Machine Learning Accurately Predicts Patient-Reported Outcomes 1 Year After Breast Reconstruction.

Abstract

Objective: To develop and evaluate machine learning algorithms for predicting patient-reported outcomes following breast reconstruction.

Summary of background data: Machine learning may inform PROs in breast reconstruction, possibly enhancing shared decision-making and tailoring patient care.

Methods: Data on patient characteristics, reconstructive technique, and BREAST-Q scores from women undergoing breast reconstruction at Memorial Sloan Kettering Cancer Center (MSKCC) between January 2010 and March 2024 was retrospectively collected. Five machine learning algorithms were developed and validated on this data to predict improved versus not improved BREAST-Q scores after reconstruction. Models were externally validated models using multicenter data from the Mastectomy Reconstruction Outcomes Consortium (MROC). Models were evaluated using the area under the receiver operator curve, sensitivity, specificity, and Brier score.

Results: A total of 4,776 patients (2,687 from MSKCC, 2,089 from MROC) were included in model development and validation. Machine learning algorithms demonstrated AUCs of 0.97 for physical wellbeing of the abdomen, 0.86 for satisfaction with breast, 0.79 for sexual wellbeing, 0.78 for physical wellbeing of the chest, and 0.74 for psychosocial wellbeing. Variables that contributed the most to model predictions across all domains were preoperative BREAST-Q scores, timing of radiation, BMI, age, and reconstructive technique.

Conclusions: Machine learning algorithms can accurately predict PROs before breast reconstruction. Ultimately, this data-driven approach may streamline shared decision-making and enhance patient-centered care.