A Novel Model to Predict Progression to Death After Withdrawal of Care in Potential Donation-After-Circulatory-Death Heart Donors.

BACKGROUND: Predicting successful heart donation after circulatory death (DCD) remains a challenge. We developed a model to predict progression to circulatory death after withdrawal of care in potential DCD heart donors. METHODS AND RESULTS: Adult DCD heart offers at a single institution over a six-month period were retrospectively reviewed. Univariate logistic regression was used to assess associations between pre-withdrawal-of-care variables and a binary outcome of circulatory death. The discriminatory performance of multivariate logistic regression and four supervised machine-learning models in predicting circulatory death was assessed by stratified cross validation. The top-performing model type was developed using the full dataset and externally validated in a cohort of potential donors at a distant center. Of 234 included offers, 150 (64.1%) progressed to circulatory death. Factors associated with circulatory death included Glasgow Coma Scale, brainstem reflexes; mean arterial pressure, vasoactive inotropic score, ventilator triggering, positive end-expiratory pressure, PaO2/FiO2 ratio, and serum sodium. A multivariate logistic regression model demonstrated sensitivity of 0.85, specificity of 0.52, and AUC of 0.77 in the development cohort, and sensitivity of 1.00, specificity of 0.50, and AUC of 0.88 in external validation. CONCLUSIONS: We present a novel model to predict progression to circulatory death among potential DCD heart donors.