Related Experiment Video
Updated: Jun 27, 2026
Predictive Immune Modeling of Solid Tumors
Published on: February 25, 2020
Development and validation of a predictive model for chronic or persistent immune thrombocytopenia in children
Ximei Yang1, Xianghui Wan2, Lei Ying1
1Department of Laboratory Medicine, Jiangxi Provincial Children's Hospital, Nanchang, China.
Background:
Pediatric immune thrombocytopenia (ITP) is an autoimmune condition involving platelet destruction. While most cases resolve spontaneously, 20%-30% progress to chronic disease (cITP), necessitating early risk prediction. Current prognostic models for childhood ITP lack platelet-specific immunological markers. This study aimed to develop and validate a predictive model for chronic/persistent ITP(c/pITP) integrating clinical factors and anti-platelet glycoprotein antibodies.
Methods:
We retrospectively analyzed 381 children with newly diagnosed ITP (2022-2024). Patients were randomly allocated into training (n = 266) and test (n = 115) sets (7:3 ratio, seed = 42). Missing data were addressed via multiple imputation (m = 5). Predictors were selected through univariate analysis (P < 0.1), LASSO regression with 10-fold cross-validation, and bootstrap stability assessment (λ.1se, B = 1,000, selection frequency ≥ 90%). Candidate models were compared using bootstrap optimism correction and the Equivalent Parsimony Principle. The final model was evaluated on the independent test set for discrimination (AUC), calibration (Hosmer-Lemeshow test), and clinical utility (decision curve analysis).
Results:
Fourteen highly stable predictors were identified; the top three were platelet count (PLT, 100%), recent infection or vaccination history (IOV, 100%), and anti-GPIIb antibody status (99.9%). The final 4-variable model (age, IOV, PLT, anti-GPIIb) demonstrated good discrimination in the test set (AUC = 0.743; 95% CI: 0.650-0.836), with no evidence of overfitting (training AUC = 0.732). Calibration was satisfactory (Hosmer-Lemeshow P = 0.859). At a low-risk threshold (predicted probability ≤ 15%), the negative predictive value reached 94.7%, facilitating the rule-out of chronic progression. Decision curve analysis confirmed positive net benefit across threshold probabilities of 0%-60%.
Conclusion:
This concise model incorporating anti-GPIIb antibody enables early risk stratification, primarily to identify children unlikely to develop c/pITP.
