/UCSD, San Diego, CA

²University of California San Diego, San Diego, CA

Purpose/Problem: Diagnosis education for third-year medical students (MS3) is variable. Many existing curricula teach concepts of diagnostic reasoning through case-based didactics, but assessment is absent or limited. Additionally, the integration of medical knowledge from other health professionals relevant to diagnosis is lacking. We created a longitudinal curriculum in the internal medicine clerkship that teaches the mechanics of diagnosis, emphasizes the role of the interprofessional healthcare team in diagnosis, offers faculty-guided practice, and provides formative and summative assessment via direct observation.

Description of Program, Assessment, or Study: Diagnosis education in the Internal Medicine Clerkship at the University of California, San Diego (UCSD) School of Medicine was formalized for the 2021-2022 academic year in collaboration with the UCSD Master Clinician Program (MCP), a distinguished group of clinician-educators who occupy a non-evaluative coaching role. In orientation, students are introduced to the concepts of problem representation, diagnostic schema, and illness scripts in an interactive case-based workshop and provided resources for ongoing practice. In a subsequent session, students solve a clinical case with two MCP faculty members and are taught how cognitive biases can lead to diagnostic error during the discussion. The role of the interprofessional team in diagnosis is explored through a formal introduction to its members (e.g., RNs, speech pathologists, physical therapists), discussing strategies to promote proactive and effective collaboration. Formative assessment of students’ diagnostic reasoning occurs weekly through direct observation of patient care by MCP faculty on ward rounds followed by same-day 1:1 feedback using the Assessment of Reasoning Tool. The Interdisciplinary Education Perception Scale (IEPS) and Readiness for Interprofessional Learning Scale (RIPLS) questionnaires are used at the beginning and end of the clerkship to assess student readiness for interprofessional collaboration. Finally, students undergo summative assessment by participating in a standardized patient encounter (SPE) wherein they generate a problem representation, diagnostic schema, and hypothesis-driven differential diagnosis, and defend their leading diagnostic consideration.

Outcomes: We aim to provide students with a comprehensive assessment of their diagnostic skills through feedback guided by the ART, scores on the IEPS and RIPLS questionnaires, and SPE performance.

Discussion: Diagnosis curricula for MS3s fall short in interprofessional education and assessment. Our curriculum aims to provide a longitudinal experience to grow and sustain their skillset. Collaboration with the MCP and use of SPEs are unique strengths of this curriculum but may pose challenges to generalizability as they depend on faculty availability and access to simulation resources.

Development, Implementation, and Evaluation of a Critical Thinking Curriculum Using Malpractice Claims Database Cases

C. Royce ¹, A. Sullivan¹, K. Huth², A. Simpkin³, J. Rhatigan⁴, R. Schwartzstein¹

¹Beth Israel Deaconess Medical Center, Boston, MA

²Boston Children’s Hospital, Boston, MA

³Massachusetts General Hospital, Boston, MA

⁴Brigham and Women’s Hospital, Boston, MA

Purpose: How medical students acquire critical thinking skills remains debated. Many writers argue these are emergent skills resulting from acquisition of knowledge and experience through reflective practice. Others maintain that principles of critical thinking are a discrete set of skills that are learnable by pre-clinical students despite their lack of extensive knowledge and experience. The authors developed a critical thinking curriculum based on malpractice claims cases for students ending their pre-clinical training and evaluated students’ ability to retain the information and their assessments of its utility.

Methods: The authors used published summaries of litigated malpractice claims where diagnostic and cognitive errors were coded as a primary root cause. They created a case-based curriculum focusing on strategies to reduce cognitive errors, consisting of 3 hours didactic teaching and 5 hours case discussion administered to pre-clerkship medical students at a single medical school as part of a transition to clinical medicine course. Students completed post-tests and satisfaction surveys.

Outcomes: Of 135 students enrolled, 92 provided complete or partial responses. Forty-eight to sixty-two percent of students reported learning “a moderate amount” in each of six sessions, while 33%-46% reported “learning a lot.” Seventy-two percent felt it was “the right amount” of training, 45% reported the training was “good” and 33% reported it was “very good” or “excellent.” Prior to training, self-reported critical thinking skills were “fair” (58%) or “good” (35%), and only 1% “very good”; after training, a majority reported skills were “good” (62%) or “very good” (24%). Students scored an average of 18.5 out of a possible 24 points on the multiple-choice knowledge post-test. Half of students answered open-ended questions. Discussion of real cases, applying diagnostic reasoning and understanding consequences of cognitive errors were the most common response to “most valuable aspect of curriculum”. Other themes included appreciation of metacognitive practices and learning about bias mitigation strategies.

Discussion: Pre-clinical students found a critical thinking curriculum based on malpractice claims cases to be a valuable part of their education and demonstrated retention of key principles. Students valued the use of real-world cases and the discussion of clinical impact of cognitive errors.

Significance of Findings: This study contributes to the literature surrounding acquisition of clinical reasoning skills, demonstrating a short exposure to critical thinking concepts including metacognitive practice and bias mitigation skills improved self-reported critical thinking skills.

An Interprofessional, Case-based Workshop on the Co-production of Diagnosis for Medical and Pharmacy Students

M. Symoniak ¹, V. Lang²

¹St. John Fisher College, Rochester, NY

²University of Rochester, Rochester, NY

Purpose: National competencies in diagnosis recognize the collaborative role of interprofessional teams, not just individuals, in diagnostic reasoning and reducing diagnostic error. The goal of this intervention was to teach trainees in pharmacy and medicine to identify how they can collaborate in the diagnostic process.

Description of Program: Fifty-four third-year pharmacy and 100 second-year medical students participated in a 2-hour on-line workshop. Didactic content included an overview of diagnostic reasoning, diagnostic error, and pharmacists’ scope of practice. In small groups, students worked through a written case collaboratively. Each group included 8-12 medical and pharmacy students facilitated by a medicine or pharmacy faculty member. The case was co-designed by faculty in pharmacy and medicine. It began with an older patient on multiple medications who was recently discharged from the hospital and developed lightheadedness. The case unfolded in two sequential aliquots. Students from each profession were provided with different, but complementary, information with each aliquot. Medical students received information from a primary care physician’s perspective, while pharmacy students received information from a community pharmacist’s perspective. Students needed to communicate with each other in order to arrive at a correct diagnosis. Students completed reflections on how communicating with their interprofessional counterparts impacted their diagnostic reasoning. Comments were coded by two faculty independently, and discrepancies resolved by consensus.

Outcomes: See Table.

Discussion: Nearly all students in both professions found that working through the case together impacted their diagnostic reasoning, but in different ways. Most pharmacy students recognized the additional information from the physician’s perspective and broadened their differential diagnoses. Most medical students connected information from the pharmacist perspective (e.g., mechanism of action or data from medication reconciliation) with diagnostic hypotheses to justify their differential diagnoses. Interprofessional planning of the session and delivery of case information in aliquots specific to each profession led to high engagement of students from both professions.

Significance of Findings: An interprofessional co-production of diagnosis experience can be calibrated to a level where students of two different professions learn how to work together to improve diagnostic reasoning.

Framework for Using Longitudinal Diagnostic Codes to Support a Diagnostic Learning Healthcare System

J. Koola ¹, S. Nemati¹, R. El-Kareh¹*

*(Presenting Author)

¹UC San Diego, San Diego, CA

Statement of Problem: Improving diagnostic performance requires organizations to review diagnostic processes and outcomes, but current case identification methods have significant limitations. We sought to develop computable methods to identify potential diagnostic opportunities.

