Validation of Claims-based Algorithms to Identify Interstitial Lung Disease in Patients with Rheumatoid Arthritis

Abstract

Background/Purpose: The prevalence of interstitial lung disease (ILD) among rheumatoid arthritis (RA) patient ranges between 5-10%. ILD leads to high morbidity and mortality in patients with RA and has economic implications. Validation of ILD among patients with RA in administrative datasets will enable future studies to describe clinical courses and treatment patterns on a large scale using real-world data. Therefore, we aimed to validate claims-based algorithms to identify ILD in RA patients.

Methods: We selected RA patients aged ≥65 years who had ≥2 diagnostic codes for RA and ≥1 disease-modifying antirheumatic drugs in Medicare Parts A/B/D linked to the Partners Healthcare electronic medical records from 2012-to 2014. To identify ILD in RA patients, we evaluated 8 claims-based algorithms using a combination of ICD-9 diagnosis codes and procedure codes related to the diagnosis or management of ILD. The algorithms were: (1) ≥1 diagnosis code for ILD by any physician; (2) ≥2 diagnosis codes for ILD by any physician; (3) ≥1 diagnosis code for ILD by any physician and ≥1 procedure code within 6 months from the ILD code; (4) ≥1 diagnosis code for ILD by any physician and ≥1 diagnostic procedure after the ILD diagnosis code; (5) ≥1 diagnosis code for ILD by pulmonologist or rheumatologist; (6) ≥2 diagnostic codes for ILD by pulmonologist or rheumatologist; (7) ≥1 diagnostic code for ILD by pulmonologist or rheumatologist and ≥1 procedure code within 6 months from the ILD code; (8) ≥1 diagnosis of ILD by pulmonologists or rheumatologist and ≥1 relevant diagnostic procedure after the ILD code. We performed manual medical record review and used presence of ILD on the clinical radiology report of chest computerized tomography (CT) scans or lung biopsy as the gold standard. In ambiguous cases of ILD, three rheumatologists and one pulmonologist discussed the case to determine if ILD was present. We calculated the positive predictive value (PPV) of each algorithm as the percentage of ILD confirmed by the gold standard ILD diagnosis over the denominator of patients who had chest CT or lung biopsy data available.

Results: A total of 5,214 RA patients were included in this study. In each algorithm, 181-993 patients were identified. Of those, 40.9-58.0% had adequate records of chest CT or lung biopsy reports. The PPV of algorithm 1 was 43.4% and requiring more than 2 ILD diagnosis codes improved the PPV to 52.0% in algorithm 2. When procedure codes related with the diagnosis or management for ILD was added to diagnostic code (algorithms 3 and 4), PPV did not increase, although the algorithms considering the time sequence between procedure and diagnosis codes improved the PPV slightly. However, the PPV of algorithm 5 (≥1 diagnosis code by pulmonologist or rheumatologist) increased to 61.5%. Requiring more than 2 ILD diagnostic codes by specialists improved the PPV to 72.4% in algorithm 6 (Table).

Conclusion: Among RA patients, the algorithm that required ≥2 ICD-9 diagnosis codes for ILD by pulmonologist or rheumatologist demonstrated adequate PPV of 72.4% to identify ILD. These results demonstrating support the use of the claims-based algorithm for RA-associated ILD for generating more generalizable real-world evidence.