Multivariate Modeling of Student Performance on NBME Subject Exams.

Area of Science:

• Medical Education
• Biostatistics
• Health Professions Education

Background:

• Currently, tools to predict medical student performance on clinical subject exams are limited.
• National Board of Medical Examiners (NBME) self-assessments offer a potential data source for predictive modeling.

Purpose of the Study:

• To develop and evaluate statistical models correlating NBME self-assessment performance with student success on clinical subject exams.
• To assess the predictive value of these models across different medical disciplines.

Main Methods:

• Multivariate regression models were created using NBME self-assessment data from medical students.
• Models controlled for factors including self-assessment timing, USMLE Step 1 scores, and academic quarter.
• Student performance on six clinical subject exams (Medicine, Surgery, Family Medicine, OB-GYN, Pediatrics, Psychiatry) was the outcome variable.

Main Results:

• Linear regression requirements were met, with statistically significant models (p<0.001) across all subjects.
• Model predictive power (R-squared) ranged from 0.1799 to 0.4915, indicating weak to moderate correlation.
• The strength of prediction varied considerably depending on the specific clinical subject and performance metric (percent correct or percentile).

Conclusions:

• Statistical models using NBME self-assessments demonstrate weak to moderate predictive value for student performance on clinical subject exams.
• The models' accuracy is highly dependent on the specific medical discipline being assessed.
• Future research should focus on refining models with additional variables and evaluating their impact on reducing student failure rates.