Description of the Intervention or Program: The Longitudinal Analysis of Codes to Identify Diagnostic Opportunities (LUCID) project takes a set of timestamped diagnostic codes from a population of patients and outputs: 1) a list of cases that may represent diagnostic opportunities (avoidable delays in diagnosis) and 2) visual representations of the diagnostic trajectories for the cases to facilitate efficient case review (Figure). In our pilot, we identified three target conditions with avoidable diagnostic patient harm: acute venous thromboembolism, colorectal cancer and spinal abscess. For patients with the diagnosis-of-interest, we collated encounter data and converted ICD-10-CM (International Classification of Diseases, 10th Revision, Clinical Modification) codes to AHRQ CCS (Clinical Classification Software) categories. We focused on new signs, symptoms, and diagnoses immediately preceding the diagnosis-of-interest as potential signals. This “lookback window,” 180 days for colorectal cancer and 14 days for thromboembolism and spinal abscess, represents a potential time period where earlier diagnosis might have been possible. We create temporal, visual representations of the diagnostic trajectories using CCS categories and highlight the target condition. LUCID will employ three analytic approaches to scrutinize the “lookback window” compared to the patient’s overall health history: 1) simple comparisons between CCS categories; 2) logistic regression; and 3) machine learning algorithms. We then perform structured diagnostic case reviews using validated processes. The insights from these reviews are forwarded to our organizational quality group and inform iterative improvements in our analytic and visualization approaches.

Findings to Date: Roughly 4-20% of patients with a diagnosis-of-interest had a diagnostic code from a new CCS category within their “lookback window.” Our assessment of the learning opportunities from the case reviews is on-going.

Lessons Learned: Lessons include: 1) modern electronic health record systems store diagnostic codes in multiple places—complicating extraction of longitudinal codes; 2) data visualization using CCS categories can focus case reviews; and 3) insights from case reviews can enable rapid improvement of analytic approaches.

How the Predictive Value of a Test Affects the Epidemiology of Pandemics

M. Gusack ¹

¹MANX Enterprises, Ltd., Huntington, WV

Statement of Problem: The recent Covid-19 pandemic generated the fastest response in modern history leading to the worldwide roll out of a variety of tests for the virus. However, the use of early testing modalities to identify this new infectious disease is fraught with complications. In particular, the validity of a test for a new virus is, in part, dependent on the epidemiology of the pandemic. Yet the tests can influence and confound epidemiologic data critical to assuring diagnostic accuracy as well as effectiveness of policy making.

Description of the Intervention or Program: Presented is an overview of the circular interaction between the sensitivity and specificity of new laboratory tests designed to detect a novel infectious organism and the epidemiology that informs these tests and is influenced by them. A series of epidemiologic models were developed and assessed against a variety of different levels of sensitivity and specificity of one or more tests to determine how this interrelationship may lead to diagnostic and policy making error. The models examine a number of variables that may influence the epidemiology of a pandemic including but not limited to: • Transmissibility • Infectivity • Virulence • Period of viral sheading • Effect of personal protective equipment • Social distancing • Quarantine • Medical intervention, and • Vaccination

Findings to Date: The introduction of one or more variables into the model illustrates the potential for adversely affecting epidemiologic data including but not limited to: • For any sensitivity and specificity of a test, prevalence of the infectious organism will determine the number of false positives and false negatives • Early in a pandemic, low prevalence leads to higher false positives and overestimation of the spread of the infection while understating the fatality rate • Late in a pandemic, high prevalence leads to higher false negatives leading to underestimation of the degree of societal immunity while overstating the fatality rate • The use of multiple testing methodologies exhibiting different sensitivity and specificity will further confound the epidemiologic picture

Lessons Learned: Responding to pandemics requires: • Understanding the circular interaction between characteristics of screening test(s) and epidemiology of the spread of the infectious organism • Including the impact of tests on epidemiologic models used to track the pandemic • Devising a means to correct epidemiologic data to provide more accurate measurement of a pandemic to better direct efforts to mitigate its effects on patients and society.

A QI Initiative to Improve Vital Sign Reassessment and Sepsis Recognition in a Community ED

C. Mangus ¹, P. Papineni², M. Schneider², M. Roebuck², M. Jaggi², P. Mahajan¹

¹University of Michigan, Ann Arbor, MI

²Hurley Medical Center, Flint, MI

Problem: Sepsis, a life-threatening systemic response to infection, is a leading cause of pediatric morbidity and mortality. Delayed identification of sepsis is a significant cause of misdiagnosis-related harm. Vital sign abnormalities may be the first and, in many instances the only, sign of early sepsis. Reassessing vital signs at emergency department (ED) discharge for patients with abnormal initial vital signs is an important process measure to reduce the misdiagnosis of pediatric sepsis. However, only 30% of children with abnormal triage vital signs in our community pediatric ED had their vitals repeated prior to ED discharge.

Intervention: We implemented a quality improvement (QI) initiative to improve the percentage of pediatric patients discharged from our ED with documented discharge vital signs from the baseline rate of 30% to >80% to enhance the timely and accurate diagnosis of pediatric sepsis and reduce patient harm. Our intervention was grounded on the Institute for Healthcare Improvement’s Model for Improvement and utilized the Plan-Do-Study-Act (PDSA) cycle to implement and test interventions. Initial steps included meetings with key stakeholders to update the pediatric vital sign parameters in the electronic medical record (EMR). Then, a nursing alert was built into the EMR that prompted the nurse to repeat vital signs prior to ED discharge if a patient had abnormal vital signs during ED triage. Change (or delta) in length of stay and rate of hospitalization were used as balancing measures and assessed the impact of the intervention on ED processes and outcomes.

Findings to Date: Initiation of the EMR alert to repeat vital signs improved discharge vital sign acquisition to 44.5%. A second PDSA cycle with data review and input from key stakeholders revealed that mild abnormalities in blood pressure measurements alone accounted for 41% of the nursing alert triggers. As such blood pressure abnormalities have not been identified as relevant for sepsis detection, our next PDSA iteration will remove abnormal blood pressure measurements as a trigger for the “repeat vital sign” alert.

Lessons Learned: As this QI project continues, we have seen slow, incremental improvement in repeat vital sign assessment. Input from key stakeholders, including bedside nurses and information technology specialists, has been imperative in identifying clinically relevant interventions for detecting sepsis without significantly increasing provider workload.

Implementation and Outcomes of a Peer Consult Service for Pediatric Hospitalists During the COVID-19 Pandemic

L. O’Neill ¹, P. Bhansali¹, M. Rush¹, S. Stokes¹, S. Todd¹, N. Shah¹

¹Children’s National Hospital, Washington, D.C.

Statement of the Problem: The COVID-19 pandemic challenged hospitalists on the frontlines of patient care while simultaneously removing in-person peer support due to increased teleworking. A peer consult program can bridge this gap by providing clinical decision support during a time of increased clinical uncertainty. We describe the development, implementation, and outcomes of a peer consult service for a pediatric hospital medicine (PHM) division.

Description of Program: The peer consult service supports >75 members of the PHM division who practice at a tertiary care children’s hospital, six affiliate community sites, and a sub-acute facility. Eight PHM faculty with diverse experiences and areas of expertise volunteer as consultants, with two assigned per week. PHM faculty on clinical services requested peer consultation for challenging clinical scenarios. The service functioned in a “curbside” manner and consultants did not have medical record access. Survey data were collected from both the requesting and consultant physicians to describe the characteristics of the consults, satisfaction with the process, and changes to patient care.

Findings: Over the 52-week time period from May 2020 - April 2020, 24 consultations were completed for 26 patients, an average of 2 consults per month. We received completed surveys from 20 of 22 (83%) requesting physicians and 22 of 24 (92%) consulting physicians. The service received consults from the tertiary inpatient hospitalist services (n=15), community sites (n=7), and surgical co-management service (n=2). Consultants spent an average of 22 minutes per consult (range 2-75), with more time spent if more than one patient was discussed. The majority of consultations were completed by telephone (n=19). Requesting hospitalists unanimously reported very high satisfaction with the timeliness of response and overall process of the consultation (average 5 on 5-point Likert scale). Hospitalists requested peer consultation for diagnostic (n=11) or management uncertainty (n=7), discharge planning concerns (n=4), managing subspecialty consult recommendations (n=4), and challenging patient psychosocial situations (n=4). Peer consultation changed management in 40% of the consults by prompting formal subspecialty consultation or new diagnostic testing. Requesting hospitalists commented that the consultation provided a sense of reassurance and validation of their plan of care.

Lessons Learned: We established a peer consult service that provides just-in-time clinical decision support across various practice sites. If a more formalized peer consult service is desired, medical record access would be necessary to allow for independent review during consultation. Future steps include studying the impact of the program on diagnostic accuracy and patient outcomes.

