A retrospective investigation of surgical training assessments has revealed that statistical learning algorithms can predict intraoperative competence in general surgery residents with measurable accuracy, while exposing consistent algorithmic disparities across gender. The study, published in BMC Medical Education, evaluated multiple classification models designed to track resident performance in operating rooms. Although advanced computational architectures successfully predicted surgical competence ratings, the researchers found that every single modeling approach yielded lower demographic parity values for female trainees, classifying female residents as competent less frequently than male peers.
To examine whether predictive modeling could support equitable and data-informed entrustment decisions, the investigative team analyzed a large repository of surgical performance ratings. The study drew upon SIMPL OR evaluation data collected over nearly a decade, spanning from September 2015 through May 2024, to establish how different algorithmic approaches handle real-world clinical assessment data.
The analyzed dataset represented a substantial cross-section of surgical education, incorporating 123,894 individual operating room evaluations across general surgery residency programs. These comprehensive assessment records captured performance evaluations from 3,793 surgical trainees assessed by 2,755 supervising faculty members.
To evaluate predictive capabilities, the researchers trained and validated eight distinct statistical learning algorithms using a rigorous ten-fold group cross-validation framework. The tested cohort comprised logistic regression, linear discriminant analysis, random forest, k-nearest neighbors, gradient boosting machine, support vector machine, generalized linear mixed model, and generalized neural network mixed model architectures. Model performance was systematically measured using standard validation metrics, with the analytical design placing specific emphasis on both classification accuracy and fairness across trainee gender.
Across all eight computational approaches examined in the study, postgraduate training year emerged as the single most influential determinant governing the assessment of intraoperative competence. In comparative evaluations, the generalized neural network mixed model attained the highest predictive precision, registering an Area Under the Receiver Operating Characteristic Curve of 0.787 and an overall classification accuracy of 0.722. In addition to this top-performing neural architecture, both the gradient boosting machine and the k-nearest neighbors models demonstrated strong and balanced predictive performance across the evaluation dataset.
The investigation also established that conventional logistic regression performed comparably to more computationally intricate approaches in many analytical cases. However, fairness analyses revealed persistent discrepancies that spanned every technique evaluated. Demographic parity values were consistently lower for female trainees across all eight statistical learning models, demonstrating that female surgical residents were classified as competent less frequently than male residents.
The authors concluded that machine learning models, most notably generalized neural network mixed models, offer meaningful promise for forecasting intraoperative competence and guiding objective, data-informed entrustment decisions in general surgery residency. Nevertheless, the researchers emphasized that institutional model selection must not depend solely on raw predictive performance metrics like accuracy and area under the curve. Surgical educators and curriculum directors must carefully weigh algorithm interpretability, clinical feasibility, and specific educational intended use alongside fairness considerations before integrating automated predictive systems into surgical training environments.
Subject of Research: Machine learning models for predicting intraoperative competence and evaluation fairness in general surgery residency
Article Title: Predicting assessment of intraoperative competence in general surgery residents: evaluating accuracy and fairness of statistical learning methods
Article References: Sun, T., Kim, S. Y., Moore, G., & Smith, B. K. (2026). Predicting assessment of intraoperative competence in general surgery residents: evaluating accuracy and fairness of statistical learning methods. BMC Medical Education. https://doi.org/10.1186/s12909-026-10690-0
Image Credits: AI Generated
DOI: 10.1186/s12909-026-10690-0
Keywords: general surgery residents, intraoperative competence, statistical learning, SIMPL OR, generalized neural network mixed model, logistic regression, demographic parity, algorithmic fairness, postgraduate year, entrustment decisions, gradient boosting machine, k-nearest neighbors, random forest, linear discriminant analysis, support vector machine, group cross-validation, surgical education
News Source: Blake Davidson. (November 12, 2026). Neural Network Models Predict Surgical Competence but Reveal Persistent Gender Disparities. Scienmag.



