Do Physicians' Implicit Views of African Americans Affect Clinical Decision Making?

M. Norman Oliver, MD, MA; Kristen M. Wells, MPH, PhD; Jennifer A. Joy-Gaba, PhD; Carlee Beth Hawkins, MA; Brian A. Nosek, PhD

Disclosures

J Am Board Fam Med. 2014;27(2):177-188. 

In This Article

Results

Variation in degrees of freedom for inferential tests is due to nonresponses for some items. The average age of the participants was 39.16 years (standard deviation [SD], 11.52 years), and the mean number of years in medical practice was 12.05 (SD, 10.26 years) (Table 1).

Analytical Approach

The main dependent variables were diagnosis of osteoarthritis and TKR decision. Both of these were measured on 5-point Likert scales and analyzed using linear regression. TKR decision also was measured as a dichotomous variable (recommend TKR or not) and analyzed using logistic regression. Most of the racial attitudes and stereotypes (both implicit and explicit) were measured as relative preferences for white versus black people, and these items were analyzed for significant difference from the 0 point (indicating no preference between black and white people) with 1-sample t tests (H0 = 0). Warmth and medical cooperativeness also were assessed separately for black and white people, and differences between these means were tested with paired-samples t tests. Patient race was a between-participants manipulation, so differences in attitudes and stereotypes, as well as beliefs about treatment and one's own biases, based on whether participants viewed a white patient or a black patient in the vignette, was tested with 2 samples t tests.

Diagnosis of OA and TKR Decision. On average, participants reported that it was "somewhat likely (60% to 80%)" to "very likely (>80%)" that the vignette character Mr. Jackson's knee pain was due to OA. In total, 207 doctors (44%) recommended TKR for Mr. Jackson. A similar question asked participants to characterize their recommendation for Mr. Jackson using a 5-point scale, from 1 (would definitely not recommend TKR) to 5 (would definitely recommend TKR), and the average response was the midpoint of the scale (unsure). The 5-point scale had a bimodal distribution and was highly correlated with the dichotomous measure (r = 0.75; P > .0001). We report analyses of the dichotomous items; substantive results and interpretations were the same using the continuous variable.

TKR Recommendation by Race and Exposure to Implicit Bias. Previous research has shown that completing an IAT can be an intervention for reducing explicit biases.[39] To test whether experiencing the IAT could also influence clinical decision making, we manipulated the order of the IATs and the TKR vignette and decision. We hypothesized that completing the IAT before making the TKR decision would reduce the effect of patient race on TKR recommendations. There was a main effect of the order of the IAT (χ2(df = 1, n = 470) = 5.33; P = .02): participants were more likely to recommend TKR when they completed the IAT before the decision (50%) versus after the decision (39%). However, the order did not interact with patient race to predict TKR recommendation (χ2[df = 1, n = 470] = 0.00; P = .96). Furthermore, the main effect of patient race on TKR recommendation was not statistically significant (P = .73), indicating that there was no racial bias for TKR recommendation regardless of IAT order. Doctor age and sex were added as covariates to the model in which patient race predicted TKR recommendation. Age significantly predicted TKR recommendation (χ2(df = 1, n = 456) = 5.81; P = .016): older doctors were less likely to recommend TKR. Sex did not significantly predict TKR recommendation (χ2(df = 1, n = 456) = 0.55; P = .458), nor did the addition of the covariates change the patient race effect (χ2(df = 1, n = 456) = 0.32; P = .569). Patient race does not predict TKR decisions, even when restricting the sample to white doctors only (n = 371) (χ2(df = 1, n = 324) = 0.02; P = .888).

Implicit Racial Attitudes and Stereotypes. A 1-sample t test revealed that participants implicitly preferred white people to black people (mean, 0.43; SD, 0.34; t(71) = 10.64; P < .0001; d = 1.26). This IAT was administered only to the directly recruited sample. Similarly, participants showed a stronger implicit association with medical cooperativeness and white people than black people (mean, 0.23; SD, 0.41; t(425) = 11.76; P < .0001; d = 0.57). In both cases, as expected, IAT scores did not differ, despite whether they were administered before or after the vignette (P = 0.12 and 0.62, respectively) or whether the vignette presented a black or white patient (P = 0.91 and 0.74, respectively) (Table 2).

Next we tested whether implicit racial bias predicted TKR recommendation and whether this effect was moderated by the implicit bias education intervention. We also tested whether implicit cooperativeness stereotype predicted TKR recommendation. Our base model predicted TKR recommendation in a logistic regression from the main effects of patient race and order of the IAT and their interaction; then we added the main effect of the moderator along with all 2-way interactions and the 3-way interaction. Neither of the moderators produced significant main effects, nor did they moderate the effect of patient race on TKR recommendation (Table 3).

Self-reported Attitudes. Overall, a 1-sample t test revealed that participants explicitly preferred white people to black people (mean, 0.28; SD, 0.78; t(426) = 7.46; P < .0001; d = 0.36). Paired t tests revealed that participants reported significantly higher feelings of warmth toward white people (mean, 7.31; SD, 1.69) than black people (mean, 6.99; SD, 1.80; t(426) = 4.96; P < .001; d = 0.24).