A Non-Invasive Tool to Quantify Autonomic Dysfunction, A Prognostic Indicator in COVID-19

A. Rajanna ¹, J. C R¹, S. Rajoli¹

¹BMCRI

Background: The Outbreak of SARS-CoV-2 has caused a major pandemic posing a threat to the millions of lives all over the world. The evidence shows that there is a relation between the autonomic nervous system and coronaviruses and likewise, levels of inflammatory markers - C-reactive protein (CRP) and autonomic dysfunction. Autonomic dysfunction is elicited using heart rate variability which in turn quantified using autonomous regulatory index (ARI). Hence this study was conducted to determine if ARI measured could be used as a non-invasive measure of autonomic dysfunction among COVID-19 subjects.

Materials & Methods: An exploratory study was conducted among randomly selected 22 COVID-19 male patients aged more than 18 years, admitted to COVID ward, Victoria Hospital, Bengaluru for 5 days, using ANS Recorder, a non-invasive heart rate variability recorder heart rhythm data were collected, one test per day continuously for 5 days, an ANS Analysis Software instrument was used to record, analyze and interpret the heart rate variability in terms of ARI and CRP levels were measured. Data was analyzed using SPSS version 18.0. A P value of < 0.05 was considered statistically significant.

Results: The occurrence of autonomic dysfunction in COVID-19 patients using the Patented NEUROCOR Precision HRV® Solution was found to be among 50.0%. The median scores of average ARI indices were significantly lesser among those with higher health risk (28.39) compared to those with lower health risk (65.95) (P<0.05). The Median ARI index showed a weak negative correlation (r = -0.13, P>0.05) with CRP (P>0.05). ARI index showed a significantly excellent predictive ability in detecting the higher health risk with the areas under the curves (AUC) being 0.93 with an optimal cut-off of 40.85 with maximum sensitivity and specificity of 100.0% and 93.0%.

Conclusion: Autonomous Regulatory Index (ARI) index with significantly excellent predictive ability in detecting the higher health risk can be used as a non-invasive measure of autonomic dysfunction among COVID-19 subjects.

Cognitive Accessibility for Digital Health

L. Wills ¹, T. Bailey¹, R. Ivimey-Cook¹

¹Globefox Health, Farnham, Surrey, United Kingdom

Our good health and wellbeing depends on having access to the right medical attention and care at the right time. People need to access critical information and to be able to identify and communicate their concerns and experiences effectively. Currently, many struggle with this, either because they have disabilities, or because of reduced capacity or cultural barriers… or because they simply can’t use digital services. While a great efforts are now going into the development and production of resources for people with learning disabilities, most common cognitive issues are not properly considered, and adequate measures are rarely put in place. This is highly problematic, as alongside better recognized issues such as dementia, head injuries and neurological disorders, conditions such as menopause, heart disease, head injury, and gut disorders can all impair cognition. Stress, some medications, and substance abuse can also lead to significant impairments, which the individual may not even be aware of. Cognitive Support is often needed to overcome: • broken ‘journeys’ • too many steps to follow • unreliable technology or interactions • overly complex processes and procedures • use of medical jargon and acronyms • unclear, treatment plans, provision and pathways • repetitive, intrusive or highly irrelevant requests • services concepts and options that don’t quite fit or are poorly framed • not having regular access to the internet due to practical offline barriers • lack of appropriate, accessible online support or help There is a common assumption that it’s OK to ‘muddle through’ things especially where people are not confident in being able to complete a task or process successfully. This is not ideal in any situation but is unacceptable when people’s health and wellbeing is at stake. A preliminary study used service design, user centric approaches, and rapid ethnography to capture and examine the experiences of people using digital health tools and services, and outline remedies. Critical findings include: • 12.5% of people said that they were confident in using digital health • 66% had cognitive issues that affected their ability to use digital health • 41% of people said that tiredness was a factor.

Measuring Diagnostic Error: Patient Complaints and Grievances

M. Kanter ¹, L. Lurvey¹, K. Litman¹, M. Hugh¹, M. Au¹, M. Arteaga-Garavito¹*

*(Presenting Author)

¹Kaiser Permanente, Pasadena, CA

Measuring the incidence of diagnostic error is challenging and typically requires extensive chart reviews post-adverse events. Identifying a standardized measure could reveal patterns in the diagnostic process. The Southern California Permanente Medical Group assessed a reporting system of patient’s complaints and grievances to be used as a standardized measure for diagnostic error. Like claims data, complaints and grievances measure diagnostic error from the patient’s perspective but may have greater volume and be more timely. A random sample of 200 patient complaints were evaluated to determine if the elements of diagnostic error were present. The National Academy of Medicine defines diagnostic error as the failure to establish an accurate, or timely diagnosis or communicate that explanation to the patient. The sample was selected from patient complaints received between November 2019 - March 2020. Complaints are categorized into multiple types but we focused on those that would likely be related to diagnostic error including: 1) diagnosis, 2) delay in diagnosis, 3) diagnostic work-up lacking, 4) test results delayed, missing or unavailable and 5) failure to keep patient or family informed. Each selected case was reviewed by a clinician and non-clinician reviewer to consider both physician’s and patient’s perspectives Patient reviewers focused their assessment of the complaint submitted by the patient, while clinician reviewers focused on the information available in the medical record. Results from this review showed that 88 percent of cases reviewed from the patient perspective had at least one of the elements of diagnostic error, versus 18 percent of cases reviewed from the physician perspective. Specificity of issue types 1 to 4 ranged between 93 and 69 percent from patient’s perspective, and between 20 and 15 percent from physician’s perspective. Issue type 5 complaints were mostly not related to diagnostic error. These results suggest patient complaints may support the measurement of diagnostic errors by capturing communication breakdowns from the patient’s perspective that may not be observed if solely analyzing information available in the patient’s chart. The patient’s perspective should be the gold standard regarding accurate and timely communication, and whether explanation of the health problem was well understood. Trended complaints data showed sensitivity to changes in procedures such as the coronavirus pandemic, specifically around the communication of test results early in the pandemic. These findings suggest that patient complaints are sensitive to changes in care delivery. Further analysis on specific diagnosis associated to the patient complaint is being undertaken.

Development and Testing a New Framework to Improve Diagnostic Reasoning in Medical and Health Professionals

T. Batteson ¹, J. Carlson¹, M. Chirica¹

¹Rosalind Franklin University of Medicine and Science, North Chicago, IL

The Institute of Medicine (IOM) categorized four types of error: diagnostic, treatment, preventive, and “other”; recommending that considerable focus should be given to diagnostic ‘blind spots’ that ‘continue to harm an unacceptable amount of people’ and cause high societal costs. Approximately 80% of diagnostic errors reported are related to cognitive errors often categorized as; faulty knowledge, faulty data gathering, and faulty synthesis. These cognitive errors lead to diagnostic error through factors such as overconfidence bias, anchoring, and/or premature closure around inaccurate diagnostic hypotheses. There is growing interest in solutions that help providers minimize inaccurate diagnoses including more robust training around cognitive bias related to the development of diagnostic hypotheses and the use of cognitive forcing strategies such as the use of Diagnostic Reminder Systems (DRS) when determining diagnoses. This study aimed to explore the impact of a DRS on student and resident diagnostic reasoning and confidence. We developed and implemented a quasi-experimental method to assess the impact of a readily available web-based DRS (Isabel PRO) on diagnostic accuracy in medical students (M2 and M3), physician assistant student (PAS-1 and PAS-2), and medical resident (PGY1 and PGY2) populations. Fourteen case simulations were created based on diagnoses known to be frequently missed or delayed in actual clinical settings. Experienced standardized patients (SPs) were trained for the standardized simulation and portrayed the patient cases during clinical skills exams or exercises delivered as part of their education. Paired sample t-tests were conducted on the data in-order to assess pre- and post-diagnostic accuracy across all cases and educational years. A statistically significant change occurred, with mean accuracy scores increasing (pre-accuracy score, M=2.63; post-accuracy score, M=2.93), t(518) = 14.93, p <.001. Overall diagnostic accuracy significantly improved after using the Isabel DRS for the majority of cases studied across each level of education. Using a DRS system appears to have improved novice practitioners’ diagnostic decisions and accuracy and reduced the impact of their cognitive biases. This suggests that using a DSR system in pre-licensure education could benefit how learners approach diagnostic reasoning by training them to check their cognitive processes. The two groups that appeared to benefit this the most were Group 5 (M2) and Group 3 (PAS-2) suggesting that their level of training and knowledge was at the optimum stage to introduce a cognitive checking system. Longitudinal research needs to be conducted to assess this further.

