Abstract / Summary
Non-relapse mortality (NRM) remains a leading cause of transplant failure in patients with high-risk or refractory acute leukemia (HR/R-AL) undergoing allogeneic hematopoietic stem cell transplantation (allo-HSCT). We developed and internally validated a competing-risk nomogram for predicting NRM, designed for early post-transplant risk stratification. This single-center retrospective study enrolled 587 h/R-AL patients who underwent first allo-HSCT between January 2015 and December 2021. Patients were randomly allocated to a training cohort (n = 411, 70%) and a validation cohort (n = 176, 30%). Using the Fine-Gray competing-risk framework with relapse as the competing event, least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation was applied for predictor selection. A nomogram was constructed from the final multivariable model. Model discrimination was evaluated by time-dependent area under the receiver operating characteristic curve (AUC) and the concordance index (C-index), with internal validation via bootstrap resampling (B = 200). Calibration was assessed by optimism-corrected calibration slope, and clinical utility by decision curve analysis (DCA). A head-to-head comparison with the Hematopoietic Cell Transplantation-specific Comorbidity Index (HCT-CI) was performed. Six predictors were retained: age, pre-transplantation disease status (complete remission vs. non-CR), HCT-CI score (low/intermediate/high), white blood cell engraftment time (> 21 vs. ≤21 days), platelet engraftment time (> 14 vs. ≤14 days), and acute graft-versus-host disease grade (II–IV vs. 0–I). In the training cohort, the 1-, 2-, and 3-year AUCs were 0.786 (95% confidence interval [CI] 0.731–0.840), 0.764 (0.710–0.818), and 0.746 (0.690–0.803); in the validation cohort, 0.818 (0.739–0.896), 0.799 (0.720–0.877), and 0.806 (0.728–0.884), respectively. The optimism-corrected C-index was 0.715 (apparent 0.730) in the training cohort and 0.767 (95% CI 0.703–0.831) in the validation cohort. Bootstrap-corrected calibration slopes were 0.94, 0.92, and 0.97 at 1, 2, and 3 years in the training cohort; external validation slopes were 1.30, 1.14, and 1.16, respectively. DCA demonstrated that the nomogram provided positive net benefit across a clinically relevant range of threshold probabilities (5–50%), consistently outperforming both “treat-all” and “treat-none” default strategies. The nomogram significantly outperformed HCT-CI in discrimination (C-index difference: 0.114, P = 0.020 in training; 0.191, P = 0.008 in validation). This competing-risk nomogram integrates pre-transplant characteristics with early post-transplant events to provide reliable NRM risk stratification after allo-HSCT for HR/R-AL patients.