Machine learning algorithm as a prognostic tool for Epstein-Barr virus reactivation after haploidentical...

Area of Science:

• Hematopoietic Stem Cell Transplantation (HSCT)
• Infectious Disease
• Machine Learning in Medicine

Background:

• Epstein-Barr virus (EBV) reactivation is a major complication following haplo-identical related donor (HID) hematopoietic stem cell transplantation (HSCT).
• Graft-versus-host disease (GVHD) prophylaxis with anti-thymocyte globulin (ATG) is commonly used in HID HSCT, but its impact on EBV reactivation risk requires further understanding.
• Predictive models are crucial for identifying patients at high risk of EBV reactivation to enable timely intervention.

Purpose of the Study:

• To develop and validate a comprehensive machine learning model to predict Epstein-Barr virus (EBV) reactivation after HID HSCT.
• To identify key clinical and biological factors associated with EBV reactivation in this patient population.
• To stratify patients into low- and high-risk groups for EBV reactivation.

Main Methods:

• A cohort of 470 consecutive acute leukemia patients undergoing HID HSCT with ATG for GVHD prophylaxis was analyzed.
• Patients were divided into training (60%) and validation (40%) cohorts.
• A machine learning model was developed incorporating variables such as age, gender, disease status, HLA disparity, EBV serostatus, graft composition, and conditioning regimen.

Main Results:

• The developed model demonstrated significant predictive accuracy for EBV reactivation.
• Key predictors included age, gender, disease status, HLA disparity, EBV serostatus, and graft characteristics.
• The model successfully stratified patients into low- and high-risk groups, with significantly different 1-year cumulative incidences of EBV reactivation (11.0% vs. 24.5% overall).

Conclusions:

• A comprehensive machine learning model can accurately predict EBV reactivation post-HID HSCT with ATG prophylaxis.
• This model aids in identifying at-risk patients, facilitating personalized monitoring and management strategies.
• The model also showed potential in predicting relapse and survival outcomes.