Predicting Plastic Surgery Residency Match Tier From Publication Metrics: A Machine Learning Analysis

In recent years, matched plastic surgery applicants have shown a marked increase in research output, yet it remains unclear whether these academic metrics can predict the tier of the residency program into which an applicant matches. This study utilized machine learning (ML) to evaluate whether a comprehensive set of preresidency publication measures could jointly predict the matched program tier. In this retrospective, cross-sectional analysis, data were aggregated from scientific databases including PubMed and Scopus for 914 applicants who matched into ACGME-accredited integrated plastic surgery residency positions during the 2019 to 2024 National Resident Matching Program cycles. Five ML classifiers (logistic regression, decision tree, k-nearest neighbors, Gaussian naïve Bayes, and multilayer perceptron) were trained to predict program tiers under reputation-based and research-based Doximity ranking schemes. Variables included publication volume (total, first-author, and PRS-journal-specific), subspecialty categorization, and institutional variables. Performance was assessed using accuracy and F1-score (a balanced measure of precision and recall) in binary and tertiary classification settings. Performance was above chance but limited across all schemes. For binary reputation tasks, the best model achieved 61.0% accuracy and 71.4% F1-score. Under the research scheme, performance peaked at 57.5% accuracy and 66.0% F1. For tertiary classification, the best reputation result was 45% accuracy and 44% F1, and the best research result was 43% accuracy and 43% F1. No algorithm consistently outperformed the others. Preresidency publication metrics alone weakly predict the matched residency program tier, and factors beyond research output likely play a more decisive role in residency placement.