Original Research
JIAO Xiaolin, NING Shaoxiong, YUAN Mei, CHAO Yao, DUAN Lixiang
Objective: To establish a predictive model of platelet transfusion efficacy in patients with myelodysplastic syndrome (MDS) based on machine learning algorithms, and to validate its predictive performance and clinical utility in a patient cohort.
Methods: A total of 160 MDS patients in Yuncheng Central Hospital Affiliated to Shanxi Medical University from January 2021 to December 2024 were enrolled as the training set and divided into an ineffective group and an effective group according to platelet transfusion efficacy. Characteristic variables for constructing the prediction model of platelet transfusion efficacy in MDS patients were sequentially screened using least absolute shrinkage and selection operator (LASSO) regression analysis and logistic regression analysis. Four machine learning algorithms, namely extreme gradient boosting (XGBoost), decision tree, random forest and logistic regression, were respectively employed to construct the prediction models. Predictive performance of each model was evaluated using sensitivity, specificity, the area under the curve (AUC) of receiver operating characteristic (ROC) and Youden index. The decision curve analysis was further used to evaluate the clinical practicability of the optimal model. In addition, 73 patients with MDS in Yuncheng Central Hospital Affiliated to Shanxi Medical University and the Second Hospital of Shanxi Medical University from January to May 2025 were selected as the time validation set to evaluate the clinical generalizability of the optimal model.
Results: Among the 160 patients with MDS in the training set, the rate of platelet transfusion inefficiency was 34.38% (55/160). The proportions of fever, splenomegaly, platelet antibody positive and platelet transfusion times ≥5 times, as well as the levels of serum interleukin (IL)-1β and IL-8 in the ineffective group were higher than those in the effective group (all P < 0.05). Logistic regression analysis showed that splenomegaly, platelet antibody positive, platelet transfusion times and the levels of serum IL-1β and IL-8 were independent risk factors for ineffective platelet transfusion in MDS patients (all P < 0.05). The AUC values of XGBoost model were 0.946 [95% confidence interval (CI): 0.899-0.975] in the training set and 0.947 (95% CI: 0.868-0.986) in the validation set, which were both significantly higher than those of the random forest model [0.871 (95% CI: 0.809-0.919) and 0.830 (95% CI: 0.723-0.907)], decision tree model [0.856 (95% CI: 0.792-0.907) and 0.814 (95% CI: 0.705-0.895)] and logistic regression model [0.849 (95% CI: 0.784-0.901) and 0.804 (95% CI: 0.695-0.888)] (all P < 0.05). Additionally, in both the training and validation sets, the XGBoost model demonstrated the sensitivities of 90.91% and 91.67%, and the specificities of 89.52% and 91.84%, which were both significantly higher than those of the other three predictive models (both P < 0.05). The XGBoost model ranked the factors that increase the risk of ineffective platelet transfusion in MDS patients in order of importance, namely platelet antibody positive, platelet transfusion frequency, serum IL-8 level, serum IL-1β level and splenomegaly. The decision curve showed that when the threshold probability ranged from 0.01 to 0.99, applying the XGBoost model to predict the efficacy of platelet transfusion in MDS patients consistently demonstrates that the clinical benefits derived from correct interventions outweigh the losses caused by misjudgments, which has good clinical practical value.
Conclusion: The efficacy of platelet transfusion therapy for MDS patients is not ideal. The XGBoost model, constructed based on five clinical indicators including platelet antibody positivity, platelet transfusion frequency, serum IL-8 level, serum IL-1β level and splenomegaly, has the best comprehensive predictive performance for the platelet transfusion effect in MDS patients and has good clinical practical value. It can provide references for clinical adjustment, optimization of treatment plans and improvement of platelet transfusion efficiency.