Implicit Bias in Diagnostic Testing for Pulmonary Embolism: The PERCEIVE-PE Study

Although evidence-based algorithms guide testing once pulmonary embolism (PE) is suspected, far less is known about how clinicians arrive at the decision to suspect PE in the first place. This gap is critical since PE remains frequently misdiagnosed. We hypothesized that PE testing is influenced by implicit biases regarding patient sex and race.

We designed an electronic survey that described the history and examination of 7 fictional adult ED patients with chest symptoms. Using a 2x2x2 factorial design, participants randomly received 1 of 8 versions of each case that differed by sex (female/male), race (Black/White), and a case-specific clinical factor. Each case had 2 outcomes. Question 1 (Q1) asked participants to rate their concern for 5 conditions (including PE) on a 0-to-100 visual analog scale (VAS). In Question 2 (Q2), participants chose tests they would order in each case; we used a binary outcome of whether PE tests (e.g., D-dimer or computed tomography) were ordered or not. Pilot testing by 16 ED physicians showed good test-retest validity (Pearson’s r = 0.84) and criterion validity (higher VAS score in cases with higher Wells scores). We used linear mixed-effects regression (Q1) and logistic mixed-effects regression (Q2) to examine each factor, with fixed effects for case (sex; race; case order) and participant traits (sex; race; age; clinical role). Crossed random intercepts for participants and cases were used to account for repeated measures.

We invited 2011 ED physicians and advanced practice providers via in-person appeals at a conference and email invitations to staff at 37 EDs in 12 health systems. The study had a 43% response rate (n=863) and 78% completion rate (n=673), including participants from 27 US states and Canada. Cases with female patients had a higher reported PE risk (β = 5, 95% confidence interval [CI] [3.2, 6.8], p < 0.001) and higher odds of testing (odds ratio [OR] = 1.6, 95% CI [1.4, 2], p < 0.001) compared with males. Patient race was not significantly associated with perceived PE risk (β = -0.7, 95% CI [-2.5, 1.1], p=0.43) or odds of PE testing (OR = 0.9, 95% CI [0.8, 1.1], p = 0.48).

Implicit sex bias, but not racial bias, appears to influence PE diagnostic evaluation. Since females have a lower incidence of PE than males, this heuristic could contribute to low-value testing (via over-testing) in females and diagnostic failures (via under-testing) in males.