In addition, in a 1-sample t test participants explicitly reported that white patients were more medically cooperative than black patients (P < .0001). Furthermore, when asked about white and black patients independently, paired t tests revealed significantly more agreement that white patients were more medically cooperative (mean, 6.55; SD, 1.38) than black patients (mean, 6.22; SD, 1.42; t(423) = 5.80; P < .0001; d = 0.28). Even so, participants reported providing similarly aggressive care for white patients and black patients (mean, −0.04; SD, 0.41); t(426) ≥ 1.89; P = .06; d = 0.06).

Finally, using a 1-sample t test, participants reported strong agreement that subconscious biases may influence their decision making (mean, 3.64; SD, 1.12; t(427) = 11.79; P < .0001). Participants agreed that if black patients received differential medical treatment it could be because of their own preferences (mean, 3.26; SD, 1.25; t(214) = 2.99; P = .003; d = 0.20). In contrast, participants neither agreed nor disagreed that white patients received differential medical treatment because of their own preferences (mean, 3.14; SD, 1.26; t(210) = 1.64; P = .10; d = 0.11).

Interestingly, a 2-sample t test revealed that participants who viewed a black patient in the vignette reported less agreement that biases could influence their decisions than participants who viewed a white patient (P = .02), suggesting that exposure to a black patient may have increased reactance to the possibility of biases influencing behavior. However, there were no other significant differences in any attitudes or beliefs about bias by patient race condition (all P ≥ .08). For a complete list of means, see Table 4.

Explicit attitudes were standardized (SD, 1; retaining the rational zero point of "no preference" or "no difference" between blacks and whites) and averaged to create an aggregate score, indicating favorability for whites compared with blacks. (The questions included were, "Which of the following best describes your view about patients' cooperativeness with medical advice about interventions such as TKR?," "Please rate how cooperative you feel white patients are on average with medical advice about interventions such as TKR," "Please rate how cooperative you feel black patients are on average with medical advice about interventions such as TKR," "I provide less/similar/more aggressive care for white patients than I do for black patients," "I strongly prefer whites to blacks," "Please rate how warm or cold you feel toward white people," and "Please rate how warm or cold you feel toward black people.") For those topics that had separate items for blacks and whites, a difference score was calculated before standardizing. An intervention-by-patient race analysis of variance predicting this aggregate explicit attitude measure revealed no effect of intervention (P = .48), patient race (P = .92), or their interaction (P = .55) predicting explicit attitudes.

Furthermore, we examined whether explicit attitudes predicted TKR recommendation, using logistic regression with patient race, the aggregated explicit attitudes, the intervention condition, and their interactions as the independent variables predicting TKR recommendation (Table 5). The results suggest that explicit attitudes did not predict the TKR recommendation as a main effect or interaction with the other predictors.

Beliefs About the Patient. For a complete list of participants' beliefs about the patient in the scenario, see Table 4. Participants answered questions about treating the particular patient whose photograph appeared in the survey questionnaire. The race of the patient was never mentioned explicitly. Participants reported that they would feel more comfortable working with the black patient (mean, 4.34; SD, 0.82) compared with the white patient (mean, 4.16; SD, 1.02; t(457) = 2.06; P = .04; d = 0.10). Results revealed no other significant differences based on the race of the patient (all P ≥ .07).

We next tested whether the beliefs about the patient were affected by his race and the intervention. Beliefs about the patient were averaged to create an aggregate score indicating favorable beliefs about the patient. An intervention-by-patient race analysis of variance revealed no effect of intervention (P = .70), patient race (P = .82), or their interaction (P = .11).

We also examined whether beliefs about the patient predicted TKR recommendation using logistic regression, with recommendation for TKR as the dependent variable and patient race, aggregated patient beliefs, and the intervention condition as independent variables (Table 6). The results revealed a significant 3-way interaction (χ2[df = 1, n = 451] = 4.22; P = .04), suggesting that in 3 of the conditions (those who received the vignette first and/or who viewed a white patient), physicians who reported more favorable beliefs about the patient were less likely to recommend TKR. In contrast, physicians who received the IAT first and viewed a black patient reported more favorable beliefs about the patient and were more likely to recommend TKR.

Beliefs About the Educational Value of Learning About Bias. See Appendix 2 for a full correlation matrix. Questions were centered so that a rating of zero indicated neither agreement nor disagreement, whereas positive values indicated agreement and negative values indicated disagreement. After completing the study, a 1-sample t test revealed that participants agreed that subconscious biases could influence their decisions regarding patient care (t(379) = 13.04; P < .0001; d = 0.67). Participants agreed that learning about subconscious biases could improve patient care (t(377) = 26.80; P < .0001; d = 1.38).

Two-sample t tests revealed that participants who viewed a white patient were significantly more likely to agree that subconscious bias influences their decisions (mean, 3.83; SD, 1.01) than participants who viewed a black patient (mean, 3.62; SD, 1.13; t(378) = 1.93; P = .05; d = 0.20). There were no other significant differences based on patient race (all P ≥ .06), suggesting that whether participants viewed a black or white patient did not influence their remaining beliefs about the study (Table 7).

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....