Predictors of 1-year outcome after cardiac re-transplantation: Machine learning analysis

Abstract

Background: As cardiac re-transplantation is associated with inferior outcomes compared with primary transplantation, allocating scarce resources to appropriate re-transplant candidates is important. The aim of this study is to elucidate the factors associated with 1-year mortality in cardiac re-transplantation using the random forests algorithm for survival analysis.

Methods: We retrospectively reviewed the United Network for Organ Sharing registry and identified all adult (> 17 years old) recipients who underwent cardiac re-transplantation between January 2000 and March 2020. The random forest algorithm on Cox modeling was used to calculate the variable importance (VIMP) of independent variables for contributing to 1-year mortality.

Results: A total of 1294 patients underwent cardiac re-transplantation. Of these, 137 patients were re-transplanted within 1 year of their first transplant, while 1157 patients were re-transplanted more than 1 year after their first transplant. One-year mortality was significantly higher for patients receiving early transplantation compared with those receiving late transplantation (Early 40.6% vs. Late 13.6%, log-rank P < .001). Machine learning analysis showed that total bilirubin (> 2 mg/dl) (VIMP, 2.99%) was an independent predictor of 1-year mortality after early re-transplant. High BMI (> 30.0 kg/m² ) (VIMP, 1.43%) and ventilator dependence (VIMP, 1.47%) were independent predictors of 1-year mortality for the late re-transplantation group.

Conclusion: Machine learning showed that optimal 1-year survival following cardiac re-transplantation was significantly related to liver function in early re-transplantation, and to obesity and preoperative ventilator dependence in late re-transplantation.

Keywords: United Network for Organ Sharing (UNOS); heart (allograft) function/dysfunction; heart transplantation; machine learning; retransplantation.