Why Residents Quit

National Rates of and Reasons for Attrition Among Emergency Medicine Physicians in Training

Dave W. Lu, MD, MSCI, MBE; Nicholas D. Hartman, MD, MPH; Jeffrey Druck, MD; Jennifer Mitzman, MD; Tania D. Strout, PhD, RN, MS


Western J Emerg Med. 2019;20(2):351-356. 

In This Article


Study Design and Setting

This was a retrospective observational study using de-identified complete national data from the annual American Medical Association (AMA) National Graduate Medical Education (GME) Census.

Study Population

The study population included all categorical residents without prior United States graduate medical education training who entered ACGME-accredited EM programs between academic years 2006–2007 and 2015–2016. The attrition group consisted of any resident at any level who left his or her program during a specific year.


We calculated the attrition rate for any year using aggregated national data as the percentage of all residents who left their programs. To ensure anonymity the de-identified dataset included resident characteristics that were limited to gender, race/ethnicity, and medical school type (i.e., allopathic, osteopathic, and international). Per the census database, a primary status and reason for each resident departure was reported by the program director (PD). Attrition statuses included the following: 1) dismissal; 2) transfer to another EM program; 3) transfer to a non-EM program; 4) transfer unknown; and 5) withdrawn. Transfer "unknown" means whether it was unknown by the PD at the time of the report to the AMA National GME Census to what specialty the trainee transferred. Reasons for attrition included the following: 1) change in career plans; 2) health/family reasons; 3) military obligations; and 4) other/unknown.

There are two main ways to view resident attrition: There is attrition from the training program the resident initially enrolled in, and there is attrition from the specialty altogether. For several reasons, we chose the most inclusive definition by counting all attrition statuses, including attrition from one EM program to go to another EM program as well as attrition from the specialty altogether. First, we wanted to be consistent with prior work in other specialties so as to be able to compare our results.[10] Second, attrition from a program or a specialty results in the same negative consequences of morale, workload, and scheduling for the program and its remaining residents. Third, residents who leave one EM program to go to another EM program may highlight the unique systemic challenges he or she faced in that particular program, rather than challenges due to a poor specialty choice, which one presumes would result in attrition to another specialty. Since we were unable to parse out specific details of why each resident left his or her program based on the attrition status and reason reported by PDs, we aimed to provide the most inclusive definition of attrition to gain the most complete picture.


Our primary outcome was the annual national rate of EM resident attrition. Secondary outcomes included the main status and reason for attrition as well as resident characteristics associated with attrition.

Data Analysis

We analyzed data using SPSS for Windows v24.0 statistical software (SPSS, Inc., Chicago, Illinois). To assess for the presence of differences in attrition as well as the status and reason for attrition based upon resident characteristics (i.e., gender, race/ethnicity, medical school type), we employed chi-square analyses followed by the Marascuilo procedure where appropriate for the data.[11] To ensure differences in attrition by gender were not due to potentially changing numbers of women choosing to specialize in EM over time, we evaluated changes in the proportion of female residents using simple linear regression, with the proportion of female residents serving as the outcome variable and calendar year serving as the predictor. Comparisons of independent proportions were made using the z-test. Data are presented as counts, proportions, and 95% confidence intervals (CI) around proportions. All p-values were two-tailed; we accepted p<0.05 as statistically significant. This study was reviewed and determined to be exempt by the Maine Medical Center Institutional Review Board.