Predictive Modeling to Support Student Success on Step 2 Clinical Knowledge: A Multi-Cohort Model

The shift of USMLE Step 1 to pass/fail redirected residency programs' attention to Step 2 Clinical Knowledge (CK), creating demand for accurate early‑warning tools. Develop and externally validate a concise model that predicts Step 2 CK performance midway through clerkships and guides targeted support. Data from 422 students across three cohorts were used to build a multiple linear regression (MLR) model. Predictors included National Board of Medical Examiners (NBME) subject exam scores. Predicted scores were converted to pass probabilities (≥ 214) and mapped to low‑risk (≥ 95%) and elevated‑risk (< 95%) strata that triggered predefined remediation protocols. The multivariable model explained 67% of Step 2 CK score variance in-sample (R² = 0.673, p < 0.01) and retained strong accuracy out-of-sample (predicted-vs-actual R² = 0.612). Medicine (r = 0.703) and Pediatrics (r = 0.693) NBME subject exams were the most powerful single predictors, although all six clerkship exams contributed significantly (r = 0.612-0.659). Residuals were normally distributed (M = 0.02 ± 7.96), indicating unbiased estimates. Risk stratification flagged 9% of students for intensified remediation; every low-risk student passed, and high-risk students who completed tutoring and structured study reached the 214 benchmarks with only one single failure, yielding a 0.9% failure rate. The model delivered actionable, early guidance for targeted support. Routinely collected NBME shelf scores can power a transparent, six‑variable model that reliably forecasts Step 2 CK outcomes months in advance. Embedding this tool within an equity‑minded remediation workflow conserves advising resources and markedly reduces failure risk.