Clinical and pulmonary function-based identification of HRCT-confirmed rheumatoid arthritis-associated interstitial lung disease.

objectiveTo evaluate a pragmatic clinical and pulmonary function-based model for identifying high-resolution computed tomography (HRCT)-confirmed rheumatoid arthritis-associated interstitial lung disease (RA-ILD) in RA patients selected for chest CT evaluation and to explore, as a secondary aim, whether lung ultrasound adds diagnostic information in a selected subset.

methodsThis retrospective diagnostic accuracy study screened rheumatoid arthritis (RA) patients who had undergone chest HRCT or thin-section computed tomography. HRCT-confirmed RA-ILD was the reference outcome. Candidate predictors were prespecified according to clinical plausibility and prior RA-ILD literature, without automated variable selection or univariable P-value screening. The primary model combined clinical variables with forced vital capacity (FVC) and diffusing capacity for carbon monoxide (DLCO). Model performance was assessed using the area under the receiver operating characteristic curve (AUC), diagnostic accuracy indices, calibration, decision curve analysis, bootstrap internal validation with 1000 resamples, and additional sensitivity analyses. The lung ultrasound-enhanced model was considered exploratory.

resultsOf 554 screened RA patients with chest HRCT or thin-section CT records, 486 were included in the final HRCT analysis cohort, of whom 149 had HRCT-confirmed RA-ILD. The PFT analysis cohort included 273 patients, with 96 RA-ILD cases. In the PFT cohort, the clinical plus PFT model showed good discrimination for HRCT-confirmed RA-ILD (AUC 0.860, 95% CI 0.814-0.904), with sensitivity 79.2%, specificity 77.4%, positive predictive value 65.5%, and negative predictive value 87.3% at an exploratory Youden-derived threshold. Bootstrap internal validation with 1000 resamples yielded an optimism-corrected AUC of 0.844 and a median calibration slope of 0.899. Within the same PFT cohort, adding PFT variables improved the AUC from 0.805 to 0.860 (ΔAUC 0.055, DeLong P = 0.005). In the selected PFT plus lung ultrasound subset, adding lung ultrasound to the clinical plus PFT model did not significantly improve AUC (ΔAUC 0.036, P = 0.102), although the exploratory lung ultrasound-enhanced model had a high apparent AUC of 0.913.

conclusionIn RA patients selected for chest CT evaluation, a model combining routine clinical variables with FVC and DLCO identified HRCT-confirmed RA-ILD with good internal performance. Lung ultrasound findings should be interpreted as exploratory because of subset selection, limited sample size, and non-significant incremental AUC gain. External validation in unselected RA cohorts is required before the model can be used to guide HRCT referral decisions. Key Points • A clinical plus pulmonary function model showed good discrimination for HRCT-confirmed RA-ILD in a real-world CT-evaluated RA cohort, but the findings should not be generalized to unselected RA populations without external validation. • Pulmonary function variables improved diagnostic performance beyond clinical variables alone within the same PFT cohort. • The lung ultrasound-enhanced model showed high apparent performance but should be interpreted as exploratory because the LUS subset was selected and the incremental AUC was not statistically significant. • The findings support pragmatic risk stratification within CT-evaluated or higher-risk RA patients; lower-prevalence recalibration showed attenuated positive predictive value and net benefit in general RA screening scenarios.