# Machine Learning–Based Evaluation of Combined EBV and CMV Serostatus as Predictors of Post-Transplant Lymphoproliferative Disorder

## TL;DR

This study uses machine learning to show that combining EBV and CMV serostatus improves prediction of post-transplant lymphoproliferative disorder risk in organ transplant patients.

## Contribution

The novel contribution is demonstrating that combined EBV/CMV serostatus provides better PTLD risk prediction than EBV alone using machine learning models.

## Key findings

- PTLD incidence was 1.5% overall, with highest risk in EBV D+/R− recipients (3.2%).
- Combined [EBV D+/R−, CMV D−/R−] serostatus had more than double the PTLD risk compared to [EBV D+/R−, CMV D+/R−].
- Machine learning models identified combined serostatus, age, and race as key predictors, though discrimination was modest (AUC ∼0.61).

## Abstract

Post-transplant lymphoproliferative disorder (PTLD) is a major complication of solid organ transplantation (SOT), with the greatest risk in Epstein–Barr virus (EBV) donor-positive/recipient-negative (D+/R−) pairs. The contribution of cytomegalovirus (CMV) serostatus is less well defined. We conducted a population-based study of 47,333 abdominal SOT recipients in the United States (1995–2015) using linked SRTR data. Donor–recipient EBV/CMV serostatus was evaluated as a compound variable. The primary outcome was PTLD incidence, with secondary analyses assessing predictors of PTLD and impact on survival. Overall, 716 patients (1.5%) developed PTLD at a median of 6.1 years (IQR 2.9–9.7) after transplant. EBV D+/R- recipients had the highest incidence (3.2%), and those with compound [EBV D+/R−, CMV D−/R−] serostatus had more than double the PTLD risk compared with [EBV D+/R−, CMV D+/R−] (5.3% vs. 2.5%, p < 0.001). Logistic regression and random forest models consistently identified [EBV D+/R−, CMV D-/R-] serostatus, age, and race as leading predictors, though discrimination was modest (test AUC ∼0.61). In a matched survival analysis, PTLD was not associated with increased all-cause mortality (aHR ∼1.0). Our findings demonstrate that combined EBV/CMV serostatus improves PTLD risk prediction compared with EBV alone and emphasize the need for targeted preventive strategies.

Machine learning-based evaluation of combined EBV and CMV serostatus as predictors of post-transplant lymphoproliferative disorder. The study involves 47,333 cases, analyzing variables like patient characteristics and EBV-CMV serology. Outcomes focus on post-transplant PTLD. Data analysis methods include single variant analysis, random forest, and logistic regression. Results show a PTLD incidence of 1.5%, with EBV D+/R− having the highest incidence at 3.2%. The conclusion states that combined EBV/CMV serostatus improves PTLD risk prediction compared to EBV alone. Data from Ghazal Azarfar et al., published in Transplant International 2025.