P1307: predicting short-term and long-term mortalities from sepsis in patients who receive allogeneic hematopoietic stem cell transplantation

Topic: 22. Stem cell transplantation - Clinical Background: Many factors, including myeloablative conditioning, allogeneic haematopoietic stem cell transplantation (allo-HSCT), immunosuppression due to neutropenia, intestinal mucositis or gastrointestinal graft-versus-host disease (GVHD), and the use of central venous catheters, could predispose patients to sepsis after allo-HSCT. Previous studies revealed a high incidence of sepsis and an increasing intensive care unit (ICU) admission rate and mortality for patents with sepsis after allo-HSCT (Intensive Care Med, 2020; Crit Care Med, 2020; Intensive Care Med, 2018; Crit Care Med, 2015). Aims: To establish and externally temporally validate short-term and long-term prognostic models for patients with sepsis after allo-HSCT. Methods: All consecutive patients receiving allo-HSCT who developed sepsis after allo-HSCT at Peking University People’s Hospital between 2012 and 2021 were enrolled. Multivariate logistic regression analysis was used to identify prognostic factors of 6-month and 14-day survival, and then these identified factors were incorporated into two prediction scoring systems. Receiver operating characteristics (ROC), calibration plots, and decision curve analysis (DCA) were used to assess the models’ performance. Results: A total of 287 patients who received allo-HSCT from 2018-2021 were included in the derivation cohort, and 337 patients from 2012-2017 were included in the validation cohort. Eight independent risk factors (advanced age, ≥ grade 2 acute graft-versus-host disease (aGVHD), chronic GVHD (cGVHD), total bilirubin, lactate dehydrogenase (LDH), and organ dysfunction [renal, lung and heart]) were included in the 6-month prognostic model, and 5 independent risk factors (cGVHD, C-reactive protein, LDH, respiratory failure and coagulation dysfunction) were included in the 14-day prognostic model. The area under the ROC curve (AUC) was 0.896 (95% CI, 0.836-0.957) in the derivation cohort and 0.881 (95% CI, 0.836-0.926) in the validation cohort for predicting 6-month death and 0.872 (95% CI, 0.831-0.915) in the derivation cohort and 0.901 (95% CI, 0.866-0.936) in the validation cohort for predicting 14-day death (Figure 1), indicating that the two models both had great discrimination power for prognosis prediction regarding sepsis after allo-HSCT. Moreover, calibration plots and DCA all demonstrated the robust predictive performance of the two models. We also compared the two prognostic models with the SOFA score. The AUCs were 0.661 (95% CI, 0.598-0.723) and 0.746 (95% CI, 0.692-0.799) for predicting 6-month death with the SOFA score for the derivation cohort and the validation cohort, respectively, which were significantly lower than our 6-month prognostic model (both P < 0.001). When predicting 14-day death, the AUC was 0.649 (95% CI, 0.541-0.757) in the derivation cohort and 0.779 (95% CI, 0.717-0.841) in the validation cohort, which was inferior to our 14-day prognostic model (P < 0.001 for the derivation cohort, and P = 0.004 for the validation cohort) (Figure 1). In addition, the calibration plots and DCA also showed better performance for the two models than the SOFA score. We also validated the two models in the septic shock, pre-engraftment sepsis and post-engraftment sepsis subgroups, and all the results showed the generalizability and robust predictive performance of the two prognostic models. Summary/Conclusion: Using the two prognostic models enables the early identification of patients with high-risk 6-month and 14-day deaths so that clinicians can provide timely treatments and improve the therapeutic effects.Figure 1. Receiver operating characteristic (ROC) curve of the prognosis model in the derivation and validation cohorts. Keywords: HSCT, Sepsis, Prognostic