Artificial Intelligence and Machine Learning in the Context of Hemophilia: A Scope Review
Tatyane Oliveira Rebouças1,2, André Cardoso Tavares2, Antônio Diego Costa2
1Center For Hematology and Hemotherapy of Ceará, HEMOCE, Fortaleza, Brazil.
Introduction:
Advances in hemophilia treatment have improved life expectancy, yet challenges remain in disease management, predicting clinical complications, and optimizing healthcare resources. Artificial intelligence (AI) and machine learning (ML) have emerged as promising tools to support clinical decision-making and personalized care.
Objectives:
To map the available evidence on AI and ML applications in hemophilia and summarize their contributions to clinical practice.
Methods:
A scoping review was conducted following the Joanna Briggs Institute methodology and reported according to PRISMA-ScR guidelines. Electronic databases were searched for studies published between January 2014 and December 2024. Two independent reviewers selected studies using predefined eligibility criteria.
Results:
A total of 904 records were identified, with 98 duplicates removed. After title and abstract screening, 17 studies underwent full-text assessment. Eight were unavailable in full text, and five did not meet the criteria, resulting in four included studies. AI/ML applications included identification of hemophilia A cases in administrative databases, ultrasound-assisted diagnosis of hemarthrosis and synovitis, prediction of disease severity based on F8 gene mutations, and prediction of adverse clinical outcomes. Frequently used algorithms were Random Forest, support vector machines (SVM), logistic regression, convolutional neural networks, XGBoost, and CatBoost.
Conclusion:
Diagnostic models showed high predictive performance, whereas models for adverse outcomes demonstrated limited generalizability due to small and heterogeneous datasets. AI/ML have substantial potential to improve diagnosis, risk stratification, and personalized management in hemophilia. However, methodological heterogeneity, limited external validation, and the scarcity of studies, particularly in low- and middle-income countries, remain important barriers to routine clinical implementation.
