The following is a summary of “Enhancing the identification of rheumatoid arthritis-associated interstitial lung disease through text mining of chest computerized tomography reports,” published in the June 2023 issue of the Seminars in Arthritis and Rheumatism by Luedders et al.

The objective is to identify rheumatoid arthritis-interstitial lung disease (RA-ILD) in administrative data; algorithms with positive predictive values (PPVs) between 70 and 80% have been devised. In this cross-sectional study, the researchers hypothesized that incorporating ILD-related terms extracted from chest computed tomography (CT) reports via text mining would increase the PPV of these algorithms. They identified a derivation cohort of possible RA-ILD cases using electronic health record data from a large academic medical center (n = 114). 

They conducted a medical record review to validate diagnoses (reference standard). Natural language processing identified ILD-related terms (e.g., ground glass, honeycomb) in chest CT reports (e.g., ground glass, honeycomb). Administrative algorithms comprising diagnostic and procedural codes and specialties were applied to the cohort with and without needing ILD-related terms from CT reports. Subsequently, they analyzed comparable algorithms in a validation cohort of 536 individuals with RA. Adding ILD-related terms to RA-ILD administrative algorithms increased PPV in the derivation (improvement varying from 3.6% to 11.7%) and validation (improvement ranging from 6.0 to 21.0%) cohorts. 

This increase was most significant for algorithms with fewer requirements. The PPV of administrative algorithms incorporating ILD-related terms from CT reports exceeded 90% (maximum derivation cohort of 94.6%). As PPV increased, sensitivity decreased (validation cohort: -3.9% to -19.5%). Incorporating ILD-related terms extracted from chest CT reports via text mining improved the PPV of RA-ILD algorithms. Using these algorithms in large data sets with high PPVs could facilitate RA-ILD epidemiologic and comparative effectiveness research.

Source: sciencedirect.com/science/article/abs/pii/S004901722300046X