Modelling survival in hepatocellular carcinoma.

Area of Science:

• Oncology
• Biostatistics
• Survival Analysis

Background:

• Unresectable hepatocellular carcinoma (HCC) poses significant challenges for long-term survival prediction.
• Accurate extrapolation of overall survival is crucial for cost-effectiveness modeling in oncology.
• The Weibull distribution is frequently employed but its appropriateness in HCC requires evaluation.

Purpose of the Study:

• To identify the optimal statistical distribution for predicting long-term survival in unresectable HCC.
• To assess the suitability of the Weibull distribution as a standard survival model in this context.
• To compare the fit of various distributions using diverse data sources.

Main Methods:

• Analysis of patient-level data from two randomized controlled trials and long-term follow-up data.
• Creation of a time-to-event dataset from published Kaplan-Meier curves via systematic literature review.
• Fitting and statistical comparison of survival distributions (e.g., Weibull, lognormal, loglogistic) using log-likelihood, AIC, and BIC criteria.

Main Results:

• The Weibull distribution demonstrated a poor fit for overall survival in unresectable HCC across different data sources.
• The lognormal distribution emerged as the best-fitting model.
• The loglogistic distribution was identified as the second-best fitting model, with consistent results across datasets.

Conclusions:

• The commonly used Weibull model is not appropriate for extrapolating overall survival in unresectable HCC.
• The lognormal distribution is recommended as the most suitable model for survival extrapolation in this patient population.
• Findings are robust, supported by multiple data sources including clinical trials and real-world data.