Figure 1: Overall pre-post diagnostic accuracy scores by study cohort.

all cohorts demonstrated statistically significant improvement pre-post use of a DRS

Patient-Perceived Breakdowns in Telehealth-Diagnosis Early in the COVID-19 Pandemic. An Exploratory Mixed Methods Study

M. Brahier ¹, K. Baker², M. Penne¹, S. Davis³, M. Hill², K. M. Smith²

¹Georgetown University School of Medicine, Washington, D.C.

²MedStar Health, Washington, D.C.

³Howard University, Washington, D.C.

Background: Diagnostic error is a leading cause of patient harm in the U.S. An increased number of patients received diagnoses via telehealth during the COVID-19 pandemic. Telediagnosis is defined as diagnosis through remote interactions and transmitted data, including clear communication of the diagnosis to the patient. The rates of diagnostic error in telediagnosis for primary care encounters are currently unknown. Our study aimed to explore patient-perceived breakdowns in telediagnosis by examining patient satisfaction surveys completed within six weeks post-visit by patients with primary care encounters early in the COVID-19 pandemic.

Methods: We used an exploratory mixed methods study design to identify patient-perceived breakdowns in diagnosis for adult patients (≥18 years) participating in a primary care telehealth visit between April-May 2020. Open comments on a patient experience survey which were sentiment coded as “negative” were systematically coded using content analysis to identify patient-perceived breakdowns in the accuracy, timeliness, and communication of diagnosis and subcategorized using the DEER taxonomy. Results for telehealth encounters were also compared to those for in-person encounters.

Results: A total of 559 negative comments were analyzed, with 243 (43.5%) from telehealth encounters. Most telehealth patients were women (n=144; 59.3%), White (n=151; 62.1%) or African American/Black (n=48; 19.8%) and more often completed their survey via the internet (n=156; 64.2%) vs the mail (n=87; 35.8%; p<0.001). Telehealth patients reported an average age of 66.0±14.7 years, which was significantly older than patients with in-person visits (59.4±18.5; p<0.001). Of the 243 negative comments from telehealth visits, 52 patients (21.4%) perceived a breakdown in diagnosis. Breakdowns in diagnostic accuracy were significantly more likely in telehealth visits than in-person (50% vs 30%, p = 0.018), while breakdowns in timeliness were significantly more likely for in-person visits (33% vs 52%, p = 0.024). Breakdowns in telediagnosis were most commonly issues with physical exam/assessment (n=15; 29%), testing (n=14; 27%), and treatment (n=13; 25%).

Conclusion: We identified patient-perceived breakdowns in the diagnostic process from a small sample of patients who completed telehealth visits early during the COVID-19 pandemic. Future research and a larger sample can further elucidate the impact of telediagnosis on diagnostic error rates in primary care.

Development and Evaluation of an EHR-Supported System to Deliver Patient Outcome Feedback to Pediatric ICU-Referring Clinicians

C. Cifra ¹, C. Tigges¹, S. Miller¹, N. Curl², C. Monson¹, K. C. Dukes¹, H. S. Reisinger¹, P. R. Pennathur¹, D. F. Sittig³, H. Singh⁴

¹University of Iowa Carver College of Medicine, Iowa City, IA

²Unity Point Health-Trinity, Iowa

³University of Texas Health Science Center at Houston, Houston, TX

⁴Michael E. DeBakey Veterans Affairs Medical Center and Baylor College of Medicine, Houston, TX

Background: Critically ill children are commonly seen at emergency departments (EDs) by clinicians with variable pediatric training before transfer to the pediatric intensive care unit (PICU). Because clinicians develop expertise by learning from past performance, communicating information regarding transferred patients’ clinical outcomes to ED clinicians may be an effective way to improve pediatric emergency care. We conducted an iterative evaluation of an electronic health record (EHR)-enabled patient outcome feedback system from PICU to ED clinicians.

Methods: Using a human factors engineering approach informed by the Health Information Technology (IT) Safety Framework, we iteratively designed, implemented, and evaluated a semi-automated electronic feedback system leveraging the Epic EHR at a single institution. Over 5 months, we generated one-page update reports for all patients transferred to the PICU within 24 hours of an ED visit at the same hospital. The report included the patient’s current diagnosis, urgent interventions/procedures performed, and disposition at 3 days post-PICU admission. Reports were routed to PICU physicians who edited and sent them to ED physicians. We evaluated the system’s technical feasibility, usability, and relevance using EHR access logs, usability testing with think-aloud methods, and qualitative interviews.

Results: Eighty-eight update reports were routed to 10 PICU physicians, who sent 73 (83%) reports to 22 ED physicians. Sixty-seven (92%) reports were read. Most content was pulled automatically from the EHR (88% of mean word count). PICU physicians added a mean of 32 words/report (9% typed and 2% selected from drop-downs) and rarely edited pre-populated content. ED physicians read reports a mean of 7.1 (SD 4.9) days from seeing the patient. Usability testing with 6 PICU physicians revealed a mean System Usability Scale score of 88.3 (SD 7.9). Update reports were completed at a mean of 4.7 (SD 2.9) minutes, with more experienced users taking less time. PICU physicians indicated that the process minimally affected their workload. ED physicians were satisfied with feedback content and delivery.

Conclusions: Preliminary results showed that an EHR-enabled system to deliver timely and relevant PICU patient outcome feedback to referring ED clinicians was technically feasible, usable, and relevant. Next steps include wider implementation and evaluation.

Recognition and Description of Diagnostic Error among Pediatric Readmissions

M. Congdon ^1,2, B. Carroll^1,2, A. Costello^1,2, W. Chua^1,2, V. Fairchild¹, Y. Fatemi^1,3, M. Greenfield^1,2, D. Herchline¹, A. Howard¹, C. E. Lamberton^1,2,4, L. McAndrew^1,2, J. Hart^1,2, K. N. Shaw^1,2, I. Rasooly^1,2

¹Department of Pediatrics, Children’s Hospital of Philadelphia, Philadelphia, PA

²Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA

³Division of Infectious Diseases, Children’s Hospital of Philadelphia, Philadelphia, PA

⁴Division of General Internal Medicine, Hospital of the University of Pennsylvania, and Pennsylvania Presbyterian Medical Center, Philadelphia, PA

Background: The epidemiology of hospital-based pediatric diagnostic error (DE) has not been well described. The contribution of DE to pediatric hospital readmissions is unknown. Characterization of pediatric DE is critical to developing targeted improvement approaches.

Objective: We sought to describe the incidence and patterns of DE among pediatric patients hospitalized on the general pediatric service at an urban freestanding children’s hospital between 10/1/2018 and 9/30/2020 and readmitted within 15 days of discharge.

Methods: Each admission/readmission encounter was independently reviewed by two pediatric physicians using a validated diagnostic error identification instrument (SaferDx). Cases in which one or both reviewers identified DE were reviewed using the modified DEER Taxonomy to characterize points of failure in the diagnostic process. Interrater reliability in identifying DE will be evaluated.

Results: We reviewed 348 cases of 321 unique patients. In 44 cases (12.7%) one or both reviewers identified DE. Preliminary analysis of the first 7 cases showed that cardiac, airway, and infectious conditions were most frequently misdiagnosed. Diagnoses associated with DE included missed endocarditis, cardiac shunt, laryngeal cleft, tracheal stenosis, osteoarticular infection, Clostridium difficile colitis, anti-MOG encephalitis, and hyperventilation syndrome. Analysis by DEER taxonomy revealed that failure points occurred during the history, hypothesis generation, and testing steps in the diagnostic process (see Figure).

Conclusion: Among pediatric patients who experienced readmission there were missed opportunities for making a timely or accurate diagnosis in 12.7% of cases. Evaluating general pediatric readmissions may be helpful in identifying DE among children and serves as a feasible strategy for identifying opportunities to improve pediatric diagnosis.

