Predictive model for abdominal liposuction volume in patients with obesity using machine learning in a longitudinal multi-center study in Korea

Sci Rep. 2024 Nov 30;14(1):29791. doi: 10.1038/s41598-024-79654-y.

Abstract

This study aimed to develop and validate a machine learning (ML)-based model for predicting liposuction volumes in patients with obesity. This study used longitudinal cohort data from 2018 to 2023 from five nationwide centers affiliated with 365MC Liposuction Hospital, the largest liposuction hospitals in Korea. Fifteen variables related to patient profiles were integrated and applied to various ML algorithms, including random forest, support vector, XGBoost, decision tree, and AdaBoost regressors. Performance evaluation employed mean absolute error (MAE), root mean square error (RMSE), and R-squared (R2) score. Feature importance and RMSE importance analyses were performed to compare the influence of each feature on prediction performance. A total of 9,856 were included in the final analysis. The random forest regressor model best predicted the liposuction volume (MAE, 0.197, RMSE, 0.249, R2, 0.792). Body fat mass and waist circumference were the most important features of the random forest regressor model (feature importance 71.55 and 13.21, RMSE importance 0.201 and 0.221, respectively). Leveraging this model, a web-based application was developed to suggest ideal liposuction volumes. These findings could be used in clinical practice to enhance decision-making and tailor surgical interventions to individual patient needs, thereby improving overall surgical efficacy and patient satisfaction.

Keywords: Body fat distribution; Clinical decision support system; Liposuction; Machine learning; Obesity; Outcome assessment; Predictive value of tests; Surgical procedures.

Publication types

  • Multicenter Study

MeSH terms

  • Adult
  • Female
  • Humans
  • Lipectomy* / methods
  • Longitudinal Studies
  • Machine Learning*
  • Male
  • Middle Aged
  • Obesity* / surgery
  • Republic of Korea / epidemiology