Introduction
The clinical management of end-stage heart disease is a continually
evolving practice that has changed drastically over the last decade.
Survival after orthotopic heart transplantation (OHT) has continued to
improve.1 Despite improving longevity after
transplant, donor organ supply remains inadequate to meet
demand,2 and transplant programs face increased public
and private scrutiny of their outcomes.3Simultaneously, technologic innovations in mechanical circulatory
support platforms have demonstrated parallel improvement in clinical
outcomes,4-6 thus increasing the potential alternative
viable treatment options for heart failure patients. Therefore, an
accurate prognostic model using pre-operative data for individualized
donor and recipient selection would be of profound clinical utility in
OHT.
Prior risk models for predicting survival after OHT have displayed only
modest discriminatory capability. Examples of such algorithms include
The Donor Risk Index (DRI),7 Risk Stratification Score
(RSS),8 Index for Mortality Prediction After Cardiac
Transplantation (IMPACT)9 and International Heart
Transplant Survival Algorithm (IHTSA)10. With
increasing interest in the application of machine learning (ML) to
predictive analytics in clinical medicine,11 we aimed
to evaluate whether modern ML techniques could improve risk prediction
in OHT.