Comparison of Rule-Based Algorithms to Identify Patients With Idiopathic Inflammatory Myopathies in Electronic Health Records.

Abstract

Objective: Studying rare diseases requires assembling robust, correctly classified cohorts. We compared the performance of seven published International Classification of Diseases, Ninth Revision (ICD-9) and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) code rule-based algorithms in the electronic health record (EHR) to identify cases of idiopathic inflammatory myopathies (IIM).

Methods: We identified patients seen at a multihospital academic medical center with one or more ICD-9 or ICD-10 codes for IIM (2000-2024). From this cohort, a random sample of 250 cases was selected for chart review for IIM based on the 2017 EULAR/American College of Rheumatology IIM classification criteria. Review was performed by two reviewers and adjudicated by a third. The performance of the algorithms was compared against (1) definite and (2) definite and probable cases of IIM. Positive predictive value (PPV), specificity, sensitivity, and negative predictive value were calculated.

Results: Of 250 charts, 87 definite and 28 probable IIM cases were identified. The algorithm with the highest PPV (84%) relied on ICD 9/10 codes obtained during hospitalizations but excluded ≥80% of cases with a sensitivity of 14%. Use of two or more ICD codes ≤60 days apart obtained the second highest PPV (79%) and identified 82 cases. Use of two or more ICD codes between 30 and 365 days apart obtained the third highest PPV (78%). These top-performing algorithms included definite and probable cases of IIM.

Conclusion: Several rule-based ICD code algorithms had PPVs ranging from 77% to 84% but had lower sensitivity (14%-63%). Limitations of this study include modifications to some of the original algorithms. Future algorithms may benefit from incorporating additional EHR data and natural language processing methods.