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

Area of Science:

• Oncology
• Plastic Surgery
• Health Informatics

Background:

• Patient-reported outcomes (PROs) are crucial for evaluating breast reconstruction success.
• Machine learning (ML) offers potential to predict PROs and personalize patient care.
• Predictive models can enhance shared decision-making in breast reconstruction.

Purpose of the Study:

• To develop and validate ML algorithms for predicting PROs after breast reconstruction.
• To identify key factors influencing patient-reported outcomes.
• To assess the utility of ML in improving patient-centered care.

Main Methods:

• Retrospective collection of data from 4,776 patients undergoing breast reconstruction.
• Development and validation of five ML algorithms using BREAST-Q scores.
• External validation using multicenter data and evaluation via AUC, sensitivity, specificity, and Brier score.

Main Results:

• ML algorithms achieved high accuracy in predicting various PROs, with AUCs ranging from 0.74 to 0.97.
• Top predictors included preoperative BREAST-Q scores, radiation timing, BMI, age, and reconstructive technique.
• Models demonstrated strong performance in predicting physical wellbeing, satisfaction, and psychosocial wellbeing.

Conclusions:

• ML algorithms can reliably predict patient-reported outcomes before breast reconstruction.
• This data-driven approach supports enhanced shared decision-making.
• The findings pave the way for more personalized and patient-centered breast reconstruction care.