Machine learning predicts post-transplant muscle loss in hepatocellular carcinoma patients without sarcopenia.

Abstract

Objective: Developing a machine learning model to predict post-transplant muscle loss in hepatocellular carcinoma patients.

Background: Liver transplantation is an effective treatment for selected HCC patients. However, severe muscle loss after liver transplantation is significantly associated with increased risk of mortality and recurrence. However, effective predictive methods remain inadequate.

Methods: This study collected data from hepatocellular carcinoma patients who underwent liver transplantation over the past 2015 to 2020 at two hospitals. Propensity score matching and Cox regression analysis were conducted to establish muscle loss as an independent risk factor for recurrence. To construct the optimal predictive model for post-transplant muscle loss, we compared 50 machine learning models and use Recursive Feature Elimination to identify the most relative features.

Results: Data from a total of 248 patients were collected. Kaplan-Meier analysis revealed a significant difference in prognosis between patients with and without sarcopenia before surgery. For patients without sarcopenia, postoperative muscle loss was identified as an independent risk factor for recurrence (HR = 2.38, P = 0.005). The best model was identified as the Imbalanced Random Forest, achieving an AUC of 0.832 on the non-sarcopenia cohort.

Conclusions: A highly efficient model based on machine learning was developed to predict postoperative muscle loss in hepatocellular carcinoma patients undergoing liver transplantation, providing a valuable reference for the early detection of adverse events following the procedure.