Predicting Resection Weights of Reduction Mammaplasty: A Multi-Institutional Retrospective Analysis Using Machine...

Area of Science:

• Plastic Surgery
• Medical Informatics
• Biostatistics

Background:

• Previous single-institution study showed machine learning (ML) accurately predicted reduction mammaplasty (RM) resection weights using preoperative anthropometric variables.
• The Schnur Scale is a current standard for predicting resection weights, but its accuracy and generalizability may be limited.

Purpose of the Study:

• To evaluate ML and regression modeling for predicting RM resection weights in a diverse, multi-institutional patient population.
• To compare the accuracy of ML and regression models against the Schnur Scale for predicting individual and total breast resection weights.

Main Methods:

• A multi-institutional retrospective study included 635 patients undergoing RM for macromastia (2017-2022).
• Preoperative variables included body surface area (BSA), body mass index (BMI), sternal notch-to-nipple (SN-N), and nipple-to-inframammary fold (N-IMF).
• Seven ML and regression models were assessed for predicting resection weights, with mean absolute errors (MAE) reported.

Main Results:

• The study population had a mean age of 38.5 years, mean BMI of 32.8 kg/m², and mean BSA of 2.0 m².
• Preoperative BMI, SN-N, N-IMF, and race/ethnicity were significant predictors of resection weight.
• Six of seven models showed lower MAEs than the Schnur Scale; Elastic Net regression yielded the lowest MAEs for individual and total resection weights.

Conclusions:

• ML and regression models demonstrate superior accuracy in predicting RM resection weights compared to the Schnur Scale.
• These findings support ML and regression modeling as accurate and generalizable alternatives to the Schnur Scale in a multi-institutional setting.