Pilot investigation into the influence of racial implicit bias on physician clinical reasoning

Introduction

Clinical reasoning (CR) refers to the complex cognitive process that enables clinicians to observe, obtain, and interpret information resulting in decision-making and action (both diagnostic and management decisions) [1]. CR occurs within contexts subject to the same societal and structural influences that lead to persistent racial health disparities [2], 3]: For example, unequal access to care, racial minorities being disproportionately cared for in under resourced health systems, and delays in seeking care due to mistrust or other factors contribute to racial health disparities and are contextual factors that can affect the clinical encounter [4], [5], [6]. Racial disparities in diagnostic accuracy and timeliness of diagnosis have been demonstrated [7], [8], [9], [10], [11], [12]. Disparities are present in both diagnostic and management decisions when physicians are caring for Black patients with acute coronary syndrome (ACS) and acute injury requiring pain management [13], [14], [15], [16], [17], [18], [19].

It remains unknown if racial implicit bias impacts clinicians’ clinical reasoning (CR) and causes it to vary when seeing patients from different social groups (e.g. different races), and therefore if such variations in CR may contribute to known racial healthcare disparities. Studies exploring racial implicit bias as a contributor to disparities in CR have yielded conflicting results. Some investigators have found associations between clinicians’ implicit bias and disparities in CR [20], [21], [22], [23], [24], [25], [26], [27], [28], while others have not [29], [30], [31], [32], [33]. Characterizing the relationship between racial implicit bias and CR could identify the patient’s racial background as a contextual factor and then guide the design of CR interventions to support equitable diagnostic and treatment decisions, and ultimately, health equity.

In contrast to the conflicting research on racial implicit bias and its potential impact on physicians’ CR, there is robust evidence of the association between racial implicit bias and disparities in physician communication when seeing Black and White patients [34], [35], [36], [37]. At least three explanations for the relative strength of the evidence for the influence of implicit bias on physician communication have been proposed. First, most communication studies have been conducted in real world practice settings, in contrast to most CR studies, which evaluate physicians’ reasoning using paper or computer-based clinical vignettes. In real-world practice settings, physicians work under conditions riddled with cognitive stressors such as time constraints, environmental stressors such as noise, interruptions from both staff and the electronic health record notifications, and fatigue, among others. These stressors increase cognitive load, enhancing the probability that physicians will use heuristics – cognitive shortcuts more vulnerable to implicit bias – to guide their CR [25]. When, in real world practice settings, these stressors are combined with clinical ambiguity and increased complexity of tasks, both communication skills and CR are more likely to be influenced by racial implicit bias [38], [39], [40], [41]. On the other hand, research suggests that in low stress conditions, such as when responding to paper- or computer-based clinical vignettes, physicians are able to tap into their conscious egalitarian values and resist the influence of implicit bias on their decision-making processes [25], 29], 39]. Second, because communication may be a more automatic and affective and less cognitive and deliberate than CR it may more likely be impacted by implicit bias, regardless of the context in which it is studied. And thirdly, unlike communication challenges, the quality of CR may be more dependent on the specifics of the clinical situation [42]. For instance, CR in established conditions such as hypertension, which are linked with objective, guideline-driven approaches may be much less impacted by racial implicit bias than CR in highly ambiguous, uncertain situations [43].

To address the gaps in the literature, we previously conducted a high-fidelity simulation study to investigate the association between racial implicit bias and communication skills in physician volunteers when seeing Black or White standardized patients (SPs). Given the aforementioned known robust evidence of the association between communication disparities and racial implicit bias, we designed the simulations to include cognitive stressors and iteratively piloted and revised the simulations until they elicited the expected disparities in physician communication skills when seeing Black and White and SPs [44], 45]. In that prior work, we chose two clinical scenarios, one with ACS and the other with acute pain, due to the documented racial disparities in ACS diagnosis and treatment and also in pain assessment and management [13], [14], [15], [16], [17], [18], [19, 22], 46], 47]. All three scenarios were iteratively developed with community member, SPs, members of the investigative team, and clinical experts. The ACS case presented a diagnostic dilemma which may have been more cognitively demanding [44], it had a stronger association with disparities in communication skills ratings received by physicians when seeing a Black or White SP than the acute pain case. The acute pain case, which was a treatment decision, did result in lower communication skills ratings by female Black SPs than White SPs, but this difference was not as strongly associated with physician racial implicit bias as the ACS case.

