Abstract
BACKGROUND AND OBJECTIVES: In the early period after liver transplantation, patients are exposed to a high rate of complications and several scores are currently available to predict adverse postoperative outcomes. However, an ideal, universally accepted and validated score to predict adverse events in liver transplant recipients with hepatitis C is lacking. Therefore, we aimed to establish and validate a machine learning (ML) model to predict short-term outcomes of hepatitis C patients who underwent liver transplantation. MATERIALS AND METHODS: We conducted a retrospective observational two-center cohort study involving hepatitis C patients who underwent liver transplantation. Based on clinical and laboratory parameters, the dataset was used to train a deep-learning model for predicting short-term postoperative complications (within one month following liver transplantation). Adverse events prediction in the postoperative setting was the primary study outcome. RESULTS: A total of 90 liver transplant recipients with hepatitis C were enrolled in the present study, 80 patients in the training cohort and ten in the validation cohort, respectively. The age range of the participants was 12-68 years, 51 (56,7%) were male, and 39 (43.3%) were female. Throughout the 85 training epochs, the model achieved a very good performance, with the accuracy ranging between 99.76% and 100%. After testing the model on the validation set, the deep-learning classifier confirmed the performance in predicting postoperative complications, achieving an accuracy of 100% on unseen data. CONCLUSIONS: We successfully developed a ML model to predict postoperative complications following liver transplantation in hepatitis C patients. The model demonstrated an excellent performance for accurate adverse event prediction. Consequently, the present study constitutes the foundation for careful and non-invasive identification of high-risk patients who might benefit from a more intensive postoperative monitoring strategy.