Prediction of avatrombopag-induced thrombocytosis in pediatric immune thrombocytopenia: an AI-based real-world study

Area of Science:

• Pediatric Hematology
• Pharmacovigilance
• Artificial Intelligence in Medicine

Background:

• Avatrombopag (AVA), an oral thrombopoietin receptor agonist (TPO-RA), is effective for pediatric immune thrombocytopenia (ITP).
• Treatment-induced thrombocytosis is a significant adverse event associated with AVA, impacting treatment safety and continuity.
• Currently, no validated tools exist to predict the risk of AVA-induced thrombocytosis prior to treatment initiation.

Purpose of the Study:

• To develop and validate a predictive model for avatrombopag-associated thrombocytosis in pediatric ITP patients.
• To compare the performance of a novel predictive model against established machine learning and deep learning algorithms.
• To enhance model interpretability for clinical application in risk assessment.

Main Methods:

• A real-world study involving 74 pediatric patients with ITP treated with AVA.
• Development and validation of a predictive model using clinical data.
• Comparison with baseline models (Logistic Regression, SVM, MLP, RF, XGBoost) and deep learning models (TabPFN, FT-Transformer, HyperTab).
• Model interpretability analysis using Shapley Additive Explanations (SHAP).

Main Results:

• The FT-Transformer model demonstrated superior performance with an accuracy of 0.785 ± 0.023 and an AUC of 0.851 ± 0.021.
• SHAP analysis provided insights into individual feature contributions to thrombocytosis risk.
• The AI-driven model showed robust predictive performance on real-world clinical data.

Conclusions:

• An AI-driven predictive model for avatrombopag-induced thrombocytosis in pediatric ITP is developed and validated.
• The model offers clinically interpretable insights for individualized risk assessment.
• This tool supports safer and more precise use of avatrombopag in managing pediatric ITP.