Culture’s Impact on the Expression and Diagnosis of Anxiety

D. Dawson ¹, P. Chen², C. Tsao¹, T. Fletcher²*

*(Presenting Author)

¹Michael E. DeBakey Veterans Affairs Medical Center, Houston, TX

²Baylor College of Medicine, Houston, TX

Background: Few studies have investigated the influence of culture on the anxiety disorder diagnostic process. Under-recognition of anxiety, misidentification of anxiety symptoms, and barriers in treatment access result in greater severity of symptoms once racial/ethnic minority patients initiate care (Dawson, 2021). Illuminating the role of culture—both the patient’s and the provider’s— may allow for greater recognition of anxiety and improved diagnostic assessment.

Method: We conducted qualitative interviews with a total of 32 mental health providers working in the Veterans’ Healthcare Administration: 20 Female; 11 Male; 1 gender-queer; 18 White; 8 Black/African American; 3 Hispanic/Latinx; 3 Other. Interviews were drawn from a national, randomized list of providers. We asked providers: “How do you think culture impacts the expression and diagnosis of anxiety?” Responses were analyzed using a grounded theory approach, which allowed themes to emerge organically from the data.

Results: All 32 providers agreed that a patient’s culture impacts the expression of anxiety. All but one (31) said that their own culture influenced engagement with patients and diagnosis of anxiety. Four themes emerged: Patient’s Military Culture: Providers described military culture as an important influence that could limit a patient’s willingness to discuss weakness or express emotions, and a pervasive stigma within military culture toward seeking mental health care. Patient’s Ethnic/Racial Culture: Providers also discussed how patients’ ethnic/racial culture may influence their presentation of anxiety, level of disclosure, and even the words they might use to describe their condition. Provider’s Personal Background: Some providers discussed how their own personal or cultural connection with anxiety influenced their diagnoses and ability to immediately empathize, or not, with a patient’s anxiety. Patient & Provider Cultural Interactions: Providers who shared similar cultural backgrounds with their patients described how it allowed better engagement. Conversely, providers coming from a different culture than their patients —be it in terms of gender, race, or military experience— described how this influenced their assessment and approach to diagnosis. One provider described how cultural misunderstanding can result in misdiagnosis, while another described a strategy of finding common ground to facilitate understanding and reduce the likelihood of diagnostic error.

Conclusion: Findings indicate that it is important to have an awareness of culture and the role it plays in the diagnostic process. Providers’ ability to reflect on how their own culture interfaces with a patient’s culture can improve patient-provider relationships, diagnostic processes, and patient outcomes.

Visualizing the Patient Experience: A Systematic Review of Patient Journey Mapping and Implications for Diagnosis

J. Shi ¹, T. Scott², L. Oullette³, H. Haskell⁴, G. True⁵, H. Singh², T. Giardina²

¹Rice University, Houston, TX

²Houston VA & BCM, Houston, TX

³Texas Medical Center Library, Houston, TX

⁴Mothers Against Medical Error

⁵South Central MIRECC, Houston, TX

Background: Better patient experience is integral to patient-centered care and has been positively associated with safety and effectiveness and improved health outcomes. Patient journey mapping—often drawn from process mapping in lean design—is a visual representation of a patient’s experience interacting with the healthcare system. Such a map may uncover problem areas otherwise challenging to capture. We conducted a systematic review to explore the use of patient journey mapping methodology and additionally evaluated its use to visualize the diagnostic process.

Methods: We identified published literature up to March 2020 using Medline OVID, Embase, Scopus, and CINAHL. A trained medical librarian created a systematic literature search using MeSH terms and keywords. The final number of results after de-duplication was 1,111. At the title/abstract review stage, we excluded 894 papers based on topic and publication type (e.g., editorials, conference proceedings, abstracts). At the full text review stage (n=217), we excluded 118 articles based on publication type (e.g., conference proceedings, short reports, methods focused), 84 articles focused on system factors without patient input (e.g., improving throughput, assessing wait times, economic analyses), and 3 papers where full-text was inaccessible.

Results: Final analysis included 12 articles (Table 1). All articles included a process map or visual care pathway representing the patient journey; however, only seven articles integrated patient data (themes identified from qualitative data) into the map. Nine articles identified diagnosis as a step in the map and discussed associated barriers or gaps in care (e.g., communication, referral, and coordination issues). Only two of those nine articles conceptualized diagnosis as a process involving multiple steps and highlighted where in the process barriers occur (e.g., referral to specialist, wait time to testing and/or notification).

Conclusion: Existing literature on patient journey mapping methodology is sparse despite its potential to highlight areas of patient concern. Additionally, neither data on patient perspectives nor nuances of the diagnostic process have been well integrated visually. With additional research and development, visual mapping of the longitudinal patient journey could potentially become a powerful tool to identify areas of improvement within the diagnostic process most valued by patients.

Table 1. Articles Included in Final Analysis with Findings Related to the Diagnostic Process

The Current State of Diagnostic Error Education in US Medical Schools

D. Hall ¹, S. LaBlance¹, B. Todd²

¹Oakland University William Beaumont School of Medicine, Rochester, MI

²Beaumont Health System, Southfield, MI

Background: Diagnostic error has been identified by the Institute of Medicine as a major cause of patient harm. Despite this recent focus on the importance of diagnostic error, little is known about how medical schools are educating future physicians on diagnostic error, specifically if diagnostic error is taught, when in the curriculum it is taught, and the format for teaching. We aimed to investigate the current state of education on diagnostic error in US medical schools.

Methods: We conducted an anonymous survey of deans of medical education at US MD and DO medical schools utilizing Qualtrics. The study was distributed in February 2021 through a listserv known to reach faculty at all US medical schools. The survey was resent on two occasions to improve the response rate. The survey asked questions concerning diagnostic error education at that school, including whether or not it was taught, details about the curriculum if it is taught, and questions about obstacles the school is facing if it is not taught.

Results: 47 deans of medical education responded to at least one question of the survey out of a potential 192 deans of medical education. Of the schools that responded, 93.6% were MD schools, and 6.4% were DO schools. 83.7% of schools teach diagnostic error, while 16.3% schools do not. 68.0% of schools teach diagnostic error during both the preclinical and clinical years, 28.0% teach it during only the preclinical years, and 4.0% teach it during only the clinical years. However, 93.1% of schools felt the clinical years were an appropriate time to teach diagnostic error, while 69.0% of schools felt the preclinical years were appropriate. Small-group discussions are the most commonly used format for teaching diagnostic error, utilized by 87.5% of schools. Other teaching formats used include didactic lectures (79.2%), online educational modules (41.7%), workshops (33.3%), simulation (33.3%), and flipped classroom (29.2%).

Conclusion: To our knowledge, this is the first study investigating diagnostic error education in US medical schools. The results suggest that the majority of US medical schools do teach diagnostic error in their curriculum, and that it is more commonly taught in the preclinical years using small-group discussions. Future research should investigate the effects of diagnostic error education in medical school on patient outcomes.

Improving Diagnostic Safety by Analyzing Human Errors from Serious Adverse Events in the Emergency Department

M. Baartmans ¹, J. Hooftman ², L. Zwaan³, S. van Schoten⁴, C. Wagner⁵

¹Nivel, Amsterdam, Netherlands

²Amsterdam UMC & Erasmus MC, Amsterdam, Netherlands

³Erasmus MC, Rotterdam, Netherlands

⁴Amsterdam UMC, Amsterdam, Netherlands

⁵Nivel & Amsterdam UMC, Amsterdam, Netherlands

Background: Human errors have a large contribution to diagnostic errors in the emergency department. These errors are difficult to research reliably, because most data is gathered retrospectively without information on decision-making processes. For the current study, we have used reports that were created after a serious adverse event (SAE) related to the emergency department. These reports contain information from interviews with the involved clinicians and thus are a rich source of information. By analyzing the human errors from these reports, we can better understand why they are made and formulate more specific recommendations to improve the diagnostic process.

Methods: Twenty-three SAE-reports of diagnostic adverse events in emergency departments of Dutch general hospitals were analyzed. The human errors were identified and two researchers independently applied the Safer Dx Instrument, Diagnostic Error Evaluation and Research (DEER) taxonomy, and the Model of Unsafe acts to analyze them.

