Abstract
BACKGROUND: For most physicians trained in internal medicine (IM), residency represents the last opportunity to compare and contrast generalist versus subspecialty careers. Decisions about subspecialty practice impact physician supply and distribution in the United States. This study tested the association between randomly assigned intern year rotation duration and career choice and assessed the importance of resident characteristics versus residency experiences in predictive modeling.
METHODS: This retrospective observational study included all categorical residents graduating from a university-affiliated IM program between 2017 and 2023 (N = 338). The association between randomly assigned intern year rotation duration and career choice was evaluated with univariable and multivariable log-link Poisson regression models. "Career choice" was defined as the field in which residents were practicing one year after residency. Predictive models were built using random forest modeling to compare the relative importance of resident characteristics and residency experiences.
RESULTS: Pre-residency career interests and ultimate career choice were highly concordant (P < 0.001, Cramer's V 0.50 (95% CI: 0.41-0.51)), particularly among residents pursuing cardiology, gastroenterology, and hematology/oncology. Rotation duration and career choice were not consistently associated; in multivariable analyses, weeks on service in the intensive care unit were associated with a career in pulmonary/critical care (RR 1.40, 95% CI 1.23-1.56), but there was no association between rotation duration and career in hematology/oncology, cardiology, gastroenterology, or general medicine. Internally cross-validated predictive models revealed that pre-residency demographics and interests were highly discriminative of career choice (AUC 0.824). The addition of residency experiences did not improve model performance (AUC 0.829).
CONCLUSIONS: Pre-residency career interests are a key determinant of career choice and hold greater weight in predictive models than intern year clinical exposure.