Validation of a predictive model for platelet transfusion refractoriness in patients with hematological diseases
10.13303/j.cjbt.issn.1004-549x.2025.04.012
- VernacularTitle:血液病患者血小板输注无效的预测模型验证
- Author:
Xiulan HUANG
1
;
Shuhan YUE
1
;
Qun CAI
1
;
Liqi LU
1
;
Mengzhen HE
1
;
Qiao LEI
1
;
Caoyi LIU
1
;
Jingwei ZHANG
1
Author Information
1. Department of Blood Transfusion, Chengdu Second People's Hospital, Chengdu 610051, China
- Publication Type:Journal Article
- Keywords:
hematological disease;
platelet transfusion refractoriness (PTR);
model validation
- From:
Chinese Journal of Blood Transfusion
2025;38(4):537-545
- CountryChina
- Language:Chinese
-
Abstract:
[Objective] To validate and optimize the platelet transfusion refractoriness (PTR) prediction model for patients with hematological disorders established by our center. [Methods] The data of patients with hematological diseases who received platelet transfusions from December 2021 to December 2022 were used as the training set, and data from January 2023 to December 2023 as the validation set. The validation set data was used to validate the predictive model constructed on the training set. Relevant risk factors for PTR were collected through literature review and preliminary studies。 The patients were divided into effective and ineffective groups according to the corrected count increment (CCI) of platelet counts. Predictive factors were screened using univariate and multivariate logistic regression. The calibration of the model were assessed via calibration curves, while discrimination, accuracy, sensitivity, and specificity were evaluated using receiver operating characteristic (ROC) curves Clinical utility was further analyzed with decision curve analysis (DCA). [Results] The Hosmer-Lemeshow (H-L) goodness-of-fit test for the validation set yielded S: P=0.000, indicating that the original model needs optimization. Baseline comparisons and logistic regression identified the number of red blood cell units (RBCU) and platelet units (PLT-U) transfused as key predictors for the optimized model. The H-L goodness-of-fit test S: P values for the training and validation sets were 0.930 and 0.056, respectively; the ROC areas were 0.793 5 and 0.809 4, specificities 90.95% and 84.21%, sensitivities 59.26% and 70.04%, and accuracies 78.14% and 74.10%, respectively. DCA demonstrated clinical net benefit within a prediction probability threshold range of 0.2-0.8. [Conclusion] Transfusion volumes of RBC-U and PLT-U were inversely associated with PTR in hematological patients. The resulting PTR prediction model exhibits moderate predictive efficacy and clinical benefit.