Results: Based on the Safer Dx Instrument, we excluded two SAE-reports as they did not seem to involve a diagnostic error. Seventy-three human errors were identified in the other reports (see table), which were in most cases based on intended actions (N=69) and could be classified as mistakes (N=59) or violations (N=13). Using the DEER taxonomy, we found that most human errors occurred during the assessment and testing phases of the diagnostic process. Mistakes were most often found in combination with the assessment phase (N=48). Violations were often found in combination with the diagnostic testing phase (N=7).

Conclusion: Results indicated that interventions should be aimed at mistakes and violations, which both are intended actions and therefore might be amenable to active interventions. Use of, for example, diagnostic decision support systems, second opinions, or structural feedback in specific vulnerable phases of the diagnostic process could help to reduce mistakes. Solutions to reduce violations can best be achieved by changing organizational and contextual factors, such as lowering work pressure and crowding in the emergency department. Besides the overall recommendations, we believe that these analyses, would be useful for hospitals to use after a (diagnostic) adverse event to help formulate more specific interventions for improving diagnostic safety.

Linguistic Indicators of Diagnostic Uncertainty in Clinical Documentation for Hospitalized Children

T. Marshall ¹, L. Nickels², E. Edgerton², P. Brady¹, J. Lee², P. Hagedorn¹

¹Cincinnati Children’s Hospital, Cincinnati, OH

²University of Cincinnati, Cincinnati, OH

Background: Diagnostic uncertainty remains challenging to study as there is no validated method for identifying it in clinical practice. We previously developed a novel electronic health record label for hospitalized children with uncertain diagnoses (UD). This study aims to identify linguistic patterns in clinical documentation associated with diagnostic uncertainty.

Methods: This case-control study included patients on the hospital medicine service at Cincinnati Children’s Hospital Liberty Campus. Cases were defined as patients prospectively labeled by physicians with an UD during their admission. Control patients did not receive the UD label and were matched up to 4:1 to cases on age strata, biological sex, and date of admission to adjust for differences in diagnostic considerations. An electronic query was used to extract clinician admission and progress notes and create two corpora for cases and controls. We performed a keyword analysis to identify discrepancies in term frequency between these corpora, followed by a qualitative concordance analysis for terms of interest and investigating the context surrounding their use to identify patterns indicative of diagnostic uncertainty. A physician and a linguist reviewed the resulting list of potential linguistic indicators to validate observed patterns. We utilized MetaMap, a publicly available natural language processing (NLP) program, to identify and categorize medical terminology into semantic types (i.e., sign or symptom). Final analyses included only semantic types which categorized accurately and those relevant to the discourse patterns of diagnostic uncertainty. Following identification of confirmed linguistic indicators (Table 1) and optimization of MetaMap data, we tested the performance of several machine learning (ML) algorithms in identifying diagnostic uncertainty in clinical documentation.

Results: Our final corpora included 664 case and 2406 control notes. A random forest model using the confirmed linguistic indicators and MetaMap terms had sensitivity and positive predictive value of 87.9% and 97.2% respectively, with accuracy and cross validation scores of 87.0 % and 83.9%.

Conclusion: This study used a combination of NLP and ML methods with human validation to build a model to detect diagnostic uncertainty in clinical documentation. This methodology represents a promising approach to detecting, studying, and ultimately mitigating diagnostic uncertainty in clinical practice.

Association Between the COVID-19 Pandemic and Follow-Up of Cancer-Related Abnormal Test Results

A. Meyer ¹, H. Singh¹, A. Zimolzak¹, L. Wei¹, A. Marinez¹, D. R. Murphy¹

¹Michael E. DeBakey VA Medical Center and Baylor College of Medicine, Houston, TX

Background: The COVID-19 pandemic decreased patients’ interactions with healthcare, delaying care for both routine follow-up and serious medical issues. Significant reductions in new cancer diagnoses have been found, suggesting delays in cancer diagnoses. We hypothesized that the pandemic reduced follow-up of cancer-related abnormal test results, which could lead to delayed cancer diagnoses.

Methods: We utilized validated electronic triggers to identify delays in follow-up from national Department of Veterans Affairs clinical data before and during the COVID-19 Pandemic (3/11/2019-12/31/19 and 3/11/2020-12/31/2020, respectively; using the World Health Organization’s pandemic declaration as pandemic start). These triggers scan clinical data and identify patients with delayed follow-up of abnormal test results (i.e., red-flags) for five possible cancers, including bladder (urinalysis with high-grade hematuria), breast (abnormal mammograms), colorectal (positive fecal occult blood tests [FOBT] and laboratory results consistent with iron deficiency anemia [IDA]), liver (alpha-fetoprotein [AFP]), and lung (chest imaging suggestive of malignancy). Follow-up was considered delayed after 60 days, except for chest imaging, where 30 days was used. Outcome measures included test quantity (reported using descriptive statistics) and proportion of abnormal tests lacking appropriate follow-up (compared between periods utilizing tests of proportions). Statistical process controls were used to examine temporal effects of the pandemic on test result follow-up proportions (from 3/11/2019-12/31/2020).

Results: Test quantities decreased between pre-pandemic and pandemic periods by 15.2-30.6% depending on the test. Proportion of abnormal results lacking follow-up increased for FOBT (3.5%), AFP (0.8%), and chest imaging (2.2%; Table). Controls charts revealed significant increases in delays at two stages: ‘Early’- around the transition from pre-pandemic to pandemic periods for urinalyses, mammograms, FOBT, laboratory results consistent with IDA, and chest imaging; and ‘Late’- 30-45 weeks into the pandemic for mammograms, FOBT, and laboratory results consistent with IDA.

Conclusion: Delays in follow-up of abnormal test results suspicious for cancer during the COVID-19 pandemic had a bi-phasic pattern. While early delays were expected, the late period of delays suggest potential strain of healthcare services as patients reentered the healthcare system. Outcomes related to delays in cancer diagnoses need additional study and interventions are needed to mitigate the impact of pandemic-related delays in care.

Assessing the Assessment - Design and Deployment of a Tool for Evaluating Clinical Notes’ Assessments’ Quality

M. Mirica ¹, M. Khazen², J. Ramos³, S. Hussain⁴, G. Schiff¹

¹Brigham and Women’s Hospital, Boston, MA

²University of Tel Aviv, Tel Aviv, Israel

³Emory Medical School, Atlanta, GA

⁴Mass General Hospital, Boston, MA

Background: While the importance of ambulatory diagnostic errors is now being increasingly recognized, there is a need to advance diagnostic safety to both target high risk areas and advance our ability to measure the quality of the diagnostic process. Urgent care visits for acute symptoms and accompanying clinical assessment documentation areas fraught with uncertainties and quality issues, particularly related to unexpected or potentially missed diagnoses. We designed and deployed a new instrument to assess the quality of documented diagnostic assessments found in urgent care notes to better understand address key diagnostic assessment issues.

Methods: We conducted a retrospective chart review of 100 urgent care encounters with 20 primary care physicians occurring January-December 2018. We iteratively designed and used the Assessment of the Assessment tool (AAT) to review the quality of the recorded diagnostic assessment. AAT covered a variety of diagnosis assessment quality domains including: adequacy of addressing the chief complaint, differential diagnosis, clinical uncertainty, contextual/psychosocial issues/impact, diagnostic follow-up plan, situational awareness and safety nets, avoiding legal liability/pejorative or stigmatizing language, and global subjective assessment of the note’s diagnostic assessment. A total of 18 items were scored on a 7-point Likert scale which reflected both presence of an item and quality of the documentation (0-Absent, 1-Minimal, 2-Less than good, 3-Neutral, 4-Good, 5-Very good, 6-Excellent). 16 training encounters’ charts were used to calibrate the tool, refine operational definitions and scoring of the items, and resolve any disagreements between the coders. After the calibration, a clinically trained researcher scored the 100 urgent care charts.

Results: Ratings of don’t miss diagnoses were the highest (mean 2.7 [SD 1.7]), followed by quality of differential diagnosis (mean 2.1, SD [1.7]), situational awareness of red flags (mean 2.0 [SD 1.4]), and uncertainty 1.9 [SD 1.6]. The highest-scored individual item was avoiding legal liability pejorative red flags (mean 6.0, SD [0.0]), while noting pitfalls received the lowest scores (mean 0.5, SD [1.4]).

