Predicting Breast Reconstruction Readmission, Reoperation, and Prolonged Length of Stay: A Machine Learning Approach.

Background: Predicting short-term postoperative complications after breast reconstruction is critical for improving patient outcomes and reducing costs. This study investigated the utility of machine learning (ML) algorithms to predict complications in breast reconstruction patients.

Methods: Data were collected from patients who underwent autologous, implant, and tissue expander-based reconstruction in the National Surgical Quality Improvement Program (NSQIP) database (2020-2022). Six ML models were trained to predict 30-day readmission, 30-day reoperation, and prolonged length of stay (LOS). Model performance was assessed using the area under the receiver operating characteristic curve, sensitivity, specificity, and Brier score. SHapley Additive exPlanations (SHAP) values ranked predictors influencing model outcomes.

Results: A total of 27 718 patients (5584 autologous; 8170 implant; 13 964 TE) were included. Top-performing models showed moderate to strong predictive performance across cohorts for all complications. AUCs ranged from 0.614 to 0.861. The highest AUCs were achieved for prolonged LOS in implants patients (AUC 0.861) and for 30-day readmission in the delayed autologous cohort (AUC 0.859). Key predictors of complications included operative time, BMI, age, reconstruction timing, and ASA class.

Conclusion: ML can predict short-term postoperative outcomes in breast reconstruction patients. With further model refinement and data quality optimization, these models may improve preoperative risk stratification and patient outcomes in breast reconstruction.