Abstract
Background: Continuous renal replacement therapy (CRRT) is increasingly employed for critically-ill patients with hematologic disorders. However, Current research on coagulation risk assessment remains limited to specific treatment modalities or anticoagulation methods, resulting in models that are either clinically impractical or prone to evaluation inaccuracies, thereby hindering their utility in clinical practice.This study aimed to construct and validate an independent risk prediction model for extracorporeal circulatory line coagulation in critically ill patients with hematologic disorders on CRRT.
Methods: A two-stage study design was used. Firstly, 134 CRRT patients (66 patients in the coagulation group and 68 patients in the non-coagulation group) in a tertiary care hematology specialty hospital were included in a retrospective study from December 2023 to December 2024, and binary logistic regression was applied to screen the risk factors and build a prediction model; subsequently, external validation was performed in a prospective cohort study (70 patients from January to May 2025). Model efficacy was assessed by area under the ROC curve (AUC) for discrimination, Hosmer-Lemeshow test and calibration curve for calibration, and decision curve analysis (DCA) for clinical utility.
Results: The incidence of extracorporeal circulatory line coagulation in the retrospective cohort was 49.2% (66/134). Multifactorial analysis identified lymphoma (OR = 2.540), mechanical ventilation (OR = 3.123), elevated postfiltration ionized calcium (OR = 1.106), elevated platelet count (OR = 1.012), decreased mean arterial pressure (OR = 0.973), and severe anemia (OR = 0.500) as independent risk factors. Accordingly, a prediction model was developed: p = 1 / [1 + exp(1.163 + 0.930×lymphoma - 0.028×mean arterial pressure + 1.380×mechanical ventilation - 2.243×postfiltration ionized calcium - 0.700×severe anemia + 0.012×platelet count)]. Model validation showed that the AUC of the modeling group was 0.821 (95% CI: 0.750-0.892), with a sensitivity of 82.20% and a specificity of 76.00%, while the AUC of the validation group was 0.817 (95% CI: 0.709-0.924), with a sensitivity of 80.00% and a specificity of 76.00%. The Hosmer-Lemeshow test confirmed that goodness of fit in the modeling group (X2=8.167, P=0.417) and validation group (X2=4.056, P=0.852), the calibration curves showed that the predicted probabilities were in agreement with the actual probabilities, and the DCA suggested that the model had a net clinical benefit in the risk threshold range of 0.1-0.8.
Conclusion: The risk prediction model constructed in this study has good discrimination, calibration, and clinical applicability, and can provide a quantitative tool for the early identification of CRRT extracorporeal circulatory line coagulation in critically ill hematologic patients.
