Development and validation of data-driven, decision tree-based algorithms for identifying Behçet's disease in claims...

Area of Science:

• Rheumatology
• Medical Informatics
• Data Science

Background:

• Behçet's disease diagnosis in Japan relies on clinical criteria.
• Accurate identification of Behçet's disease patients using administrative data is challenging.
• Existing diagnostic algorithms may lack precision and external validation.

Purpose of the Study:

• To develop and externally validate novel, data-driven algorithms for identifying Behçet's disease patients in Japan.
• To assess the utility of variable selection methods and decision tree models for this purpose.
• To improve the accuracy of claims-based research for Behçet's disease.

Main Methods:

• Retrospective cross-sectional study of 13,538 patients from six tertiary hospitals.
• Claims data linked to chart-confirmed Behçet's disease diagnoses.
• Utilized Least Absolute Shrinkage and Selection Operator, Boruta, and Recursive Feature Elimination for feature selection.
• Developed rule-based algorithms from decision tree models and evaluated diagnostic performance.

Main Results:

• Diagnosis codes alone showed high sensitivity (1.000) and specificity (0.992) but modest positive predictive value (PPV).
• Incorporating prescriptions for sulphamethoxazole-trimethoprim and colchicine significantly improved PPV (0.793 in test set, 0.865 in external validation).
• The developed algorithms maintained high sensitivity and specificity while enhancing PPV.

Conclusions:

• Integrating prescription data with diagnosis codes enhances the accuracy of identifying Behçet's disease patients in claims data.
• A data-driven framework combining variable selection and decision tree analysis offers a validated and scalable approach.
• This methodology supports more reliable claims-based research for Behçet's disease.