Associate Professor of Pediatrics and Epidemiology Rutgers University Rutgers University New Brunswick, United States
Background: Juvenile idiopathic arthritis (JIA) is an uncommon, heterogeneous, and potentially disabling chronic condition in children. Prior efforts to validate diagnostic algorithms for JIA within administrative data have been performed in geographically limited, potentially non-generalizable settings (e.g., specific clinics or health care systems).
Objectives: To validate diagnostic algorithms for incident JIA within a large US commercial claims database.
Methods: Using commercial health plan data (2013-2020), we identified children diagnosed with JIA before age 18 following ≥12 months of baseline continuous enrollment without JIA diagnosis or immunosuppression. JIA diagnoses were based on ICD-9-CM (696.0, 714, 720) or ICD-10-CM codes (L40.5, M05, M06, M08, M45) applied to 3 previously validated definitions: 1) diagnosis by a rheumatologist plus orders for ≥2 specific laboratory tests; 2) ≥2 outpatient diagnoses 8-52 weeks apart; 3) 1 inpatient diagnosis. Charts from a random subset of subjects meeting each definition were abstracted and independently adjudicated by clinical experts; discrepancies were resolved by a third expert or, where necessary, consensus. Using probable or definite incident JIA as the reference standard, we estimated the positive predictive value (PPV) of proposed definitions and refined definitions based on rule-based criteria.
Results: Of 92 ICD-9-based charts, 59 (64%) had JIA, 53 (58%) incident JIA. Of 90 ICD-10-based charts, 74 (82%) had JIA, 37 (41%) incident JIA. Among the 3 definitions, definition 1 was most accurate, with PPV for incident JIA 76% (ICD-9) and 51% (ICD-10). Refinement by requiring 3-4 lab orders improved performance: ICD-9, PPV 77-81%, Sensitivity (Se) 82-61%; ICD-10, PPV 81-91%, Se 72-56%. The PPV of definition 2 for incident JIA was 46% (ICD-9) and 36% (ICD-10). Refinement by requiring ≥5 outpatient diagnoses, any JIA treatment, and excluding initial diagnoses by eye doctors produced acceptable performance for ICD-9 (PPV 83%, Se 83%) but not ICD-10 (PPV 53%, Se 89%). In a small sample of inpatient charts (≤5 per set), definition 3 had low PPVs (20-33%).
Conclusions: Compared to ICD-10-based algorithms, ICD-9-based algorithms captured a lower proportion of children with JIA but a higher proportion of children with incident JIA. Best-performing ICD-9- and ICD-10-based algorithms required diagnosis by a rheumatologist and multiple orders for specific laboratory tests, yielding PPVs exceeding 80%. For claims databases without taxonomic data on clinical specialty, clinically informed algorithms for JIA performed adequately based on ICD-9, but not ICD-10, codes. Further algorithmic refinement or quantitative bias analysis may be warranted to use ICD-10-based algorithms for JIA without taxonomic data.
