Variations in institutional staffing and clinical practice are predictive of center-specific 1-year survival...

The accuracy of various risk models to predict early post-transplant mortality is limited by the type, quality, and era of the data collected. Most models incorporate a large number of recipient-derived and donor-derived variables; however, other factors related to specific institutional practices likely influence early mortality. The goal of this study was to determine if the addition of institutional practice variables would improve the predictive accuracy of a recipient/donor risk model in a modern cohort of heart transplant recipients.

Between 1999 and 2007, 3,591 primary heart transplants were performed at the 26 institutions participating in the Cardiac Transplant Research Database. Multivariable regression analysis in the hazard domain was used to identify recipient, donor, and institutional practice variables that were predictive of 1-year mortality. The derived model was used to predict institutional outcomes and compare them with observed outcomes first without and then with the inclusion of the institutional practice variables.

Eleven individual plus 2 interaction recipient variables and 2 individual plus 2 interaction donor variables were predictive of increased mortality. The addition of institutional practice variables to the model identified 4 variables associated with decreased mortality: greater number of transplant cardiologists, a thoracic surgery fellowship, a surgery or cardiology attending taking donor call, and routine surveillance for antibody-mediated rejection. By using a p-value > 0.10 as a robust measure of similarity, the addition of institutional practice variables increased the number of institutions with similar predicted vs. observed mortality from 18 of 26 institutions (69%) to 26 of 26 (100%), demonstrating improved predictive accuracy of the model.

Multiple recipient and donor variables influence early survival but do not fully explain the difference in predicted and observed outcomes at the institutional level. Variations in staffing and clinical practice contribute to risk, and the addition of these variables to our risk model improved predictive accuracy.