AI Risk Prediction Tools for Autologous Breast Reconstruction

Area of Science:

• Plastic Surgery
• Surgical Oncology
• Biostatistics

Background:

• Autologous breast reconstruction offers a durable, natural option post-mastectomy.
• Complications such as flap loss, infection, and delayed healing remain significant risks.
• Predictive models are needed to identify patients at higher risk for adverse outcomes.

Purpose of the Study:

• To develop and compare traditional statistical and machine learning (ML) models for predicting 90-day postoperative complications following autologous breast reconstruction.
• To identify key predictors of complications including infection, hematoma, seroma, delayed wound healing, and flap compromise.

Main Methods:

• Retrospective analysis of 2,128 patients undergoing abdominal-based autologous breast reconstruction.
• Development of multivariable logistic regression and supervised ML models.
• Evaluation of model performance using Area Under the Curve (AUC) and identification of key predictors via SHAP analysis.

Main Results:

• A 22.3% overall complication rate within 90 days, with infection (10%) and hematoma (6%) being most common.
• ML models demonstrated superior predictive performance (AUCs 0.64-0.73) compared to logistic regression (AUCs 0.60-0.66).
• Higher BMI and neoadjuvant chemotherapy were associated with increased complication risks; age, BMI, and pre-reconstruction radiation were key ML predictors.

Conclusions:

• Both statistical and ML models can predict complications after autologous breast reconstruction.
• ML models show promise for enhancing individualized risk assessment and potentially personalizing treatment strategies.
• Further research is necessary to improve model accuracy and clinical applicability for patient care.