Artificial Intelligence for the Prognostication and Management of Heart Transplant: A Scoping Review

Abstract

<h3>Purpose</h3> Current clinical guidelines related to the identification and management of heart transplant (HT) outcomes are based on small studies or expert consensus. With a growing interest in artificial intelligence (AI) to optimize prognostication and management of patients post-HT, our scoping review aims to summarize the current state of AI in the setting of HT, and its potential impact on clinical practice. <h3>Methods</h3> We systematically searched MEDLINE, Embase, and Inspec for observational and interventional studies that applied AI methodologies to a HT population. Blinded abstract screening and full text review were completed in duplicate. Disagreements were resolved by consensus or independent third-party adjudication. <h3>Results</h3> A total of 1351 publications were initially identified. Among these, 78 met our inclusion criteria, primarily addressing themes of prognosis (n=53), diagnosis (n=23), and management (n=2) of HT patients (Fig 1A). AI models were applied to various clinical applications, including the classification and prediction of pre- and post-HT outcomes (e.g., graft function, hospitalization, re-HT, and survival), detection of acute cellular rejection and cardiac allograft vasculopathy, and modelling of blood levels of immunosuppressive agents. In 18 studies, AI models were directly compared against conventional logistic regression, and in half of these studies AI outperformed conventional models. Of note, 31% (n=24) of the included studies were full manuscripts, whereas 69% (n=54) were peer-reviewed conference abstracts (Fig 1B). <h3>Conclusion</h3> AI presents a unique opportunity to enhance organ allocation, donor-recipient matching, and post-HT survival and management. Unfortunately, with the current state of evidence, there is a lack of primary studies that transition from abstract presentation to manuscript publication, perhaps suggesting more robust validation is required prior to the ubiquitous integration of AI into clinical practice.