Detection of rheumatoid arthritis-associated interstitial lung disease: a systematic review and meta-analysis.

BACKGROUND: Rheumatoid arthritis-associated interstitial lung disease (RA-ILD) often has an insidious onset with few or no respiratory symptoms, so early disease may be overlooked. Timely diagnosis and monitoring are therefore crucial. High-resolution computed tomography (HRCT) is the reference standard for RA-ILD, but cost and radiation limit its use as a routine screening tool. Several lower-cost modalities-such as serum biomarkers, machine learning models, and lung ultrasound (LUS)-have been investigated, but their diagnostic value has not been systematically appraised. OBJECTIVE: To evaluate the accuracy of serum biomarkers, LUS, and biomarker-based prediction models, most of which were based on logistic regression, for early RA-ILD diagnosis, and to explore their roles as adjunctive screening tools in HRCT-based diagnosis. METHODS: We systematically searched PubMed, the Cochrane Library, Embase and Web of Science for studies assessing the diagnostic accuracy of serum biomarkers, machine learning models, or LUS in RA-ILD. Risk of bias was assessed using QUADAS-2. A bivariate mixed-effects model was used to pool sensitivity, specificity and construct summary receiver operating characteristic (SROC) curves. RESULTS: Twenty-six studies involving 4,544 participants were included. Among individual biomarkers, KL-6 showed pooled sensitivity and specificity of 0.82 (95% CI 0.68-0.91) and 0.82 (95% CI 0.71-0.89), respectively. LUS showed pooled sensitivity and specificity of 0.96 (95% CI 0.72-0.99) and 0.97 (95% CI 0.66-1.00), respectively. Biomarker-based prediction models, most of which were based on logistic regression, showed pooled AUCs of 0.837 (95% CI 0.795-0.880) in training cohorts and 0.815 (95% CI 0.759-0.871) in validation cohorts. However, these findings should be interpreted cautiously due to variations in study design, diagnostic thresholds, reference standards, and risk of bias, and a lack of sufficient external validation for several prediction models. CONCLUSIONS: Serum biomarkers, biomarker-based prediction models, and LUS may provide useful auxiliary information for early RA-ILD detection. KL-6 demonstrated the most consistent diagnostic performance among individual biomarkers. The high pooled estimates for LUS should be interpreted cautiously because they were derived from a limited number of studies with heterogeneity. At present, these tools should be regarded as adjunctive tools for risk stratification and triage rather than replacements for HRCT. They may help identify patients with RA who are more likely to require confirmatory HRCT and closer follow-up.