Validation of claims-based algorithms for identifying selected interstitial lung disease (idiopathic interstitial pneumonias and drug-induced interstitial lung disease).

BACKGROUND

Interstitial lung disease (ILD) is a heterogeneous group of more than 200 diseases that cause fibrosis or inflammation of the pulmonary parenchyma. ILD is a leading cause of death, yet data on its precise incidence and prevalence are limited due to the complexity of diagnostic criteria. To address this gap, we developed and validated claims-based algorithms to identify selected ILD (restricted to idiopathic interstitial pneumonias and drug-induced ILD) in Japan.

METHODS

We identified potential ILD cases from 2010 to 2020 from electronic medical record databases at two large healthcare institutions using a primary claims-based algorithm for high positive predictive value (PPV) and a relaxed algorithm for improved sensitivity. We calculated sensitivity and assessed the validity of the algorithms by comparing them to two gold standard definitions: (1) physician's diagnosis on the medical record, and (2) adjudication by a team of ILD experts based on abstracted medical record data including chest computed tomography images.

RESULTS

Among 7638 potential ILD cases, we sampled 460 patients. The estimated PPV was 87.0% (95% CI 83.2-90.8%) for the primary algorithm and 71.4% (64.0-78.7%) for the relaxed algorithm, based on confirmed ILD responses in the expert adjudication as the gold standard. Estimated sensitivity was 30.4% (27.4-33.8%) and 59.4% (54.1-65.1%) for the primary and relaxed algorithms respectively.

CONCLUSIONS

The algorithms developed in this study may be useful for identifying selected ILD from administrative data in post-marketing database studies and other clinical and epidemiologic research. The appropriate algorithm may be selected based on the specific research objective.