Abstract
BACKGROUND: Studies across medical specialties have shown that scores on residency self-assessment examinations (SAEs) can predict performance on certifying board examinations. OBJECTIVE: This study explored the predictive abilities of different composite SAE scores in physical medicine and rehabilitation and determined an optimal cut-point to identify an "at-risk" performance group. METHODS: For our study, both predictive scores (SAE scores) and outcomes (board examination scores) are expressed in national percentile scores. We analyzed data in graduates of a physical medicine and rehabilitation residency program between 2008 and 2014. We compared mean, median, lowest, highest, and most recent score among up to 3 SAE scores with respect to their associations with the outcome via linear and logistic regression. We computed regression/correlation coefficient, P value, R (2), area under the curve, sensitivity, specificity, and predictive values. Identification of optimal cut-point was guided by accuracy, discrimination, and model-fit statistics. RESULTS: Predictor and outcome data were available for 88 of 99 residents. In regression models, all SAE predictors showed significant associations (P ≤ .001) and the mean score performed best (r = 0.55). A 1-point increase in mean SAE was associated with a 1.88 score increase in board score and a 16% decrease in odds of failure. The rule of mean SAE score below 47 yielded the highest accuracy, highest discrimination, and best model fit. CONCLUSIONS: Mean SAE score may be used to predict performance on the American Board of Physical Medicine and Rehabilitation-written examination. The optimal statistical cut-point to identify the at-risk group for failure appears to be around the 47th SAE national percentile.