ABSTRACT
Heart failure (HF) is an uncommon but serious cardiovascular complication after allogeneic hematopoietic stem cell transplantation (allo-HSCT). Unfortunately, knowledge about early mortality prognostic factors in patients with HF after allo-HSCT is limited, and an easy-to-use prognostic model is not available. This study aimed to develop and validate a clinical-biomarker prognostic model capable of predicting HF mortality following allo-HSCT that uses a combination of variables readily available in clinical practice. To investigate this issue, we conducted a retrospective analysis at our center with 154 HF patients who underwent allo-HSCT between 2008 and 2021. The patients were separated according to the time of transplantation, with 100 patients composing the derivation cohort and the other 54 patients composing the external validation cohort. We first calculated the univariable association for each variable with 2-month mortality in the derivation cohort. We then included the variables with a P value <.1 in univariate analysis as candidate predictors in the multivariate analysis using a backward stepwise logistic regression model. Variables remaining in the final model were identified as independent prognostic factors. To predict the prognosis of HF, a scoring system was established, and scores were assigned to the prognostic factors based on the regression coefficient. Finally, 4 strongly significant independent prognostic factors for 2-month mortality from HF were identified using multivariable logistic regression methods with stepwise variable selection: pulmonary infection (P = .005), grade III to IV acute graft-versus-host disease (severe aGVHD; P = .033), lactate dehydrogenase (LDH) >426 U/L (P = .049), and brain natriuretic peptide (BNP) >1799 pg/mL (P = .026). A risk grading model termed the BLIPS score (for BNP, LDH, cardiac troponin I, pulmonary infection, and severe aGVHD) was constructed according to the regression coefficients. The validated internal C-statistic was .870 (95% confidence interval [CI], .798 to .942), and the external C-statistic was .882 (95% CI, .791-.973). According to the calibration plots, the model-predicted probability correlated well with the actual observed frequencies. The clinical use of the prognostic model, according to decision curve analysis, could benefit HF patients. The BLIPS model in our study can serve to identify HF patients at higher risk for mortality early, which might aid designing timely targeted therapies and eventually improving patients' survival and prognosis.
