Artificial Intelligence in Post-Liver Transplantation: A Scoping Review of Comparative Model Performance

TL;DR

This review maps AI applications in post-liver transplant care, highlighting clinical predictions and methodological gaps.

Contribution

The study provides a comprehensive scoping review of AI in post-liver transplantation with a focus on model performance and gaps.

Findings

Most AI studies in post-LT care focus on clinical predictions like graft survival and rejection.

Only a few studies evaluated operational or system-level AI applications.

External validation and real-world implementation of AI models remain limited.

Abstract

Objective: To map and characterize artificial intelligence (AI) applications in post-liver transplantation (LT) care, summarize comparative performance where available, and identify methodological and translational gaps. Methods: We conducted a scoping review in accordance with PRISMA-ScR. A comprehensive search of electronic databases was performed from inception through 1 April 2025. We included primary studies evaluating AI applications in the post-LT period (model development, validation, or implementation). Comparative studies were defined as those reporting head-to-head evaluation of at least two algorithmic models for the same task with quantitative performance metrics. Single-model studies were retained for evidence mapping but analyzed separately. Reviews and the other non-primary literature were included for contextual mapping. Results: The search yielded 3088 records. After deduplication, 2408 were screened, 191 full texts were assessed, and 65 studies were included. Of these, 52 reported primary outcome data. Clinical prediction studies (n = 43) focused on graft survival, rejection, fibrosis, oncologic recurrence, mortality, and composite outcomes. Operational studies (n = 3) evaluated early warning or bedside decision-support systems, and system-level studies (n = 6) examined benchmarking, donor–recipient matching, explainability, fairness, and cross-domain modeling. Most studies were retrospective and single-center, with internal validation commonly reported and external validation uncommon. Conclusions: AI research in post-LT care is expanding, with a predominant focus on clinical prediction. However, limited external validation, heterogeneous methods, and scarce real-world implementation constrain clinical readiness. Standardized evaluation and prospective integration are needed to determine whether AI tools can support decision-making and improve post-transplant outcomes.

Figures