engleski

Use of machine learning models for identification of predictors of survival in patients undergoing liver transplantation for hepatocellular carcinoma

Background: Hepatocellular carcinoma (HCC) is one of the leading indications for liver transplantation, however, selection criteria remain controversial. We aimed to identify survival factors and predictors for tumor recurrence using machine learning methods. We also compared a machine learning model to Cox regression model. Methods: 32 donor and recipient general and tumor specific parameters were analyzed from 170 patients who underwent liver transplantation for HCC between March 2013 and December 2018 at the University Hospital Merkur. Survival rates were calculated using the Kaplan-Meier method, and multivariate analysis was performed using the Cox proportional hazards regression model. Data was also processed through machine learning Random Forest (RF) method, which included preprocessing, variable selection, Random Forest variable significance, resampling, training and cross- validation of the RF model. Accuracy and concordance index were used for evaluation metrics. Results: Two year recipient and graft survival was 78% and 75%, respectively. The best predictive accuracy of our RF model was 0.75 while the best concordance index was 0.80. RF analysis yielded several relevant predictors of survival: donor CRP, bilirubin and sodium levels, recipient MELD and recipient age. Most significant predictors of HCC recurrence were recipient AFP level and donor CRP and sodium levels. Some of the analyzed parameters were shown to be detrimental for survival both in Cox multivariate analysis and in the RF models. In contrast to the RF model, Cox analysis showed an association between donor age and recipient and graft survival, while donor BMI and donor male sex were identified as risk factors for HCC recurrence. Conclusions: The purpose of a machine learning model for prediction of post transplant HCC recurrence is to identify the patients that would benefit from liver transplantation. Further research including prospectively collected data and additional parameters is needed to confirm our results and improve the existing model.

machine learning, liver transplatation, hepatocellular carcinoma

nije evidentirano