[Development and validation of a risk assessment model for interstitial lung disease in patients with rheumatoid arthritis].

OBJECTIVES

To develop and validate a risk assessment model for interstitial lung disease (ILD) in patients with rheumatoid arthritis (RA).

METHODS

A retrospective study analyzed clinical data from 312 patients with RA treated at Guanghua Hospital Affiliated to Shanghai University of Traditional Chinese Medicine between January 1, 2021 and June 30, 2024. Patients were divided into an RA-only group and an RA-ILD group based on the presence of ILD. Demographic characteristics, laboratory test results, clinical manifestations, disease activity scores, medication history, joint X-ray findings, and traditional Chinese medicine (TCM) syndrome types were collected as potential predictors. Variables were screened using univariable analysis and LASSO regression, and binary logistic regression was used to build the model and construct a nomogram using the rms package in R software. The discrimination and clinical utility of this model were assessed via receiver operating characteristic (ROC) and decision curves, and calibration was evaluated using Bootstrap-resampled (1000 iterations) calibration curves. A LightGBM machine learning model was also developed based on the selected predictors, and its performance was evaluated using ROC and precision-recall (PR) curves. Five-fold cross-validation was employed to further assess the robustness of the predictors.

RESULTS

Multivariate logistic regression identified sex, age, anti-cyclic citrullinated peptide (CCP) antibody, disease activity score in 28 joints (DAS28) based on erythrocyte sedimentation rate (ESR), Krebs von den Lungen-6 (KL-6), international normalized ratio (INR), activated partial thromboplastin time (APTT), and methotrexate use as independent predictors of RA-ILD (all P<0.05). The model showed excellent discrimination with an area under the curve (AUC) of 0.976. Calibration curves showed strong agreement between predicted and observed probabilities, with an optimism-corrected slope approaching unity (0.9973). Decision curve analysis demonstrated a high net clinical benefit. The LightGBM model yielded an ROC AUC of 0.8464 and a PR AUC of 0.9417. In five-fold cross-validation, all ROC AUC values were greater than 0.65.

CONCLUSIONS

The developed risk assessment model for ILD in patients with RA indicates high predictive ability and clinical utility.