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

Area of Science:

• Transplant Oncology
• Immunology
• Biostatistics

Background:

• Posttransplant lymphoproliferative disorder (PTLD) is a significant malignancy in thoracic transplant recipients, impacting survival rates.
• Accurate risk prediction models are crucial for PTLD prevention and effective management strategies.

Purpose of the Study:

• To develop and validate risk prediction scores for PTLD in adult thoracic transplant recipients.
• To identify pretransplant variables associated with PTLD risk within the first five years post-transplant.

Main Methods:

• Analysis of 160 pretransplant variables from 89,139 adult heart, lung, and heart-lung transplant recipients using SRTR and UNOS data.
• Development of risk scores using the FasterRisk algorithm, compared against statistical and machine learning models.
• Cross-validation to assess model performance for 1, 3, and 5-year PTLD risk prediction.

Main Results:

• The developed model achieved cross-validated AUCs of 0.776 (1-year), 0.711 (3-year), and 0.689 (5-year).
• Increased PTLD risk was associated with steroid induction, prior malignancy, and younger recipient age (18-27 years).
• Decreased PTLD risk was linked to positive Epstein-Barr virus (EBV) status, heart transplants, African American ethnicity, and basiliximab induction.

Conclusions:

• The proposed risk scores offer enhanced understanding of PTLD risk factors post-thoracic transplantation.
• Individualized PTLD risk prediction is feasible within the first five years, aiding clinical decision-making.
• Identifying high-risk patients allows for targeted monitoring and potential interventions to improve outcomes.