Validation of an algorithm to identify incident interstitial lung disease in patients with rheumatoid arthritis

M Meehan,F Xie,N Annapureddy,M I Danila,J Zhuo,J Oates,J Lobo,J R Curtis,C Clinton,B R England,A Shah

doi:10.1186/s13075-021-02655-z

Abstract

Background/purposeInterstitial lung disease (ILD) is an important problem for patients with rheumatoid arthritis (RA). However, current approaches to ILD case finding in real-world data have been evaluated only in limited settings and identify only prevalent ILD and not new-onset disease. Our objective was to develop, refine, and validate a claims-based algorithm to identify both prevalent and incident ILD in RA patients compared to the gold standard of medical record review.MethodsWe used administrative claims data 2006–2015 from Medicare to derive a cohort of RA patients. We then identified suspected ILD using variations of ILD algorithms to classify both prevalent and incident ILD based on features of the data that included hospitalization vs. outpatient setting, physician specialty, pulmonary-related diagnosis codes, and exclusions for potentially mimicking pulmonary conditions. Positive predictive values (PPV) of several ILD algorithm variants for both prevalent and incident ILD were evaluated.ResultsWe identified 234 linkable RA patients with sufficient data to evaluate for ILD. Overall, 108 (46.2%) of suspected cases were confirmed as ILD. Most cases (64%) were diagnosed in the outpatient setting. The best performing algorithm for prevalent ILD had a PPV of 77% (95% CI 67–84%) and for incident ILD was 96% (95% CI 85–100%).ConclusionCase finding in administrative data for both prevalent and incident interstitial lung disease in RA patients is feasible and has reasonable accuracy to support population-based research and real-world evidence generation.

Highlights

Rheumatoid arthritis (RA) is associated with a variety of extra-articular pulmonary manifestations, the best characterized being interstitial lung disease (ILD)
Cohort identification and characteristics After applying inclusion and exclusion criteria to the Medicare data and restricting to the five participating medical centers, we identified 578 RA patients receiving any care at the participating locations (51 at Vanderbilt University Medical Center (VUMC), 51 at Medical University of South Carolina (MUSC), 243 at University of North Carolina (UNC), 170 at Duke, and 63 at University of Alabama at Birmingham (UAB))
The 39 cases that did not have sufficient information for clinical adjudication of ILD status were excluded from further analysis, leaving 234 RA patients with sufficient data to be evaluated for ILD (Fig. 1)

Summary

Introduction

Rheumatoid arthritis (RA) is associated with a variety of extra-articular pulmonary manifestations, the best characterized being interstitial lung disease (ILD). Case ascertainment of RA-ILD at a population level has been challenging, and algorithms using administrative databases for the identification of RA-ILD cohorts have previously been applied to large healthcare databases without validation [8,9,10,11,12,13]. A recent study using Veterans Affairs (VA) administrative data created and validated several variants of claimsbased algorithms to identify prevalent RA-ILD compared to the gold standard of medical record review [15]. A second study, conducted in a single-state medical center network, validated an algorithm to identify prevalent RAILD in claims data and showed a PPV as high as 72% but did not evaluate incident ILD [16]. The generalizability of a prevalent ILD algorithm in more diverse environments is unknown

Objectives

Methods

Results

Discussion

Conclusion