Objective: This multi-site study developed and validated a predictive computable phenotype to identify children with rheumatic disease using on-demand, self-service electronic health record (EHR) reporting tools.Method: Four sites using the same EHR created standardized queries. Inclusion covariates included combinations of diagnosis, age, department (pediatric rheumatology), encounter type (office visit, telemedicine), visit type (new patient, return visit), and care team. The queries were initially derived and tested retrospectively over two weeks, with query results manually compared to clinic schedules by pediatric rheumatologists. The highest-performing query was then validated both prospectively and retrospectively. Query performances are reported using descriptive statistics and logistic regression with clustering by site.Results: Of the 674 patient visits included, 589 (88%) patient visits were prospectively evaluated, 517 (77%) patient visits were retrospectively evaluated, and 433 (64%) patient visits were included in both queries. Prospective queries had a sensitivity of 86%, specificity of 96%, positive predictive value (PPV) of 92%, and negative predictive value (NPV) of 92%, with an area under the curve (AUC) of 0.92. Retrospective queries had a sensitivity of 80%, specificity of 97%, PPV of 95%, and NPV of 87%, with an AUC of 0.91. Age and return visit type, but not female sex, were statistically associated with a rheumatic diagnosis and being present on both queries in univariate analysis. Conclusions: An EHR-based computable phenotype can efficiently and accurately identify eligible patients with rheumatic disease. Such queries can potentially be utilized for pre-visit planning and research recruitment, reducing the need for manual chart review.