Portal de Periódicos da CAPES

Implicit Bias Among Physicians

2008; Springer Science+Business Media; Volume: 24; Issue: 1 Linguagem: Inglês

10.1007/s11606-008-0821-8

1525-1497

Neal V. Dawson, Hal R. Arkes,

Social and Intergroup Psychology

To the Editor: Green and colleagues1 claim to have provided evidence that unconscious (implicit) race bias among physicians is causally associated with fewer recommendations for appropriate thrombolytic treatment for African-American male patients who present with symptoms suggestive of acute coronary syndromes. We attempt to demonstrate why this claim is not substantiated. From a sample of 776 internal medicine and ED residents, Green and colleagues obtained data from 220 who were unaware of the purpose of the study or who were not otherwise excluded. They were shown a vignette about a 50-year-old man with a history of hypertension and smoking who presented to the ED with chest pain. With the vignette the face of a white or black man of approximately age 50 was randomly paired. The vignette was written to be consistent with the presentation of myocardial infarction, in particular containing an EKG reading that was “suggestive of anterior myocardial infarction.” The subject was told he/she did not have access to a cardiac catheterization lab and that the patient had no absolute contraindications to thrombolysis. Subjects were asked to a) assess the likelihood that the patient’s pain was due to CAD using a five-point scale where 1 = very unlikely ( 80%); b) state “yes” or “no” to “would you recommend thrombolysis” for this patient; c) state the strength of that recommendation on a five-point scale from 1 = would definitely recommend to 5 = would definitely NOT recommend; and d) give their opinion about the effectiveness of thrombolysis for acute MI on a five-point scale from 1 = very ineffective to 5 = very effective. Subjects then completed three implicit association tests (IATs) to “measure bias that may not be consciously recognized.” The “IAT measures the time it takes subjects to match members of social groups (e.g., blacks and whites) to particular attributes (e.g., good, bad, cooperative, stubborn).” A difference in reaction times to associate good or bad concepts with black or white faces is the measure of “implicit bias.” We obtained opinions from content experts to develop plausibility arguments and consulted published literature to critique the methods and results of Green et al. Nine attending-level general internists plus two disparities researchers (one nephrologist, one pulmonary-critical care physician) reviewed the study scenario [A] without a patient face [B] and were blinded to the published study and its results. (Letters in brackets refer to the Appendix where details of the bases for the critiques are presented.) The internists provided opinions about the top 5 diseases in their differential diagnosis (DDx), estimated a probability for each disease in the DDx (adding to 100%), and stated their opinion as to whether thrombolysis would be beneficial (+), neutral (0), or harmful (-) (90 mm visual analog scale) for each of the diseases in their DDx (see Table 1). Table 1 Physicians’ Estimated Probability of Alternative Diagnoses and the Judgment as to Whether Thrombolysis would have a Beneficial, Neutral, or Harmful Effect if the Alternative Diagnosis were Correct Five major areas of concern were identified: 1) the design of the judgment task response set provided no opportunity for subjects (resident physicians) to list competing diagnoses. No differential diagnoses were solicited, nor was there a recording of the risks/benefits of thrombolytic therapy for the competing diagnoses, e.g., pericarditis or dissecting aortic aneurysm [C]; 2) main study results were based on an interpretation of cross-sectional data as if the data were longitudinal (see Fig. 1) [D]; 3) although randomization was performed by allocating a white or black face randomly with each scenario, a non-randomized variable (IAT score) was interpreted as if it had been the unit of randomization [E]; 4) Green et al. conflated measurement issues with interpretation; the progression from the Introduction through the Methods to the Results section of the terms “increasing time for association,” “racial preference,” “racial bias,” to “pro-white/pro-black scores” does not represent a sequence of synonyms [F]; and 5) no serious discussion of alternative hypotheses occurred [G]. Figure 1 Relationship between physician race preference Implicit Association Test (IAT) score and thrombolysis decision as a function of patient’s race. This figure is based upon Figure 3d in Green et al. Higher numbers on the y-axis denote greater propensity ... The figure contains the same data as Green et al.’s Figure 3d, which purports to summarize the prejudicial impact of the “pro-white bias” on thrombolytic treatment of patients. The basis for the conclusion in Green et al. is the significant “decrease” (p < .05) in the thrombolytic treatment of black patients and the “increase” (p < .11) in the thrombolytic treatment of white patients as “implicit bias” increases. The resulting interaction was significant (p = .009). The figure depicts an extremely unusual state of affairs in which those with the lowest levels of bias as assessed by the IAT treat the races differently, and the most biased physicians treat the races nearly equivalently. This pattern would seem to be contrary to what most observers would consider to be a manifestation of bias. We offer an interpretation of the data that differs from that offered by Green et al. First, we are reluctant to deem as implicitly biased those persons on the right-hand side of the figure who treat the races equivalently. Our unwillingness to rely on the measure of bias utilized by Green et al. is also motivated by the fact that those physicians who are supposedly least biased are the ones who treat patients differently as a function of patient race. We respectfully suggest that this general pattern is completely contrary to the way in which “bias” is generally construed. Concern 1 (no competing diagnoses) is a fatal flaw as it does not allow the investigators to have any criterion from which to declare the treatment choice to be appropriate or inappropriate (the “delta scores” are uninterpretable). African-Americans are more likely than whites to present with symptoms that strongly mimic coronary disease even in the absence of significant coronary obstruction on angiography2. Our sample of physicians indicated that thrombolysis would be harmful if pericarditis or aortic dissection were the actual disease entity. We suggest that this flaw and other methodological shortcomings (see Appendix) nullify the conclusion that “unconscious (implicit) race bias among physicians” predicts the inappropriate under-utilization of thrombolytic therapy among African-American male patients (and thus does not support the predictive validity of “implicit bias”); nor is the claim “that physicians’ unconscious biases may contribute to racial/ethnic disparities in use of medical procedures such as thrombolysis for myocardial infarction” supported.