Abstract Objective To develop and validate data-driven algorithms for identifying patients with dermatomyositis (DM) and polymyositis (PM) using Japanese administrative claims data. Methods This multicentre retrospective cross-sectional study included outpatients from six university hospitals between November and December 2023. Administrative claims data covering one year were linked with chart-confirmed diagnoses. Twenty-six candidate variables, including diagnosis codes, laboratory tests, prescriptions, and administrative billing items, were evaluated. Three feature-selection methods were applied to identify relevant predictors. Decision tree analysis was used to construct simplified rule-based algorithms, which were validated in independent internal and external cohorts. Results Among 8,199 training, 3,512 testing, and 1,827 external validation patients, 576 (7.0%), 244 (6.9%), and 136 (4.2%) carried diagnosis codes for DM/PM, of whom 352, 150, and 98 were confirmed as true cases, respectively. The performance of diagnosis codes alone yielded positive predictive values (PPVs) of 0.602 in the testing set and 0.713 in the external validation set. Anti–double-stranded DNA antibody testing, intractable disease management fees, and diagnosis codes for Sjögren’s syndrome were identified as key discriminative variables. The optimal algorithm demonstrated sensitivity of 0.878, specificity of 0.984, PPV of 0.761, and F1-score of 0.815 in the external validation cohort, providing superior accuracy compared to the use of ICD-10 codes alone. Conclusions ICD-10 codes alone were insufficient for accurate identification of DM/PM in Japanese claims data. Integrating laboratory tests and administrative information substantially improved PPV, providing a practical and generalizable framework for claims-based research in Japan.
憲映 et al. (2026) studied this question.