In a recent study published in medical frontierResearchers developed and validated a machine learning-based model to improve early detection of rheumatoid arthritis-associated interstitial lung disease (RA-ILD) using routinely available clinical and laboratory data. The goal was to create a practical risk stratification tool that could be applied in primary care and general hospital settings where access to advanced imaging and expert assessment is limited.
Researchers conducted a retrospective analysis of 410 patients with rheumatoid arthritis, of whom 100 (24.39%) had confirmed RA-ILD. After initial data preprocessing, the cohort was divided into training and validation sets. We first screened a broad set of clinical variables using univariate analysis, and variables of potential significance were further refined by LASSO regression to identify the most informative predictors.
Seven key characteristics were ultimately selected to build the predictive model: age, smoking history, lymphocyte count, lactate dehydrogenase, rheumatoid factor, cancer antigen 125 (CA125), and carbohydrate antigen 199 (CA199). We then used these variables to build five different machine learning models: CatBoost, Logistic Regression, Support Vector Machine, Decision Tree, and Random Forest.
Model performance was evaluated using standard classification metrics such as precision, precision, recall, F1 score, and area under the receiver operating characteristic curve (AUC). Among all models, CatBoost showed the strongest overall performance, achieving an AUC of 0.784 (95% CI, 0.656-0.885) and the lowest Brier score of 0.158, indicating good calibration and discrimination. The decision tree model performed similarly, showing the highest recall (0.653) and F1 score (0.603) with AUC 0.783 (95% CI, 0.661-0.818), suggesting high sensitivity in identifying RA-ILD cases.
To improve interpretability, the authors applied SHapley Additive exPlanations (SHAP) analysis to the CatBoost model. This approach identified CA199, CA125, and age as the most influential predictors in determining RA-ILD risk. These findings suggest that tumor-associated markers, along with demographic and inflammatory indicators, may convey meaningful signals about pulmonary involvement in rheumatoid arthritis.
Overall, this study concludes that both CatBoost and decision tree models provide promising approaches for classifying RA-ILD using simple clinical data. Although performance was inconclusive and moderate, the results highlight the potential of machine learning tools to assist with early risk stratification. The authors emphasize that this approach may ultimately support early detection and referral of patients at risk for RA-related lung disease, although external validation in large, independent cohorts is required before clinical application.
