Two decades of diagnostic safety research: advances, challenges, and next steps

Algorithmic (trigger-based) event detection

Healthcare operations generate vast data sets, and most modern electronic data warehouses can be searched to identify events of interest that might otherwise be undiscovered or underreported. Electronic triggers (e-triggers) are algorithms applied to the electronic health record (EHR) to identify patients with red-flag signals or symptoms who are at risk of misdiagnosis. E-triggers have been applied to identify diagnostic errors associated with specific care environments (e.g., unexpected return visits or care escalations in emergency departments [17]) and diseases (e.g., cancer [18]). E-trigger tools can allow health systems to monitor event rates, study contributing factors, and identify targets for improving diagnostic safety. Certain e-triggers can help detect diagnostic delays before they result in harm. Another approach is the Symptom-Disease Pair Analysis of Diagnostic Error (SPADE) method for large administrative data sets (i.e., billing, insurance claims) to map frequently missed diagnoses (e.g., stroke) to one or more previously documented high-risk symptoms (e.g., headache) [19]. One limitation is that algorithmic methods rely on data that may have inaccuracies or not provide a complete representation of clinical care.

Voluntary reports

Physicians are not always aware of their errors, but they are invaluable sources of insight into the errors they do discover. For instance, voluntary surveys of physicians have revealed that diagnostic errors occur frequently, even in common conditions [20], [21]. Healthcare organizations face steep obstacles to normalizing event reporting in routine practice [22]. Diagnostic errors are often still attributed to individual physicians’ skill or intellect, and thus are especially difficult to elicit through voluntary reporting [22], [23]. Physicians are more willing to bring attention to errors when reporting is easy, efficient, and aligned with a culture that prioritizes psychological safety. An intervention that encouraged physician reporting using reminders, scheduled reflection time, and monthly progress reports was successful in increasing reports of suspected diagnostic errors in a hospital medicine service [24]. Technological solutions, including mobile apps and EHR integration, have also been developed to facilitate clinician reporting [25].

Applying a diagnostic safety lens to other safety events

Diagnostic errors may occur as part of other safety events but are not recognized as such. For example, during care escalation events, a patient may be treated for an acute condition (i.e., shortness of breath), but failure to accurately identify the underlying cause (i.e., pneumonia vs. pulmonary embolus) may result in a diagnostic error. Since diagnostic errors may be obscured during complex events, investigators have used diagnostic safety measurement tools to enhance the investigation of events that may not have been initially classified as diagnosis related. Several resources have been introduced to improve data gathering, analysis, and learning from such safety events [26], [27].

Patient-reported events

Research studies have analyzed patients’ complaints [28] and used methods to capture patients’ perspectives related to their diagnosis [29]. Policies to give patients access to their medical records, now mandated by law in the U.S. [30], have opened new doors for patients to identify care discrepancies and errors in the medical record. These methods are limited in that patients and clinicians may differ in their understanding of what constitutes an error. Event reporting mechanisms for patients require further refinement, with a broader set of pathways for event investigation and adjudication.

Addressing measurement challenges

Further consensus is needed around a shared mental model and standardized definition of a diagnostic error and/or diagnostic safety event across the research enterprise as well as the application of tools to measure these events. For example, one area that could especially benefit from standardization is diagnostic safety event reporting. Implementation of reporting standards within healthcare organizations (e.g., for internal monitoring and quality improvement) and externally (e.g., reporting out through patient safety organizations) could accelerate diagnostic safety measurement and improvement activities. Additional areas for development include standards on the timeliness of diagnosis of certain conditions, especially those prone to being missed or delayed. Resilience engineering and Safety II, as applied to diagnosis, is also a nascent yet rapidly evolving area of study that will require standardized definitions for measuring individual and organizational resilience concepts relevant to the diagnostic process [31].

