Self composition and linguistic bias in letters of recommendation for early-career oncologists: an opportunity to improve equity in funding opportunities and career advancement

Purpose: Financial support via grant mechanisms can be an essential component for the career advancement of physician scientists. One potential source of bias in the evaluation of grant applications is the utilization of letters of recommendation (LORs), which may contain linguistic biases that affect the perception of future success for a given applicant. Furthermore, LORs are often partially or fully written by applicants themselves, which may compromise their value as evaluation tools. In this survey-based observational study, we aim to systemically characterize the impact of self-composition and linguistic bias in LORs used to support applications of recent ASCO Young Investigator Award (YIA) recipients.

Methodology: This study is currently pending IRB approval at MD Anderson Cancer Center and thus surveys have not been distributed at the time of abstract submission. The target survey population will be recipients of the ASCO YIA from 2021-2023, a list of whom is publicly available. There have been 283 awardees during this period. Individuals will be contacted initially via email and a repeat email will be sent one week later to those who have not responded after the first request. A ten-dollar Amazon gift card will be sent to all responding individuals.

A Research Electronic Data Capture (REDCap) survey will be distributed to the target population. The instrument will collect self-reported demographic and background awardee information, including affiliated institution, advanced training degree (MD, DO, MD/PhD), specialty (medical, surgical, or radiation oncology), gender, race, ethnicity, and years since completion of medical school. The primary survey items will collect data regarding the self-composition of any component of the LOR submitted on behalf of the applicant. Additional information will be collected regarding the length of mentorship, gender, race, and institution of letter writers. We will request that individuals provide an electronic copy of a submitted LOR from their application, after which we will assess for linguistic biases using Linguistic Inquiry and Word Count (LIWC) software.

Results: At the time of abstract submission, the protocol is currently pending IRB approval and thus surveys have not been distributed at this time. Results will be updated and completed by the meeting date. We hypothesize that there will differences in the prevalence of self-composition of LORs according to applicant gender and length of mentorship from letter writer. Further, we hypothesize that there will be linguistic biases in LORs, with LORs for women and other minority populations being less likely to contain language that emphasizes standout applicant characteristics.

Conclusions: Self-composition and linguistic biases in LORs may contribute to inequity in the allocation of grant funding for early career oncologists. These data may be used to improve fairness and equity in allocation of funding to support research and career development of oncologists.

References: