SP02. Predicting Reoperation in DIEP Flap Surgery With Machine Learning: A Random Forest Approach

PURPOSE: Deep Inferior Epigastric Perforator (DIEP) flap surgery is a complex procedure in breast reconstruction with potential postoperative complications. Accurate prediction of these complications can enhance patient outcomes and surgical planning. This study aims to develop and validate a machine learning model to predict postoperative complications in DIEP flap surgery patients. METHODS: Data from 13,326 DIEP flap surgery cases in the National Surgical Quality Improvement Program (NSQIP) from 2016-2022 was analyzed. The dataset included 276 features covering demographic information, preoperative conditions, intraoperative variables, and postoperative outcomes. After preprocessing, including handling missing values, categorical encoding, and feature normalization, a Random Forest Classifier was trained to predict the likelihood of reoperation, a key complication indicator. RESULTS: The model, evaluated using an 80-20 train-test split, demonstrated exceptional predictive accuracy with 99.92% accuracy, 100% precision, 99.88% recall, and a 99.94% F1 score. The confusion matrix revealed 998 true negatives, 0 false positives, 2 false negatives, and 1,666 true positives. Feature importance analysis identified patient age, comorbidities, intraoperative blood loss, and operative time as key predictors. CONCLUSION: This study presents a highly accurate machine learning model for predicting postoperative complications in DIEP flap surgery. The Random Forest model, trained on comprehensive NSQIP data, provides a reliable tool for enhancing surgical outcomes through better preoperative risk assessment. This high level of accuracy suggests that machine learning can be valuable in preoperative risk stratification and personalized patient care. Implementing such predictive models in clinical practice can aid surgeons in identifying high-risk patients, optimizing surgical planning, and potentially reducing postoperative complications. Future work will focus on expanding the model to predict other complication types and integrating it into clinical decision support systems.