Patient co-development

Funding agencies and project sponsors can promote the engagement of diverse patients and families as partners in research and measurement activities by incentivizing projects that adopt co-design/co-development approaches while ensuring that funding is available to compensate patients and family partners. The use of participatory methods, such as the co-production of tools and processes and co-designing interventions, can enhance or facilitate patient and family engagement in diagnostic safety research and quality improvement initiatives.

Multidisciplinary perspectives in research

Given the sociotechnical complexities of diagnostic safety, multidisciplinary teams and approaches are needed to accelerate progress. Multidisciplinary research teams should aim to include additional disciplines beyond clinical medicine. For example, human factors experts enhance diagnostic safety research by addressing cognitive, environmental, and organizational elements that influence clinician decision-making and workflows, but few studies have applied rigorous human factors and cognitive science approaches to optimize interface designs or cognitive support to clinicians. Other disciplines that can substantially contribute to diagnostic safety work include health informatics, social sciences, cognitive science, and implementation science, among others.

Cross-cutting and disease-agnostic research

Many studies in diagnostic safety have aimed to address diagnostic concerns in specific high-risk conditions (e.g., cancer, sepsis, pulmonary embolism). Focusing on one disease or concepts such as the “Big 3” (e.g., cancer, cardiovascular, and infectious disease) leads to a disease-centered approach that is useful for the given condition but may limit generalizability to other conditions. For example, disease-focused approaches are likely to overlook important insights on cross-cutting issues, such as the management of diagnostic uncertainty in patients with undifferentiated symptoms. We recommend that, in addition to disease-focused approaches, future efforts focus on developing more disease-agnostic approaches that consider systems, processes, and cognitive issues that are cross-cutting and underlie most diagnostic errors. Adoption of cross-cutting methods and strategies can also help researchers and healthcare organizations apply these methods to understudied populations and respond more readily to patient-centered priorities or emerging system or population needs.

Use of sociotechnical approaches in the context of health information technology

A variety of sociotechnical interventions have been developed to help close the loop on clinician communication (e.g., orders, test results, and referrals). However, these interventions have been adopted slowly in real-world practice settings, and methods to support their implementation are needed. The body of research to study and improve system usability and address poorly designed EHR interfaces that burden clinicians and lead to diagnostic errors is also small. Novel methodologies for studying diagnostic decision-making in EHR-based environments include eye-tracking and computational ethnography (e.g., collecting digital trace data or metadata) that help better understand users’ true EHR interactions and information-seeking behaviors.

Cognitive science of diagnostic errors

Prior studies implicate cognitive contributing factors in most diagnostic errors. However, studies of cognitive processes in medicine have been conducted mainly in experimental or simulation settings. We suggest two main areas for future research on cognition and diagnostic decision-making. First, we need to understand how various elements of the clinical practice environment, such as documentation demands and EHR interface designs, affect clinicians’ cognitive load and diagnostic accuracy [32]. For instance, experimental studies suggest that negative shifts in effect (e.g., during a stressful encounter) may temporarily reduce cognitive resources devoted to clinical problem-solving [33], increasing the risk of error. Testing this and related hypotheses in real-world settings will require new experimental designs, which may incorporate methods such as wearable technologies or other unobtrusive observation methods to identify higher-risk states.

Second, we need to develop interventions to help clinicians become aware of their mistakes and devise strategies to remediate knowledge or reasoning gaps. For instance, experts have called for the development and implementation of systematic feedback on diagnostic performance to improve clinical reasoning and calibration [34]. However, clinicians currently have few incentives to incorporate these strategies. While individual clinicians are often motivated to become better diagnosticians, relatively few find it feasible to overcome the barriers of limited time, resources, and support to implement new diagnostic excellence practices. Health systems, professional societies, and state/federal agencies can provide these needed resources to ensure clinicians’ participation in diagnostic improvement activities. Furthermore, any addition to clinicians’ existing burden should be supported by strong evidence favoring the new practice and, ideally, accompanied by de-implementation of practices that do not benefit the clinician or patients.

