Utilizing electronic health record pre-consultation data to create a predictive algorithm for diagnosis of chronic pediatric rheumatic conditions

Abstract

Objectives: To develop a predictive algorithm for diagnosing chronic pediatric rheumatic conditions using patient-reported, historical, and referral data in the electronic health record (EHR) to address current lengthy consultation wait times. Methods: All new rheumatology patient evaluations from 2021 to 2023 were retrospectively reviewed to identify the reason for the visit, patient-recorded outcomes, and international classification of disease codes. The data sample was randomly split into 80% derivation and 20% validation sets. Logistic regression evaluated the association of diagnosis and referral data; variables with p < 0.2 in univariate were included in a multivariate model. Complete data are reported. Results: Of the 3139 subjects, 2064 (66%) were female, with a median age of 13 [IQR 8, 16]. Patients diagnosed with inflammatory arthritis numbered 319 (10%), while 55 (2%) were diagnosed with systemic lupus erythematosus (SLE). The median time from the first visit to diagnosing inflammatory arthritis and SLE was 88 days [35, 210] and 42 [17, 132], respectively. In univariate analysis, a referral reason for swelling was positively associated with a new inflammatory arthritis diagnosis. In contrast, antinuclear antibody positivity, rash, and lupus were positively associated with a new SLE diagnosis. Referral data had low sensitivity and high specificity for both inflammatory arthritis and SLE diagnoses, with areas under the curve of 0.59 and 0.65, respectively. Conclusion: Utilizing the EHR to create a predictive algorithm for chronic rheumatic disease presents a promising solution to existing patient care challenges. This approach suggests that integrating such models to the referral process could help expedite access to pediatric rheumatology services. (Table presented.)