Machine Learning-Driven Pharmacovigilance Models for Drug-Induced Liver Injury (DILI) Risk Prediction and Personalized Therapeutic Dose Optimization in Chronic Liver Disease Patients

• Ada John

  ladoke Akintola university of technology

  Author

Drug-Induced Liver Injury (DILI) remains a leading cause of clinical trial attrition, post-marketing drug withdrawals, and significant patient morbidity, particularly among the growing population of chronic liver disease (CLD) patients who face elevated susceptibility to hepatotoxic insults. Traditional preclinical models fail to detect approximately 40–45% of hepatotoxicity cases that emerge in clinical trials, while conventional pharmacovigilance systems predominantly rely on retrospective reporting, creating a critical gap in proactive DILI risk identification and personalized dose management . This study presents a comprehensive machine learning-driven framework that integrates multi-modal patient data—including demographic profiles, clinical biomarkers (ALT, AST, total bilirubin, ALP), genetic polymorphisms, drug physicochemical properties, and real-world pharmacovigilance data from FAERS—to predict DILI risk and optimize therapeutic dosing in CLD patients. A hybrid ensemble model combining Gradient Boosting Machine, Random Forest, and deep neural networks achieved 89.4% accuracy (AUC: 0.94) on external validation, significantly outperforming traditional logistic regression (72.1% accuracy). Key predictors identified include baseline liver enzyme elevations, CYP2C9 and CYP2D6 polymorphisms, drug lipophilicity (logP > 3), and concomitant medication burden. The model generated dose recommendations that reduced predicted hepatotoxic events by 34.2% while maintaining therapeutic efficacy in simulation studies. 

• pdf

Copyright (c) 2026 Ada John (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.