Development and Validation of a Risk Model for Primary Graft Failure after Heart Transplantation and Comparison to the RADIAL Risk Score

Objectives: Primary graft failure (PGF) represents the leading cause of early mortality following heart transplantation. Many risk factors and scores for the development of PGF have been defined, e.g., the established RADIAL score, but these risk scores are less predictive when they are validated in multi center studies. The aim of this study is to develop and validate a risk score based on the ISHLT Transplant Registry Data.

Methods: We used the ISHLT datasets of 58,250 transplant (TX) procedures in 57,188 patients between 2005 and 2017. We excluded patients who underwent previous TX, multiorgan TX (n = 2,134), and pediatric patients (n = 6,944). Further, we only included patients with data available on age, sex, height, weight of donor and recipient, as well as PA mean, PCW, and CO of the recipient. The final dataset for analysis consisted of 23,004 patients. A patient was defined as a case of PGF if it was listed as cause of graft failure, death, or retransplantation resulting in 644 cases of PGF. Since RA pressure (part of the RADIAL risk score) was unavailable in the ISHLT registry, a modified RADIAL risk score excluding RA pressure and including all other components of the RADIAL risk score was used.

At first, 177 variables were chosen as possible predictors of PGF. Based on TX year, we split the study sample into a derivation cohort and a validation cohort. The LASSO algorithm was used for variable selection. A logistic regression model was then built using the selected variables. The performance of the model (measured by the AUC) was assessed in both the training and the test dataset, and compared to the performance of the modified RADIAL score.

Results: A model including 15 predictors was derived in the derivation cohort. The AUC of this model was 0.69. After correcting for overoptimism, the AUC decreased to 0.67. The AUC in the validation cohort was similar (0.68). The modified RADIAL score performed worse (AUC = 0.57).

Conclusion: We developed a new score to predict primary graft failure. Based on the ISHLT dataset, this new model has a higher predictive value than the established RADIAL score. This new model has the advantage of including both donor and recipient characteristics, and might therefore improve patient outcome by avoiding or early treatment of primary graft failure.