Development and validation of case-finding algorithms for identifying patients with systemic lupus erythematosus in an administrative claim database from tertiary care centers in Japan

Objective To develop and validate algorithms for identifying patients with systemic lupus erythematosus (SLE) in Japanese administrative claims databases from tertiary care centers using statistical and machine learning methods. Methods This retrospective cross-sectional study included 13,538 patients from six hospitals. One-year claims data were linked to chart-confirmed SLE diagnoses. Patients were randomly assigned to training (n = 8,811) and test (n = 3,775) sets; an external validation set (n = 952) was drawn from another hospital. Feature selection used Least Absolute Shrinkage and Selection Operator (LASSO), Boruta, and Recursive Feature Elimination. Logistic regression, random forest, and decision tree models were trained with synthetic oversampling to address class imbalance. Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUROC), and other standard performance metrics. Results The random forest model achieved the best performance (AUROC: 0.995; sensitivity: 0.971; specificity: 0.969). A simplified rule based on diagnosis code and anti–double-stranded DNA antibody testing showed high accuracy in both test and validation sets. Adding urine sediment examination modestly improved sensitivity but reduced specificity. Conclusion A claims-based algorithm incorporating diagnosis codes and standard laboratory tests accurately identified patients with SLE facilitating reliable use of administrative data in real-world research.