Costs related to diagnostic safety and the “business case” for improving diagnosis

A report from the Organisation for Economic Co-operation and Development (OECD) estimated that 17.5 % of total healthcare expenditures are attributable to diagnostic errors and overdiagnosis [35]. However, estimates of the economic burden of diagnostic error are based on broad projections. Little is known about how healthcare financing, organization, and payment models affect diagnostic performance. Furthermore, at the organizational level, little is known about the return on investment for interventions to improve diagnosis. These knowledge gaps make it difficult to provide the “business case” for investing in diagnostic safety interventions. Moreover, there are few, if any, external incentives that encourage healthcare organizations to implement diagnostic safety interventions. The NASEM report suggested potential policy and payment levers to drive improvements in this area, including the recommendation that accrediting organizations (such as the Joint Commission) require healthcare organizations to systematically monitor, identify, and learn from diagnostic errors [5]. However, scientific progress in this area has been slow and is needed to inform policy and reforms. Efforts to revise payment models to incentivize diagnostic safety measurement and improvement activities and appropriately reimburse the cognitive effort and teamwork are required to achieve diagnostic safety [36]. Furthermore, better methods are needed to quantify costs related to the impact of diagnostic error and of strategies to mitigate related harms.

Interventions for patient engagement

More effective partnerships with patients are needed to design and implement interventions to improve and monitor diagnostic safety. In clinical practice settings, strategies to improve communication and engagement during the diagnostic encounter are relevant and highly valued by patients but remain underused. Patients are also essential but often overlooked partners in the design and testing of patient-facing interfaces (e.g., patient portals) and reports to communicate diagnostic findings [37]. Finally, a variety of strategies exist to guide patient engagement in clinical encounters, research, and system improvement, and would benefit from implementation efforts. Future efforts should focus on engagement of patients who are more at risk of being disempowered, including older adults, patients with limited English proficiency, and patients with disabilities.

Artificial intelligence integration

Artificial intelligence (AI) has the potential to transform not only how individual clinicians and healthcare teams diagnose patients [38], but also how the diagnostic process and diagnostic errors themselves are measured [39]. AI-based tools to support diagnosis have varying evidence of validity, effectiveness, and clinical feasibility. For instance, while significant advances in AI have improved image interpretation in pathology and radiology, the role of generative AI in enhancing clinical diagnosis in real-world practice is still evolving. Rigorous research is needed to understand how clinicians interact with generative AI to arrive at a diagnosis, the risks and benefits of using AI for this purpose, how AI should be tailored to different settings and expertise, and patients’ and families’ perspectives on use of these AI tools. Moreover, as the use of AI in medicine becomes more ubiquitous, vigilance is needed to identify and study its use for tasks that can indirectly affect the diagnostic process. For example, physicians have raised concerns that the use of ambient AI scribes to create clinical notes may have the unwanted effect of taking away opportunities for deeper reflection that can lead to better diagnostic decisions [40]. Aside from studying the impact of AI-based interventions on diagnosis, pragmatic frameworks must also be developed to guide the implementation and evaluation of AI tools for diagnosis [41].

System-related interventions to improve diagnosis

Implementation of dedicated programs and activities to address diagnostic error remains uncommon in healthcare organizations. Diagnostic safety activities have been described in the context of existing quality and safety frameworks, such as learning health systems [42]. Additionally, broadly generalizable principles and recommendations have now emerged and are summarized in recent publicly available resources to support implementation of diagnostic safety practices. For example, to build national consensus on recommended practices for hospitals to improve diagnosis, the Leapfrog Group assembled a report describing 29 evidence-based practices that hospitals can use to prevent diagnostic harm to patients [43]. These types of recommended organizational structures and practices are only beginning to be evaluated systematically. Additional guidance has also been developed, such as the Safer Dx Checklist, which helps conduct a systematic organizational assessment of evidence-based safety practices [44].