Machine learning to forecast rituximab responses for paediatric immune thrombocytopenia: Forging a path towards
Jingyao Ma1, Chang Cui2,3, Juntao Ouyang1
1Hematology Department, Beijing Key Laboratory of Pediatric Hematology Oncology; National Key Discipline of Pediatrics (Capital Medical University); Key Laboratory of Major Diseases in Children, Ministry of Education, Beijing Children's Hospital, National Center for Children's Health, Capital Medical University, Beijing, China.
Insights
This study developed a machine learning model to predict rituximab treatment response in children with immune thrombocytopenia (ITP). The model identifies key factors to help personalize therapy and improve outcomes for pediatric patients.
Area of Science:
- Immunology
- Pediatrics
- Computational Biology
Background:
- Primary immune thrombocytopenia (ITP) is an autoimmune condition affecting children, leading to low platelet counts and bleeding risks.
- While many pediatric cases resolve spontaneously, some require second-line therapies like rituximab due to treatment resistance.
- Predicting rituximab efficacy in pediatric ITP remains a clinical challenge.
Purpose of the Study:
- To develop and validate a machine learning (ML) model for predicting the initial response to rituximab in pediatric ITP patients.
- To identify key clinical and immunological features that predict rituximab efficacy.
- To potentially optimize treatment strategies and improve patient outcomes in pediatric ITP.
Main Methods:
- Retrospective analysis of data from 156 pediatric ITP patients treated with rituximab.
- Development and evaluation of four machine learning models, including a multilayer perceptron.
- Utilized SHapley Additive exPlanations (SHAP) to interpret model predictions and identify significant features.
Main Results:
- The multilayer perceptron model demonstrated the highest predictive accuracy for rituximab response.
- Key positive predictors included antinuclear antibody titre, thyroglobulin antibody, corticosteroid response, and bleeding severity.
- Negative predictors comprised thyroid peroxidase antibody, specific T cell populations (CD3+ CD4+ IL-17+), and disease duration before treatment.
Conclusions:
- A novel ML model can accurately predict rituximab response in pediatric ITP.
- The model highlights specific biomarkers and clinical factors influencing treatment efficacy.
- This predictive tool may aid in tailoring rituximab therapy for better outcomes in children with ITP.
Abstract:
Primary immune thrombocytopenia (ITP) is an autoimmune disorder characterized by decreased platelet counts and increased bleeding risk. Although paediatric ITP often resolves spontaneously, some children do not respond to first-line treatments, thus requiring rituximab as a second-line therapy to reduce bleeding risks and corticosteroid exposure. Currently, there is no reliable method to predict the efficacy of rituximab. Our study aimed to develop a machine learning (ML) model to predict the initial response to rituximab in these patients. We analysed data from 156 paediatric ITP patients treated at Beijing Children's Hospital between 2020 and 2023 and identified 25 key predictive features. Among the four evaluated ML models, the multilayer perceptron model exhibited the highest predictive accuracy. SHapley Additive exPlanations analysis revealed that antinuclear antibody titre, thyroglobulin antibody, corticosteroid response and bleeding severity were significant positive predictors, while thyroid peroxidase antibody, CD3+ CD4+ IL-17+ T cells and the duration of disease before rituximab treatment were negatively associated with treatment responses. This ML model could be used to predict rituximab responses in paediatric ITP, which is expected to optimize treatment strategies and improve patient outcomes.