In this paper, we seek to characterize the association between racial implicit bias and physician CR when seeing Black or White SPs in otherwise identical clinical presentations. We hypothesized that physicians’ implicit bias would be associated with lower quality CR, inclusive of diagnostic and treatment outcomes when seeing Black SPs as compared to White SPs.

Subjects and methods

Study procedures

We developed and previously described a three-station simulation exercise [44]. Briefly, each scenario included cognitive stressors common to clinical environments (e.g., time constraints, interruptions, clinical ambiguity, etc.). Because the study began during the COVID-19 pandemic, necessitating virtual delivery, the first scenario was a level-setting scenario to familiarize physicians with the process of interviewing and examining SPs over a virtual platform (Zoom™). For the second scenario, we chose acute coronary syndrome (ACS) presenting with epigastric pain, nausea, and vomiting. SPs were trained to deliver a history – consistent with ACS – of a stuttering course of progressively worsening exertional epigastric discomfort lasting longer and occurring more frequently over the last few months, culminating in the presenting symptoms. SPs were not to offer this information, they had to be asked by the physician. The third scenario was a woman in acute pain who had tried alternating acetaminophen (1 g) and ibuprofen (400 mg) for up to six total doses a day (three doses of each medication) without relief of her pain that is discovered to be due a rib fracture through X-ray result read during the encounter. During the simulation, physicians interviewed and (virtually) examined SPs, discussed their diagnostic and treatment plans with the SP, and wrote a post-encounter diagnostic and treatment plan. Both physicians and SPs were blinded to the focus of the study on racial implicit bias. To prevent unblinding of physicians to the study including racial implicit bias and to foster natural behaviors when examining the SPs, the two IATs were completed after the final SP encounter. After completing the three SP encounters and two Implicit Association Tests (IATs), physician participants were debriefed on the study purpose. Study procedures are detailed in Figure 1.

Figure 1:

Flow of procedures for early career generalist physicians participating in a three-station standardized patient exercise from 2022-2023. Q, questionnaire; SP, standardized patient; CL, checklist; GRS, global rating scale; IAT, Implicit Association Test.

Results

A total of 75 physicians participated in the study. Due to technical difficulties, we are missing IAT scores for 2 physicians and one ACS CR score and two acute pain simulation scores. Table 2 lists demographic data for the full sample and Table 3 lists the physicians mean IAT and RMC-IAT scores reported by physician race and the race of the SP evaluated. Overall, Race IAT scores are more pro-White than RMC-IAT scores suggesting affective bias is stronger in this cohort than cognitive bias. Physicians who identified as White or other Race IAT scores in the moderate and strong range, respectively, while Asian and Black physicians’ scores were in the neutral range. Black SPs were seen by physicians with a mean RMC-IAT score at the top of the neutral range; White SPs were seen by physician with a RMC-IAT score at the lower end of slightly pro-White.

Overall, there were no statistically significant bivariate correlations between Race IAT or RMC-IAT and CR scores or discussion of opioid prescribing scores by the race of the SP (Table 4). Physicians’ CR scores were moderately correlated with RMC-IAT (R=0.36, p=0.075, CI −0.04, 0.66) when seeing Black SP. In contrast, there was negligible correlation of RMC-IAT on decision-making scores when seeing White SP (R=0.1, p=0.537, CI −0.44, 0.24).

