Rheumatoid arthritis-associated interstitial lung disease (RA-ILD) is a common and severe extra-articular manifestation of rheumatoid arthritis, which can significantly shorten patients' life expectancy, severely impair their quality of life, and impose an additional physical, psychological, and economic disease burden on both patients and their families. Clinically, effectively assessing the prognosis of RA-ILD patients is crucial for optimizing individualized clinical decision-making, formulating reasonable treatment plans, and improving patient outcomes, yet it poses considerable challenges due to the heterogeneity of the disease and the lack of reliable prognostic markers. This study aims to accurately predict the survival prognosis of RA-ILD patients by integrating clinical features with radiomics, so as to provide a scientific and practical tool for clinical prognosis assessment.

A retrospective study was conducted on RA-ILD patients who met the inclusion and exclusion criteria. Clinical data of all patients were collected, and univariate and multivariate Cox regression analyses were used to screen clinical risk factors associated with overall survival (OS). For radiomic feature extraction and selection, high-quality imaging data of patients were processed, and optimal radiomic features were selected via three complementary methods: variance thresholding to eliminate redundant features with low variance, univariate feature selection to screen features closely related to OS, and least absolute shrinkage and selection operator (LASSO)-Cox regression to further reduce feature dimensionality and avoid overfitting. Subsequently, three prediction models—clinical model, radiomic model, and clinical-radiomic integrated model—were constructed respectively to predict the OS of RA-ILD patients. A visual nomogram was developed based on the integrated model to intuitively predict the 5-year survival probability of individual RA-ILD patients. The performance of the three models in predicting survival prognosis was evaluated using the consistency index (C-index), with a higher C-index indicating better predictive performance. Decision curve analysis (DCA) was employed to assess the clinical utility of the models by evaluating the net benefit of each model at different threshold probabilities, while calibration curves were used to measure the calibration accuracy between the predicted survival probability and the actual survival outcome. Meanwhile, patients were divided into low-risk and high-risk groups based on the optimal threshold of the Rad-score calculated from radiomic features. Kaplan-Meier survival curves were then plotted for the two groups, and log-rank tests were conducted to evaluate two key points: 1) the significant differences in survival curves between the low-risk and high-risk groups, and 2) the differences in survival curves between the two risk groups in both usual interstitial pneumonia (UIP) pattern and non-specific interstitial pneumonia (NSIP) pattern subgroups.

A total of eligible RA-ILD patients were included in this study, and all patients were followed up for a certain period. Univariate and multivariate Cox regression analyses revealed that patient age (HR: 1.046 [95%CI, 1.013–1.180], P=0.006) and lymphocyte count (HR: 1.385 [95%CI, 1.059–1.812], P=0.018) were independent risk factors for predicting OS, based on which the clinical model was constructed. After a series of feature selection steps, 24 optimal radiomic features with high stability and strong correlation with OS were retained for building the radiomic model. The optimal threshold of the Rad-score was determined to be 0.15, and survival analysis showed that the OS of the high-risk group was significantly lower than that of the low-risk group (P<0.05). Among the three constructed models, the integrated model exhibited the best performance in predicting OS, with C-indices of 0.832 (95%CI: 0.773–0.886), 0.816 (95%CI: 0.695–0.937), and 0.812 (95%CI: 0.659–0.923) in the training set, internal validation set, and external validation set, respectively. Calibration curves showed that the predicted curve of the nomogram for 5-year survival probability was close to the reference line, indicating a high degree of model calibration and good consistency between predicted and actual outcomes. Decision curve analysis (DCA) demonstrated that the integrated model achieved higher net benefit across the three datasets, confirming its good clinical application value.

The clinical-radiomic integrated model (nomogram) constructed in this study by combining age, lymphocyte count, and Rad-score exhibited favorable predictive performance in forecasting OS and 5-year survival probability of patients with RA-ILD. This model not only has high predictive accuracy and good calibration but also has strong clinical utility, which is conducive to clinicians' accurate assessment of patient prognosis, formulation of personalized treatment strategies, and improvement of patient survival and quality of life. It provides a new and reliable tool for the prognosis assessment of RA-ILD in rheumatology clinical practice and is worthy of further promotion and application in clinical work.