Study Design and Data Sources
We conducted a retrospective cohort study of applicants to McMaster University’s Michael G. DeGroote School of Medicine from 2013 to 2018. McMaster’s Undergraduate Medical Program (initiated in 1969) is a three-year program that admits approximately 200 students each year. In 2016/17, 74.6% (4,974/6,672) of all applicants to an Ontario medical school applied to McMaster and 21.4% (206/964) of all admitted/registered medical students in Ontario medical school were for McMaster [7].
We obtained applicants' admission data for application years spanning 2013 to 2018, inclusive. Data available included application year and whether applicants were offered admission (vs. not offered/waitlisted) and, if offered, whether the offer was accepted, declined, or deferred. We excluded repeat applicants, assuming that applicants would have used the same Ontario medical school applicant identification number for subsequent applications. We classified any offer of admission as such, regardless if the offer was accepted, declined, or deferred. Also available were the applicants' demographic data, including age category, residency status (Ontario versus out-of-province), identified sex, and the postal code of their home address during high school. Admission testing criteria data included applicants' undergraduate university Grade Point Average (GPA), Medical College Admissions Test (MCAT) Critical Analysis and Reasoning (CARS; formerly Verbal Reasoning) score, Multi-Mini Interview (MMI) mean score, and the CASPer z-score. MCAT scores range from 118 to 132 with a median score of all test takers of approximately 125 for each section. As McMaster University only uses the CARS score, we only had access to this data point. CASPer z-scores were utilized as these are typically cited in the literature rather than absolute scores [8,9]. GPA is a standardized score out of 4.0, reflecting undergraduate academic success. McMaster uses a pre-admission formula of 33% for each of the MCAT, GPA and CASPer scores and a post-interview score of 70% for MMI and 15% for GPA and MCAT respectively.
We incorporated a proxy measure of applicants' neighborhood income by including the 2015 median total household income for each applicants' home postal code during high school. Specifically, using the publicly available Postal Code Conversion File (PCCF), we matched each applicants' postal code to the most appropriate dissemination area (DA, a relatively stable geographic unit comprised of 400-700 individuals) from the 2016 Canadian Census in order to capture the 2015 median total neighborhood income. This method is a well-established mechanism of estimating neighborhood income [2,10,11,12]. However, using a DA as a comparative measure of median total household income for individual applicants may drastically underestimate or overestimate individuals within the DA. Therefore, the results need to be viewed within this context. An applicant’s postal code of their home address during high school was considered ideal given that it defines the family household neighborhood area during a critical scholastic and developmental life stage.
Statistical Analysis
Applicants' demographic data and admission testing scores were compared between those who were and were not offered admission. In addition, a sensitivity analysis was completed to rule out potential confounders by running the analysis without Ontario applicants included. This was done to protect from concentrated applicants in neighborhoods in the Greater Toronto Area with disproportionately higher incomes, as too ensure generalizability. Logistic regression was used to characterize the relationship between an offer of admission and applicants' median neighborhood income (in $10,000 increments). The income variable was transformed into different variables representing quartiles of income for comparison. Age, sex, application year, and residency (Ontario vs out-of-province) were included in the analysis to control for potential confounding and modifying effects that could also influence an offer of admission.
A second logistic regression was performed, first correcting for GPA, MCAT, Casper and then MMI, GPA and MCAT, to investigate whether the results remained significant. These were modelled to reflect McMaster’s pre and post interview formula. These models only include year’s 2016 to 2018 inclusive as only the newer MCAT CARS scores were used, although sensitivity analysis using the older verbal reasoning (VR) scores yielded similar results. However, because McMaster’s formula for pre and post interview scores are determined solely by these factors, we hypothesize income would play little to no role, as there would be no discretion for income in this model.
We then used unadjusted linear regressions for age, gender, residency status and year applied, to characterize the relationship between applicants' median neighborhood income (in $10,000 increments) and admission testing scores to examine factors that might mediate admission. Assumptions of normality for these analyses were tested by checking the Q-Q and normal probability plots. GPA exhibited a non-normal distribution; therefore, the linear regression was fit using a robust covariance estimator. All data analysis was completed using the Statistical Package for the Social Sciences (SPSS) software, version 23 for Mac 9 (IBM, Armonk, NY).
Ethics Approval
Given the use of secondary de-identified data, strict information security, and institutional sponsorship/oversight, this project was granted a waiver of full review by the Hamilton Integrated Research Ethics Board on the basis of low risk / quality improvement.