Machine Learning Models for Posttransplant Lymphoproliferative Disorder (PTLD) Risk Prediction in Thoracic Transplantation

Posttransplant lymphoproliferative disorder (PTLD) is the second most common malignancy in thoracic transplant recipients and is associated with poor survival. Accurate risk prediction is essential for prevention and management. Adult (≥18 years) heart, lung, and heart-lung transplant recipients were extracted from the Scientific Registry of Transplant (SRTR) and the United Network of Organ Sharing (UNOS) databases. A total of 160 donor and recipient pretransplant variables, including demographics, laboratory tests, induction therapy, and human leukocyte antigens (HLA), were analyzed. Risk scores were developed using the FasterRisk algorithm and compared with statistical and machine learning models. Among 89,139 thoracic recipients, the model achieved cross-validated areas under the curve of 0.776, 0.711, and 0.689 for 1, 3, and 5 year PTLD risk prediction, respectively. Steroid induction and previous malignancy were associated with an increased PTLD risk. Younger age (18-27 years at 1 year; 18-23 years at 5 years) was also linked to higher risk. In contrast, positive Epstein-Barr virus (EBV) status, heart transplantation (compared with lung or combined heart-lung), African American ethnicity, and basiliximab induction were associated with a lower risk. The proposed risk scores enhance understanding of PTLD risk factors and enable individualized prediction during the first 5 years after thoracic transplantation.