Exhibit Hall | Forum 1
Purpose: Blinding is one strategy to help reduce unconscious bias in the application review process which removes or limits identifying information from the application to be used in evaluation of a candidate. We explore an automated approach to blinding coupled with human review during the annual medical physics residency application cycle.
Methods: Our institutional process for blinding included automating the download process and renaming application PDF files. A Python script that utilizes the Selenium package and the Chrome web driver was developed to crawl the website, download applications into a folder and rename each file. To anonymize the files, Adobe Acrobat was used to perform Optical Character Recognition (OCR) on the files. Once OCR was completed, a Python library was created to parse the files for gendered language and identifying information. The software looped through applications to hide the identified words or phrases including applicant name. We retrospectively compared self-reported demographic data between blinded and non-blinded cohorts from two sequential years of review.
Results: While we do not observe large differences in the gender composition between methods, we observe differences in race/ethnicity demographics between all applicants and those admitted through the initial screening process (Phase I) when blinded. We observe approximately 10% fewer Asian applicants and approximately 10% higher White applicants admitted through Phase I in the blinded applicant cohort. There are statistically significant differences in the scores in two of the scoring categories between Asian and White candidates in our blinded cohort.
Conclusion: We show the feasibility of blinding applications for a medical physics residency program. We recommend for each training program to critically evaluate their criteria for selection and review potential sources of bias in the review process. We recommend that common application programs provide an option for blinding applications.
Not Applicable / None Entered.