Personalized Survival Prediction for Hepatocellular Carcinoma Using Artificial Intelligence

Abstract

Hepatocellular carcinoma (HCC) presents a significant global health challenge. Researchers have explored machine artificial intelligence-based systems for treatment recommendation. However, these models only consider a limited range of treatment options and primarily replicate historical treatment decisions, potentially leading to suboptimal outcomes. The XGBoost-based Accelerated Failure Time (XGBAFT) model, a novel machine learning-based system, predicts survival times for nine HCC treatment options. Developed using data from 8,511 patients at two medical centers in Taiwan, the model integrates the XGBoost algorithm with the accelerated failure time framework. Its key strength lies in comparing predicted survival times and curves across treatment options, facilitating personalized decision-making. The model can also predict survival beyond a specified target survival time, with accuracy and reliability established using binary classification measures. The model achieved a concordance index of 0.831, outperforming the Cox proportional hazards model. Notably, treatment was identified as the most influential covariate, underscoring the importance of selecting appropriate and effective treatments tailored to patient characteristics for improving survival outcomes. In conclusion, the XGBAFT model demonstrates significant potential in predicting HCC outcomes and personalized survival predictions. Future research should focus on integrating advanced machine learning algorithms and improving model interpretability to support clinical decision-making and personalized care.