Table 2.

Summary of publications describing artificial intelligence application in practicing post heart transplant care.

Author	Title	Data source used	Primary outcome	Best model performance	Study conclusion
Hoda (26)	Prediction of cyclosporine blood levels in heart transplantation patients using a pharmacokinetic model identified by evolutionary algorithms	Local institutional data from the Heart and Lung Transplantation Center, University of Vienna Medical School	Cyclosporin blood level	Evolutionary algorithms model had a mean percent error of 8.0 ± 6.7%.	AI model accurately predicted cyclosporine whole blood levels in heart transplant recipients.
Chen (24)	Quantitative 3D analysis of coronary wall morphology in heart transplant patients: OCT-assessed cardiac allograft vasculopathy progression	Local patients at Transplant Center at Institute of Clinical and Experimental Medicine and the Center for Cardiovascular and Transplantation Surgery, Brno, Czech	Coronary artery intimal thickness	Exclusion regions determined by transfer learning using ImageNet network achieved an accuracy of 81.2%.	AI allows quantification of location-specific alterations of coronary wall morphology over time and is sensitive even to very small changes of wall layer thicknesses.
Peyster (22)	An automated computational image analysis pipeline for histological grading of cardiac allograft rejection	Local institutional data at Hospital of the University of Pennsylvania, Cleveland Medical Center, and the Ohio State University Wexner Medical Center	Histopathologic rejection detection	A support vector machine classification model had an AUC of 0.92.	The grader pipeline, derived using intuitive morphological features, can provide expert-quality rejection grading.
Woillard (27)	Tacrolimus exposure prediction using machine learning	local institutional data	Blood concentration of tacrolimus (TAC) following twice daily vs. daily dosing	XBBoost models to estimate TAC blood AUC based on 2 measurements showed mean prediction error close to 0; and root mean square error < 10%.	AI allows accurate estimation of TAC interdose AUC and can be used for routine TAC exposure estimation and dose adjustment.
Lipkova (21)	Deep learning-enabled assessment of cardiac allograft rejection from endomyocardial biopsies	local institutional data from the Brigham and Women’s Hospital	Histopathologic rejection detection, classification, and grading	A neural network (transfer learning from ResNet50 and attention-based multiple instance learning) model had an AUC of 0.962 for detecting allograft rejection.	The AI system showed non-inferior performance to experts and reduced interobserver variability and assessment time.
Piening (23)	Whole transcriptome profiling of prospective endomyocardial biopsies reveals prognostic and diagnostic signatures of cardiac allograft rejection	Local data from the CTOT-03 trial (NCT:0053192) population from University of Pennsylvania and the University of Wisconsin	A gene expression classifier for 0R/1R vs. 2R acute rejection	Random Forest model had an AUC of 0.947.	RNA-seq-based molecular characterization of EMBs shows significant promise for the early detection of cardiac allograft rejection.
Wei (25)	The novel proteomic signature for cardiac allograft vasculopathy	Local patients at University Hospitals Leuven, Belgium	Detection of cardiac allograft vasculopathy (CAV)	XgBoost model showed an AUC 0.71, 95% CI 0.60–0.81.	The proteomic signature might provide insights into CAV pathological processes and help study personalized treatment targets.

strong recommendation, weak recommendation/promising but not ready to implement, as suggested by the respective paper.