The international heart transplant survival algorithm (IHTSA): A new model to improve organ sharing and survival

Abstract

Background Heart transplantation is life saving for patients with end-stage heart disease. However, a number of factors influence how well recipients and donor organs tolerate this procedure. The main objective of this study was to develop and validate a flexible risk model for prediction of survival after heart transplantation using the largest transplant registry in the world. Methods and Findings We developed a flexible, non-linear artificial neural networks model (IHTSA) and classification and regression tree to comprehensively evaluate the impact of recipient-donor variables on survival over time. We analyzed 56,625 heart-transplanted adult patients, corresponding to 294,719 patient-years. We compared the discrimination power with three existing scoring models, donor risk index (DRI), risk-stratification score (RSS) and index for mortality prediction after cardiac transplantation (IMPACT). The accuracy of the model was excellent (C-index 0.600 [95% CI: 0.595 - 0.604]) with predicted versus actual 1-year, 5-year and 10-year survival rates of 83.7% versus 82.6%, 71.4% - 70.8%, and 54.8% - 54.3% in the derivation cohort; 83.7%versus 82.8%, 71.5% - 71.1%, and 54.9% - 53.8% in the internal validation cohort; and 84.5% versus 84.4%, 72.9% - 75.6%, and 57.5% - 57.5%in the external validation cohort. The IHTSA model showed superior or similar discrimination in all of the cohorts. The receiver operating characteristic area under the curve to predict oneyear mortality was for the IHTSA: 0.650 (95% CI: 0.640-0.655), DRI 0.56 (95% CI: 0.56-0.57), RSS 0.61 (95% CI: 0.60- 0.61), and IMPACT 0.61 (0.61-0.62), respectively. The de-cision-tree showed that recipients matched to a donor younger than 38 years had additional expected median survival time of 2.8 years. Furthermore, the number of suitable donors could be increased by up to 22%. Conclusions We show that the IHTSA model can be used to predict both short-term and long-term mortality with high accuracy globally. The model also estimates the expected benefit to the individual patient.