Cytokine-based HAG model accurately predicts acute GvHD risk in ATG-based haploidentical HSCT recipients

Acute graft-versus-host disease (aGvHD) remains a severe complication following haploidentical hematopoietic stem cell transplantation (haplo-HSCT), significantly contributing to non-relapse mortality. Current predictive models, primarily developed for HLA-matched donor cohorts, often fall short in accurately assessing aGvHD risk specifically in haplo-HSCT recipients receiving antithymocyte globulin (ATG)-based prophylaxis. This gap necessitates a more tailored predictive tool, leveraging biomarkers linked to epithelial injury and inflammatory signaling, to enable earlier and more precise risk stratification.

Researchers retrospectively analyzed 280 patients undergoing ATG-based haplo-HSCT across two centers, dividing them into training, internal test, and external validation cohorts. They first assessed the predictive accuracy of the established Mount Sinai Acute GvHD International Consortium (MAGIC) algorithm for aGvHD and steroid-refractory aGvHD (SR-aGvHD). Subsequently, plasma concentrations of candidate cytokines (ST2, REG3α, Elafin, and TNFRI) were measured. Predictive and causal relationships with aGvHD were evaluated using logistic regression, weighted average area under the curve (wAUC), Mendelian randomization (MR), and restricted cubic spline (RCS) analyses. A new predictive model, the HAG model, was then constructed based on identified key cytokines.

The MAGIC algorithm demonstrated effectiveness in the ATG-based haplo-HSCT setting for predicting aGvHD, achieving AUC values of 0.693 (training), 0.658 (internal test), and 0.622 (external validation). Among the candidate cytokines, a specific combination of ST2, REG3α, and Elafin — forming the HAG model — consistently showed the highest predictive accuracy for aGvHD. Mendelian randomization analysis, utilizing external genome-wide association (GWAS) datasets, further supported potential causal associations of ST2 and REG3α with aGvHD development. This robust validation across multicenter cohorts underscores the HAG model's reliability.

The HAG model, comprising ST2, REG3α, and Elafin, demonstrated superior predictive accuracy for acute GvHD compared to the MAGIC algorithm in ATG-based haplo-HSCT patients.