Machine Learning Model to Predict Postmastectomy Breast Reconstruction Complications

Area of Science:

• Plastic Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Postmastectomy breast reconstruction (PMBR) significantly enhances patient quality of life.
• Patients often lack precise, individualized information regarding complication risks associated with PMBR.
• Machine learning (ML) offers a powerful approach to analyze complex clinical data for personalized risk prediction.

Purpose of the Study:

• To develop and validate ML models for predicting major complications following PMBR.
• To utilize both structured electronic health record data and unstructured clinical notes for model training.
• To enhance shared decision-making by providing individualized complication risk estimates.

Main Methods:

• A retrospective prognostic study involving 411 female patients undergoing PMBR at two US academic centers (2012-2022).
• Extreme gradient boosting (XGBoost) and random forest models were trained on 80% of data and tested on 20%.
• Major complications were defined as unplanned reoperations or rehospitalizations within one year; model performance was assessed using AUROC and AUPRC.

Main Results:

• The XGBoost model demonstrated superior performance (AUROC 0.83, AUPRC 0.62) compared to the random forest model (AUROC 0.74, AUPRC 0.56).
• Key predictors for major complications included smoking, adjuvant radiotherapy, BMI, age, and diabetes.
• The overall major complication rate was 25.8%, with consistent model performance across different reconstruction types.

Conclusions:

• An internally validated ML model effectively predicts 1-year major complications after PMBR using diverse clinical data.
• These models are crucial for personalized risk assessment and informing patient-clinician decision-making.
• The study provides a foundation for developing prospective, externally validated decision-support tools for PMBR.