Explainable AI modeling of postoperative surgical site infection risk in autologous breast reconstruction.

Abstract

Surgical site infections (SSIs) following autologous breast reconstruction, particularly deep inferior epigastric perforator (DIEP) flap procedures, pose a substantial clinical burden and is associated with increased risk of hospital readmission. Existing prediction models lack sufficient accuracy in identifying high-risk patients. We aimed to develop a machine learning model to predict 30-day SSI risk and identify key predictors using a national surgical quality database. A total of 13,312 DIEP flap reconstructions from the American College of Surgeons National Surgical Quality Improvement Program (NSQIP) between 2016 and 2022 were analyzed. An XGBoost classifier incorporating demographic, comorbidity, operative, and laboratory variables was trained and evaluated using accuracy, recall, precision, F1 score, and area under the curve (AUC). Model explainability was assessed using SHapley Additive exPlanations (SHAP) to identify key predictors and their directional influence. The overall SSI rate was 8.14%. XGBoost classifier achieved an accuracy of 74.6%, a recall of 74.6%, and an AUC of 0.63 for predicting 30-day SSI. Key predictors of SSI included year of surgery (5.3% in 2016 and 9.3% in 2022), elevated BMI, increased body weight, prolonged operative time, and longer hospital stay. This model demonstrates the feasibility and interpretability of explainable AI in identifying SSI-related risk patterns using large-scale national data. It provides a framework for retrospective SSI risk analysis and hypothesis generation, thereby establishing a robust foundation for future preoperative-focused and prospectively validated prediction models.