Background/Purpose: Idiopathic inflammatory myopathies (IIM; including dermatomyositis, polymyositis, and inclusion body myositis) are heterogenous systemic inflammatory conditions that cause significant disability and morbidity. The study of these rare diseases (prevalence of 10-20/100,000) is limited by our ability to accurately classify cases with a high positive predictive value (PPV). Rule-based algorithms that rely on ICD codes have been used to identify IIM prevalence based on administrative claims and IIM hospitalizations with reported PPVs of 90% or higher. The objective of this study was to apply published rule-based ICD-based algorithms to electronic health record (EHR) data to identify IIM cases to build an IIM cohort.

Methods: Data was extracted from a multihospital EHR system. We identified patients with >1 ICD 9/10 codes for IIM (359.7*, 710.3, 710.4, G72.4*, M33.*, M36.*) between 1/1/2000 – 9/6/2024; from this set, a random sample of 250 patients were selected for chart review. Subjects were classified as having “definite”, “probable”, “possible”, or “unlikely” IIM per the 2017 EULAR/ACR Classification Criteria. ICD codes were extracted for all subjects. The performance of 8 previously published rule-based algorithms for IIM were compared against “definite” cases of IIM from chart review. Algorithms developed using ICD-9 codes were updated to map to ICD-10 codes. Reviews were performed by two independent reviewers (ALV, AV) and discrepancies were adjudicated by a blinded third reviewer (RLC). As the goal was to create an IIM cohort for study, the key performance characteristic of interest was PPV. Specificity, sensitivity, and negative predictive value (NPV) were also calculated.

Results: We identified 5706 patients with >1 IIM ICD 9 or 10 code from the EHR. The prevalence of definite IIM in this cohort was 34.8% (Table 1). M33.* (dermatopolymyositis) was the most common code (35.2%) in the cohort. The highest PPV achieved was 65% with >2 IIM ICD codes in the outpatient setting within 2 months (Algorithm 1) and IIM ICD codes used >2 times between 30 and 365 days (Algorithm 7, Table 2). Algorithm 7 classified 67.8% of definite IIM cases. The algorithm achieving the highest specificity at 94% had low sensitivity (16%) (Algorithm 5, Table 2).

Conclusion: The best performing rule-based IIM algorithms correctly classified 65% of definite IIM cases. The lower performance from prior reports may be due to the different patient populations and lack of direct mapping of ICD-9 codes to ICD-10 codes. Future directions include considering a broader range of EHR data including IIM treatments or data extracted using natural language processing related to IIM characteristics to improve the ability to accurately classify IIM cases at high PPV (≥80%); a crucial first step for IIM studies.

« Back to ACR Convergence 2025

ACR Meeting Abstracts - https://acrabstracts.org/abstract/the-performance-of-rule-based-algorithms-to-identify-patients-with-idiopathic-inflammatory-myopathies-in-electronic-health-records/