First nomogram for predicting interstitial lung disease and pulmonary arterial hypertension in SLE: a machine learning approach.

Abstract

Background: Interstitial lung disease (ILD) and pulmonary arterial hypertension (PAH) are severe, life-threatening complications of systemic lupus erythematosus (SLE). Early identification of high-risk patients remains challenging due to the lack of validated predictive tools. We aimed to develop and validate the first machine learning-based nomogram integrating routine clinical indicators to predict SLE-ILD-PAH risk.

Methods: Using a retrospective cohort design, we analyzed 338 SLE patients (2007-2019), including 193 with ILD-PAH and 145 controls. Univariable and multivariable logistic regression identified independent predictors, followed by nomogram construction and random forest modeling. Model performance was assessed via calibration curves and decision curve analysis (DCA).

Results: Age, C-reactive protein (CRP), anti-dsDNA, pericarditis, and SLE Disease Activity Index (SLEDAI) were independently associated with the prevalence of ILD-PAH in SLE patients. The 5 variables were selected to construct the nomogram model. Calibration curves and decision curve analysis indicated the clinical utility of the nomogram. Receiver operating characteristics (ROC) curves analysis demonstrated excellent discrimination (AUC = 0.871, 95% CI: 0.833-0.910). Forest plot analysis further confirmed the diagnostic weight of each variable.

Conclusions: We developed the first nomogram incorporating age, CRP, anti-dsDNA, pericarditis, and SLEDAI to predict SLE-ILD-PAH risk. This machine learning-enhanced tool leverages routine clinical data, enabling early risk stratification and personalized monitoring. Future studies should validate its utility in guiding therapies and improving outcomes.