Construction of Rheumatoid Arthritis-Associated Interstitial Lung Disease diagnostic model and identification of...

This study aimed to develop and validate a machine learning model integrating multi-omics and radiomics data to improve diagnostic accuracy and identify potential biomarkers for Rheumatoid Arthritis-Associated Interstitial Lung Disease (RA-ILD).

A total of 278 patients with RA were enrolled across two cohorts. Cohort 1 (63 RA-nonILD, 46 RA-ILD) provided clinical data, chest CT images, plasma, and PBMC samples for non-targeted metabolomics, transcriptomics, and 4D DIA proteomics. Cohort 1 was split in a 6:4 ratio into training and validation sets. Machine-learning algorithms (RF, LASSO, SVM) and a Transformer model were used to screen biomarkers. Diagnostic models were constructed using LASSO, RF, LightGBM, and CatBoost. A combined imaging-clinical logistic regression model was developed and externally validated in cohort 2 (102 RA-nonILD, 67 RA-ILD). Associations between key biomarkers, inflammation, lung function, and CT severity were examined, and pathways related to the radiomic feature Kurtosis were explored.

Nine radiomic features, five metabolites, two proteins, and eight genes were identified as key biomarkers. The metabolomics-based CatBoost model showed the best single-omics performance (AUC = 0.982). The multi-omics integration model outperformed all single-omics models. The imaging-clinical model demonstrated strong diagnostic accuracy in both internal (AUC = 0.963) and external validation (AUC = 0.913), and a nomogram was constructed for clinical risk assessment. Key biomarkers correlated with inflammatory indicators and lung-function decline, and high-Kurtosis-associated genes were enriched in pro-fibrotic pathways.

Integrating multi-omics and radiomics with machine learning yields a robust diagnostic strategy for RA-ILD. The imaging-clinical nomogram provides a practical tool for risk assessment, and identified biomarkers reflect disease severity and progression.