Conclusions: Overall, AAT was able to reliably capture the elements of notes’ diagnostic assessment that expert consensus suggested should be present in the chart. Various areas of strength vs. areas needing improvement were noted. In particular, better documentation of clinician’s thinking and rationale, and situational awareness of red flags and potential diagnostic pitfalls would improve the quality of diagnostic assessments, and thereby potentially prevent harmful diagnostic errors.

Missed and Delayed Diagnosis of Endocarditis

P. Polgreen ¹, A. Miller¹, A. Arakkal¹, S. Koeneman¹, J. Cavanaugh¹, P. Polgreen¹

¹University of Iowa, Iowa City, IA

Background: Prior to the introduction of effective antimicrobial therapy, endocarditis was universally fatal. Even in the post-antimicrobial era, mortality related to endocarditis remains high, and delays in diagnosis increases the likelihood of worse clinical outcomes. However, the disease is difficult to diagnose because it shares symptoms with other diseases and syndromes. The purpose of this project is to identify the incidence and duration of diagnostic delays for endocarditis and to also describe risk factors associated with diagnostic delays.

Methods: We conducted a retrospective cohort study using data from the Truven Health Analytics Commercial Claims and Encounters and Medicare Supplemental databases from 2001-2017. We identified the initial (index) endocarditis diagnosis for each patient and analyzed prior visits for symptomatically-similar diagnoses (SSD) (e.g., fever) during the year prior to the endocarditis diagnosis. We employed a change-point-detection analysis to identify the time window prior to the index endocarditis diagnosis where diagnostic opportunities first appeared. Finally, we used a bootstrap-based simulation model to estimate the frequency and duration of potential diagnostic delays.

Results: The study cohort included 224,324 patients with endocarditis. In the year prior to the index diagnosis, 197,241 (87.9%) patients had at least one SSD-associated visit, and we observed a dramatic increase in such visits in the weeks immediately before the endocarditis diagnosis. We estimate that roughly 54.3% (95%CI: 53.7-55.0%) of patients experienced at least one missed opportunity, that 361,091 visits (95%CI: 352,158-370,412) represented a missed opportunity, and the longest diagnostic delays occurred up to 98 (95%CI: 91-105) days prior to the index diagnosis. Across all patients the mean duration of delay was 18.78 days (95%CI: 17.52-20.19), and for those who experienced at least one missed opportunity, the mean duration of delay was 34.6 days (95%CI: 32.6, 36.8). Approximately 27% of patients experienced a delay longer than 30 days. On average, patients who experienced a diagnostic delay had 3.0 visits (95%CI: 2.9, 3.0) representing a missed opportunity. Missed diagnostic opportunities occurred more often in outpatient settings (83.3%) compared to inpatient (5.5%) or emergency department settings (11.2%). Diagnostic delays were more likely to occur among patients who were older, injection drug users, during weekend visits, and in outpatient or ED settings.

Conclusion: Diagnostic delays for endocarditis are common and the risk factors we identified will help inform future interventions designed to decrease diagnostic delays for this important infectious disease.

Misdiagnosis of Cellulitis in the Emergency Department: A Threat to Public Health and Patient Safety

M. Pulia ¹, R. Schwei¹, E. Harwick¹, A. Haleem¹, J. Hess¹, R. Glinert¹, T. Keenan¹, J. McBride¹, R. Redwood¹

¹University of Wisconsin-Madison, Madison, WI

Background: A recent, single center report indicated that cellulitis is misdiagnosed in up to 30% of cases in the emergency department (ED) due to the presence of non-bacterial mimics (pseudocellulitis). These diagnostic errors result in excess healthcare spending and threaten public health (bacterial resistance) and patient safety (adverse drug events). The objective of this study was to examine the proportion of ED cellulitis misdiagnosis by comparing the treating physician’s diagnosis to a consensus diagnosis from a multidisciplinary expert panel.

Methods: We prospectively enrolled patients presenting to the ED with acute lower extremity complaints that involved visible erythema. We abstracted the treating physician’s diagnosis from the patient’s medical chart as either cellulitis or pseudocellulitis. Gold standard diagnosis was determined by majority consensus among a six member multidisciplinary (Infectious Diseases, Dermatology and Emergency Medicine) physician review panel. The panelists independently reviewed the clinical documentation (except medical decision making, diagnosis and treatment), photographs of the affected skin, laboratory/imaging results. The treating physician’s diagnosis was compared to the consensus panel diagnosis to determine diagnostic error rates.

Results: The final sample included 204 patients, 59% were male and the average age was 57 years. Overall, the diagnosis assigned by the treating physician and consensus panel was discordant in 50 cases (24.5%) (Table 1). The treating physician diagnosed 138 (67.6%) of the patients with cellulitis while the expert panel assigned a cellulitis diagnosis in only 92 case (45.1%). The proportion of cellulitis overdiagnosis by the treating physicians was 34.7%. Underdiagnosis of cellulitis by the treating physician was only observed in 2 cases (0.09%).

Conclusion: Consistent with other reports, we observed that nearly 35% of cellulitis cases diagnosed in the ED were not cellulitis when compared with a consensus diagnosis. This further validates the diagnostic challenge and antibiotic stewardship threat posed by cellulitis. There is an urgent need to develop interventions that reduce diagnostic error in the evaluation of cellulitis.

Towards Understanding the Contribution of Communication Failures to Process-Based Medical Error in Adolescent Transitional Care

S. Schepps ¹, L. Garbayo¹

¹University of Central Florida, Orlando, FL

Background: Adolescent transitional care for patients with chronic conditions is habitually associated with challenges to patient-centered decision-making and other system-level barriers for a successful transition. Medical error by “omission or commission in planning or execution that contributes or could contribute to an unintended result” is studied herein in the context of adolescent transitional care of failed transitions that may lead to misdiagnosis of conditions. The project aims at identifying types of medical error associated with adolescent patients with chronic conditions going through transitions of care, clarifying how significant a role communication failures play under these conditions, and may inform how errors may be mitigated.

Methods: This work-in-progress is a patient-physician research partnership that combines literature review with a premedical student-patient self-ethnography. Literature was captured from PubMed using Boolean search terms between May 24th and May 29th of 2021.

Results: 84 articles were found, and 33 were considered relevant. Articles regarding medical error in adolescent transitional care for patients with chronic conditions provided a classification of two categories: errors in communication and errors in preparation. Both categories were found to be process-based errors. They were also found to be significant contributors to overall medical error for these patients. The unintended result of a failed transition was found often to be the product of a care team’s failure to communicate or prepare a patient for transition. The self-ethnography in progress considers the patient-author experience and discusses whether both types of error may be at least partially mitigated through an improvement in patient-centeredness in care.

Conclusion: Given the relevance of patient-centeredness in healthcare, the study’s preliminary analysis suggests that there is a need to extend patient-centeredness protocols to reduce process errors in adolescent transitional care due to miscommunication. Further research is needed to clarify how to better incorporate patient-centeredness in medical error studies and how its impacts on transitional care may be measured and evaluated for adolescents with chronic conditions.

The Effect of Type of Patient Referral Question on Diagnostic Processes: An Experimental Study

M. Speelman ¹, J. Staal¹, R. Brand², J. Alsma³, L. Zwaan¹

¹iMERR, Rotterdam, Netherlands

²Haaglanden MC, Den Haag, Netherlands

³Erasmus MC, Rotterdam, Netherlands

Background: Diagnostic errors are an important concern in healthcare. In the USA alone approximately 249.000 harmful diagnostic errors occur per year. Many of these errors are preventable, making it important to better understand causes of diagnostic error. Cognitive errors are seen as the main contributor to diagnostic errors. The method of referral, e.g., general practitioners (GPs) referral letters, can potentially induce cognitive error. When a patient is referred for further diagnostic work-up, the referral question can be broad or indicate a very specific diagnosis. The latter might cause the physician to focus only on information in the referral question and consequently consider fewer diagnoses, which can result in a diagnostic error. In this study, we examined the effect of a specific (both correct and incorrect diagnosis) versus a broad referral question (no diagnosis stated) on the diagnostic process.

