Ethnic/Racial Bias in Medical School Performance Evaluation of General Surgery Residency Applicants.

Objective Differential use of communal terms (caring/unselfish traits) versus agentic terms (goal-oriented/leadership/achievement traits) may reveal bias and has been extensively reported in letters of recommendation for residency. We evaluated bias in medical student performance evaluations (MSPE) of general surgery residency applicants. Design This is a retrospective study evaluating ethnic/race bias, as measured by differential use of agentic and communal terms, in MSPEs of residency applicants. 50% of MSPEs were randomly selected. An ethnic bias calculator derived from an open-source online gender bias calculator was populated with a list of validated agentic and communal terms. Relative frequency of communal and agentic terms was used to estimate bias. Multivariable regression was used to assess the association between the terms and ethnicity/race. Participants US medical students applying for a categorical surgery residency position at a single academic institution for a single Match cycle. Results A total of 339 MSPEs were reviewed from 119 US medical schools. Genders were equally represented (women, 51.6%); most participants were white and Asian applicants (79.1%). Overall, MSPEs were more agency biased (65.2%) than communal biased (16.2%) or neutral (18.6%). MSPEs for Black and Hispanic/Latinx applicants were more likely to contain communal rather than agentic terms (adjusted OR: 3.02, 95% CI: 1.52-6.02) when compared to white and Asian applicants. This finding was independent of MSPE writer's gender or rank. Conclusions Surgery residency applicants self-identifying as Black and Hispanic/Latinx were more likely to be described using communal traits compared to white and Asian applicants, suggesting ethnic/racial bias. Such differences in language utilized in MSPEs may impact residency opportunities for applicants who are under-represented in medicine. Educational efforts aimed at MSPE writers may help to reduce bias.