Development and Evaluation of a Machine-Learning Model for Predicting Autoimmune Hepatitis Relapse After Treatment Withdrawal Based on Clinical and Laboratory Factors

Abstract

Background: Relapse after treatment withdrawal remains a major challenge in autoimmune hepatitis (AIH), particularly because conventional risk assessment may not adequately capture interactions among multiple clinical and laboratory factors. This study aimed to develop and evaluate a machine-learning framework for predicting AIH relapse after treatment withdrawal using routinely available clinical and biochemical variables.

Methods: The proposed framework incorporates demographic, clinical, biochemical, disease-related, and treatment-related variables associated with relapse risk. Machine-learning algorithms were considered to identify nonlinear relationships and interactions among candidate predictors. The framework includes data preprocessing, feature selection, model development, internal validation, discrimination, calibration, and assessment of clinical applicability.

Results: The proposed approach provides a structured method for integrating multiple predictors into individualized relapse-risk estimation. Particular emphasis was placed on reducing overfitting, maintaining interpretability, and evaluating both discrimination and calibration. The framework can also facilitate comparison of alternative machine-learning algorithms and identification of clinically relevant predictors.

Conclusion: Machine-learning methods may provide a systematic approach for individualized prediction of AIH relapse following treatment withdrawal. Before clinical implementation, the resulting model should undergo rigorous external validation in independent patient populations.