Artificial intelligence predicts recurrent autoimmune hepatitis after liver transplantation in a multicenter cohort study

Abstract

Autoimmune hepatitis (AIH) is an important indication for liver transplantation (LT), but recurrence affects over 30% of recipients, threatening long-term survival. Current strategies to prevent recurrence and progressive graft fibrosis remain suboptimal, with limited evidence to guide selection of immunosuppressive regimens. We aimed to develop a dynamic, individualized, artificial intelligence–powered model for post-transplant recurrent AIH. Methods: We conducted a multicenter, retrospective cohort study of 706 patients who underwent LT for AIH between January 1987 and June 2020 at 33 centers in North America, South America, Europe, and Asia. We trained 4 predictive machine learning models—Logistic Regression, Random Forest, XGBoost, and Gradient Boost—to predict recurrent AIH (rAIH) using 62 clinical and laboratory variables, including demographic, biochemical features, and immunosuppressive drugs up to 1-year post-transplant. Feature importance was assessed using SHapley Additive exPlanations (SHAP) to enable interpretability at both individual and population levels. Results: AIH recurred in 16.5% of patients after LT. SHAP analysis identified younger age at LT, higher necroinflammatory activity in the explanted liver, and elevated MELD score at LT as key predictors of rAIH in the overall population. Tacrolimus-based therapy was associated with a lower risk of recurrence, while cyclosporine use conferred a higher risk. The addition of long-term prednisone to a regimen of tacrolimus and mycophenolate mofetil did not provide additional protective effect against rAIH. Conclusions: Our AI-powered clinical decision model provides personalized prediction of post-transplant rAIH. While it offers insight into modifiable and non-modifiable predictors, prospective validation is required before informing immunosuppressive decisions.