BACKGROUND: There is a paucity of literature when it comes to identifying predictors of in-state retention of graduate medical education (GME) graduates, such as the demographic and educational characteristics of these physicians. OBJECTIVE: The purpose was to use demographic and educational predictors to identify graduates from a single Michigan GME sponsoring institution, who are also likely to practice medicine in Michigan post-GME training. METHODS: We included all residents and fellows who graduated between 2000 and 2014 from 1 of 18 GME programs at a Michigan-based sponsoring institution. Predictor variables identified by logistic regression with cross-validation were used to create a scoring tool to determine the likelihood of a GME graduate to practice medicine in the same state post-GME training. RESULTS: A 6-variable model, which included 714 observations, was identified. The predictor variables were birth state, program type (primary care versus non-primary care), undergraduate degree location, medical school location, state in which GME training was completed, and marital status. The positive likelihood ratio (+LR) for the scoring tool was 5.31, while the negative likelihood ratio (-LR) was 0.46, with an accuracy of 74%. CONCLUSIONS: The +LR indicates that the scoring tool was useful in predicting whether graduates who trained in a Michigan-based GME sponsoring institution were likely to practice medicine in Michigan following training. Other institutions could use these techniques to identify key information that could help pinpoint matriculating residents/fellows likely to practice medicine within the state in which they completed their training.
BACKGROUND: There is a paucity of literature when it comes to identifying predictors of in-state retention of graduate medical education (GME) graduates, such as the demographic and educational characteristics of these physicians. OBJECTIVE: The purpose was to use demographic and educational predictors to identify graduates from a single Michigan GME sponsoring institution, who are also likely to practice medicine in Michigan post-GME training. METHODS: We included all residents and fellows who graduated between 2000 and 2014 from 1 of 18 GME programs at a Michigan-based sponsoring institution. Predictor variables identified by logistic regression with cross-validation were used to create a scoring tool to determine the likelihood of a GME graduate to practice medicine in the same state post-GME training. RESULTS: A 6-variable model, which included 714 observations, was identified. The predictor variables were birth state, program type (primary care versus non-primary care), undergraduate degree location, medical school location, state in which GME training was completed, and marital status. The positive likelihood ratio (+LR) for the scoring tool was 5.31, while the negative likelihood ratio (-LR) was 0.46, with an accuracy of 74%. CONCLUSIONS: The +LR indicates that the scoring tool was useful in predicting whether graduates who trained in a Michigan-based GME sponsoring institution were likely to practice medicine in Michigan following training. Other institutions could use these techniques to identify key information that could help pinpoint matriculating residents/fellows likely to practice medicine within the state in which they completed their training.
Authors: Ernest Blake Fagan; Claire Gibbons; Sean C Finnegan; Stephen Petterson; Lars E Peterson; Robert L Phillips; Andrew W Bazemore Journal: Fam Med Date: 2015-02 Impact factor: 1.756