Methods: We conducted a within-subject randomized experiment with 44 medical students in their clinical phase. All participants received six clinical cases presented as GPs referral letters. The referral question in each letter could be specific, i.e., either stating a correct diagnosis or an incorrect diagnosis, or broad, i.e., stating no diagnosis but a summary of the complaint. We analyzed the effect of the three types of referral questions on diagnostic accuracy, time taken to diagnose, confidence, and number of diagnoses stated in the differential diagnosis.

Results: There was no significant difference of the type of referral question on diagnostic accuracy: specific correct: 0,57 (0,27), specific incorrect: 0,51 (0,35), broad: 0,46 (0,34), χ2(2) = 1,446 (p=0,486). Time to diagnose also did not differ: specific correct: 133,79 (81,87), specific incorrect: 140,40 (74,78), broad: 120,21 (55,87), χ2(2)= 3,128 (p=0,209) and neither did confidence: specific correct: 6,21 (1,00), incorrect: 6,36 (1,13), broad: 6,23 (1,23), χ2(2) = 0,058 (p=0,971). Participants who received a broad referral question stated more diagnoses in their differential diagnosis compared with the specific correct and incorrect referral question (specific correct: 1,52 (0,96), specific incorrect: 1,42 (0,97), broad: 1,85 (1,09), χ2(2) =7,6 (p=0,022).

Conclusions: Clinical information presented as a GPs referral letter influenced the number of differential diagnoses given by medical student, without influencing the diagnostic accuracy, time to diagnose, or confidence. A broad referral question resulted in more differential diagnoses. Although accuracy was not influenced, it is important to be aware of the influence of a referral on a physician’s diagnostic process. Future research should determine the quality of referral questions in practice.

Developing and Testing a Bundle of E-Triggers to Identify Diagnostic Errors in Emergency Departments

V. Vaghani ¹, A. Gupta¹, P. Kapadia¹, W. Li¹, D. FSittig², H. Singh¹

¹Michael E. DeBakey Veterans Affairs Medical Center and Baylor College of Medicine, Houston, Tx

²UT Health, Houston, TX

Background: While missed opportunities for diagnosis (MODs) in emergency departments (ED) are likely more common than in primary care, their exact frequency is unknown. A rules-based electronic health record (EHR)-based trigger (e-trigger) algorithm has been used to identify selected patients likely to have MODs based on unexpected return visits. As a first step to determine the frequency of ED diagnostic error, we are developing and testing a “bundle” of six e-triggers that may provide a more comprehensive estimate.

Methods: Thus far, we have developed two out of six new e-triggers using the Safer Dx Trigger Tools Framework (Trigger I & II). For Trigger I, we used a symptom-disease dyad approach based on our recent work to identify stroke hospitalizations within 30 days of treat and release ER visits for dizziness in patients with 2 or more stroke risk factors. For Trigger II, we performed literature reviews and gathered expert input to develop an algorithm to detect unexpected transfer to ICU of a low-risk patient within 24 hours of hospitalization (Table 1). We iteratively developed and pilot-tested both e-triggers within the Veteran Affairs EHR data warehouse for years 2018-2019. We trained two physician-reviewers and pilot-tested the chart review process on 25 randomly selected charts using operational definitions and standardized procedures. Pilot cases were discussed to develop consensus and resolve disagreements. Physician-reviewers used structured electronic data collection instruments to standardize data collection and minimize data entry errors.

Results: The e-triggers were applied to >9 million patient records. Of 25 trigger positive cases reviewed thus far for Trigger I, 10 (40%) cases had MODs, 13 (52%) had no MODs, one (4%) had coding error, and one (4%) was inconclusive. Of 20 trigger positive cases for Trigger II reviewed thus far, just one (5%) case had MOD, 19 (95%) no MODs.

Conclusion: A symptom-disease based e-trigger appears to be valuable in identifying MODs in emergency care. However, because it focuses only on a single condition, determining MOD frequency will need additional complementary e-triggers. Findings support the need for development and testing of a future “e-trigger bundle” to ensure comprehensive diagnostic safety measurement.

Association Between Medical Malpractice and Diagnostic Errors in Japanese Emergency Rooms

T. Miyagami¹, T. Watari^2∗, T. Harada³, T. Naito¹

∗(Presenting Author)

¹Juntendo University, Japan

²Shimane University, Japan

³Showa University, Japan

Background: Diagnostic errors are common. The emergency department (ED) environment is particularly unpredictable and is, therefore, prone to diagnostic errors. In addition, non-emergency specialists often provide emergency care in Japan due to lack of certified emergency specialists, making diagnostic errors and associated medical malpractice more likely. While several studies have investigated the medical malpractice related to diagnostic errors in EDs, only a few have focused on the conditions in Japan.

Method: The purpose of this study was to elucidate contributing factors, such as initial diagnosis, final diagnosis, and situations that require attention in the ED setting, through the examination of diagnostic error-related medical malpractice in Japanese EDs. Data on medical lawsuits were retrospectively extracted from 1961 to 2017. Medical malpractice cases related to physician-related diagnostic errors occurring in the ED were extracted from the database. Main outcomes were the identification of the types of diagnostic errors, the final diagnosis assessed from the initial diagnosis of non-trauma cases, and the diagnostic classification of trauma cases.

Results: In this study, 74 cases were evaluated, and 28 were trauma related. The frequency of diagnostic errors within ED medical lawsuits was 78.9%. The mean age of the patients was 32 years, and 72.9% were men. The acceptance rate was 59.5%, and the median amount of acceptance was 224,121 (USD) (interquartile range=53,106-388,336). In 86.5% of the cases, the relevant errors were either missed or wrong diagnoses. Cognitive factors were associated with 91.7% of errors. Intracranial haemorrhage was the most common final diagnosis of trauma-related errors. The most common initial diagnoses of non-trauma-related errors were upper respiratory tract infection, non-bleeding digestive tract disease, and primary headache.

Conclusion: The number of medical malpractice claims within the ED involving missed or wrong diagnosis is higher than that of other departments. Such claims often developed from initial diagnosis of common diseases, such as upper respiratory tract infection, non-haemorrhagic gastrointestinal diseases, and headache.

A Novel Approach to AI General Medical Diagnosis for Better Accuracy, Efficiency, and Access

D. Jones ¹, A. Jones¹

¹Eureka Clinical Computing

Background: Misdiagnosis is the most common, deadly, and costly of all medical errors [1]. Research shows the diagnosis is wrong 10-15% of the time [2], causing 80,000 to 160,000 cases of permanent injury or death in the US annually [3]. AI diagnostic assistants (DA’s) have tremendous potential to improve quality, convenience, and availability of healthcare, but only if they have high accuracy, are evidence based, plug into the clinical workflow, provide recommendations in real-time, and both engage and inform the patient. We report superior accuracy of a prototype DA (MidasMed) that is designed to capture a comprehensive history directly from the patient, generate an initial probability-ranked differential diagnosis, and then guide the clinician towards a definitive diagnosis by suggesting the best next findings to discern at each step, while automatically updating the differential with each new input. The diagnostic engine addresses known weaknesses of current solutions with architectural, computational, and disease modeling innovations.

Methods: We compared the performance of physicians and six online DA’s: Ada, Babylon, Buoy, Isabel, MidasMed, and Symptomate. We used case vignettes [4] that were tested on several symptom checkers in 2015 [5], and on three physicians and one DA in 2020 [6]. As in the previous studies, we used only the simplified history findings, as could be entered by a patient or family member.

Results: The results are summarized in the table below. For each DA, we record the percent of vignettes for which it ranked the correct diagnosis #1, and in the top 3 suggested causes. (See table)

Conclusion: “Listen to your patient, he is telling you the diagnosis.” William Osler (1849-1919) Too rushed to take a comprehensive history, doctors today often miss important information provided by the patient for a prompt and accurate diagnosis. They may also compensate by ordering more tests or specialist consults, resulting in a diagnostic process that is too slow, inefficient, and expensive. Using a limited set of case vignettes, we demonstrated that a novel DA can capture the history directly from the patient, and generate an initial differential that exceeds the accuracy of physicians and other online DAs.