A pragmatic monitoring model for risk of non‐relapse mortality based on biomarkers in patients undergoing allogeneic haematopoietic stem‐cell transplantation

Biomarkers predict non-relapse mortality (NRM) primarily driven by acute graft-versus-host disease (aGVHD) after allogeneic haematopoietic stem cell transplantation (allo-HSCT), yet existing single time point models lack clinical adaptability. In a prospective observational multicentre Chinese cohort of 619 allo-HSCT patients, serum concentrations of suppressor of tumorigenesis 2 (ST2), regenerating islet-derived 3α, tumour necrosis factor receptor 1 (TNFR1), interleukin-6 and interleukin-8 were measured at multiple time points during the first 3 months post-transplantation. Cohorts were divided into modelling (n = 460, training/test) and validation (n = 159) sets. A model based on the approximately maximum concentrations of ST2 and TNFR1 reached area under the curve of 0.85 and 0.95 in the test and validation cohorts, respectively, which was the highest among the models based on other biomarker panels. The model stratified validation cohort patients into high-risk (47.37%), moderate-risk (9.76%) and low-risk (0.0%) groups with significantly different 6-month NRM (p < 0.001). Using a universal threshold, the model consistently stratified patients into moderate-high-risk versus low-risk groups for NRM at all time points (days 7, 14, 28, aGVHD onset). Cumulative 6-month NRM was significantly higher in moderate-high-risk groups (20.00%, 33.33%, 37.50%, 30.77%) versus low-risk groups (5.22%, 5.22%, 3.88%, 7.14%). The application of this model is not confined to specific time points or clinical characteristics, rendering it a flexible and pragmatic tool in a clinical setting.