Discussion

We conducted the first, to our knowledge, high-fidelity simulation study to investigate the relationship between implicit bias and CR as assessed by scoring of physician written diagnostic and management plan accuracy. Our study, with its small sample size, did not yield statistically significant results regarding our main hypotheses that physicians’ implicit bias would be associated with lower quality CR, inclusive of diagnostic and treatment outcomes when seeing Black SPs as compared to White SPs. However, we had some suggested findings that should guide future work on implicit bias and its impact on health disparities. First, our study included a potentially more (ACS diagnostic dilemma) and potentially less (rib fracture requiring pain management) cognitively demanding case. As previously seen in the literature, implicit bias did not appear to influence CR accuracy in the potentially less cognitively demanding case bias [38], [39], [40], [41]. For the potentially more cognitively demanding case, outcomes for the Race and RMC- IAT were distinct from each other. The Race IAT, which measures affective bias (prejudice) using timed associations between images of Black and White faces with valanced words such as joy and evil, had minimal to no correlation with CR. Our data suggested a possible moderate correlation between the RMC-IAT, which measures cognitive bias (stereotyping of Black patients as less medically cooperative than White patients), and physician CR accuracy scores when seeing Black SPs which was not evident when physicians were seeing White SPs. Suggesting that stereotyping (a cognitive implicit bias) rather than prejudice (an affective implicit bias) may be a factor in disparities in CR. If this finding can be confirmed it may have implications for developing interventions with a comprehensive approach to achieving equitable outcomes.

Our study highlights the importance of integrating cognitive stressors common to clinical environments into simulation-based research and the importance of adequate sample size to study the impact of cognitive implicit bias on CR. Similar to our study, Oliver and colleagues did find an association between RMC-IAT scores and the decision to refer a hypothetical patient for knee replacement by race, but Race IAT scores were not associated with this clinical recommendation [28]. CR is highly complex [1]; it is possible that physicians may develop cognitive biases which negatively influence their CR over the course of medical training. This is supported by findings in a study observing changes in medical students’ performance in simulated encounters from the first to second year of medical school [49]. In the first year, medical students performed equally well with Black and White SPs in all domains of history-taking and communication skills. In the second year, however, medical students performed equally well with all SPs in history gathering, but worse in every other domain tested with Black compared with White SPs [49]. These findings demonstrate a need to continue to investigate the impact of the curriculum and learning environment throughout medical education and training on learners’ and physicians’ stereotyping of patients and development of cognitive implicit bias that negatively impact their CR. Finally, our findings support evidence on the different impact between affective and cognitive implicit bias (i.e. prejudice and stereotyping) and add to the literature on CR and context specificity and cognitive load [1], [2], [3, 6], 50], 51].

Our study is the first to specifically investigate simulated encounters with SP race and physician racial implicit bias as contextual factors that may affect CR; our findings provide guidance for next steps in this important area of research. We have previously outlined the implications for simulation-based education and research in addressing the impact of implicit bias on clinical encounters. Our current findings expand on our prior recommendations by highlighting the importance of studying both affective and cognitive implicit bias (e.g. through Race- and RMC-IATs, respectively). CR is known to be impacted by multiple cognitive biases (e.g., anchoring, recall, and confirmation biases, among others) when caring for patients. Stereotyping may be an additional cognitive bias that further detracts from effective CR when seeing patients that hold minoritized identities – in our case Black SPs [3].

Conclusions

Racial implicit bias may impact clinical reasoning; this impact, however, is nuanced and contextual. This study suggested a potential impact of cognitive implicit bias (i.e. stereotyping) on CR accuracy when physicians are seeing Black SPs, but not when seeing White SPs – a finding that might inform future study design and power calculations in research aiming to understand the complex phenomenon of CR. Our findings support the need for a comprehensive, multi-faceted approach to future simulation-based research seeking to address the impact of racial implicit bias on clinical encounters, with the ultimate goal of contributing to health equity.