Machine Learning Models for Disease-Free Survival Analysis after Liver Resection for Hepatocellular Carcinoma: A...

Area of Science:

• Hepatobiliary Surgery
• Oncology
• Machine Learning in Medicine

Background:

• Hepatocellular carcinoma (HCC) recurrence after liver resection (LR) remains high, necessitating accurate prognostication for patient selection in clinical trials.
• Predicting disease-free survival (DFS) post-LR is critical for optimizing adjuvant treatment strategies.

Purpose of the Study:

• To evaluate and compare the performance of machine learning (ML) models against traditional Cox regression for predicting DFS after LR in HCC patients.
• To identify the most effective ML model for enhancing postoperative prognostication and clinical trial patient stratification.

Main Methods:

• Analysis of 663 patients undergoing LR for HCC across three French centers.
• Comparison of three ML models (Random Survival Forest, Gradient Boosting Survival, Fast Survival Support Vector Machine) with Cox regression.
• Performance evaluation using C-index and time-dependent AUC, with external validation on a separate cohort.

Main Results:

• Random Survival Forest (RSF) demonstrated superior discrimination and predictive accuracy in both training and external validation cohorts.
• RSF significantly outperformed Cox regression (p < 0.05) for DFS prediction.
• ML models, especially RSF, showed strong generalizability and outperformed Cox regression in external validation.

Conclusions:

• RSF significantly improves DFS prediction accuracy compared to Cox regression for HCC patients undergoing LR.
• ML-based survival models, particularly RSF, offer enhanced individualized prognostication.
• These findings support the use of RSF for refining patient selection in clinical trials for adjuvant therapies post-LR for HCC.