Predicting and validating the risk of interstitial lung disease in systemic lupus erythematosus

Objective

Our study aimed to construct a web-based calculator to predict high risk patients of interstitial lung disease (ILD) in systemic lupus erythematosus (SLE).

Methods

This retrospective study comprised training and test cohorts, including 581 and 86 patients, respectively. Univariate, least absolute shrinkage and selection operator (LASSO), random forest (RF), eXtreme Gradient Boosting (XGBoost), and logistic regression (LR) analyses were performed. A Venn diagram was used to investigate critical features. Receiver operating characteristic (ROC) analysis and decision curve analysis were used to evaluate the model’s performance. Risk stratification was performed using the best ROC cut-off value. The web-based calculator was established using Streamlit software.

Results

Characteristics such as Raynaud’s phenomenon, pulmonary artery systolic pressure, serositis, anti-U1RNP antibodies, anti-Ro52 antibodies, C-reactive protein, age, and disease course were associated with SLE complicated by ILD (SLE-ILD). LR-Venn, RF-Venn, XGBoost-Venn, LASSO-logic, RF, and XGBoost models were constructed. In training cohort, the XGBoost model demonstrated the highest area under the ROC curve (AUC, 0.890; cut-off value, 0.197; sensitivity, 0.793; specificity, 0.836) and provided a net benefit in decision curve analysis (odds ratio [OR] for SLE-ILD [high- vs. low-risk], 19.6). The model was validated in the test cohort (AUC, 0.866; sensitivity, 0.722; specificity, 0.897; OR, 22.7). Furthermore, an XGBoost model-based web calculator was developed.

Conclusion

Our web calculator (https://st-xgboost-app-kcv9qm.streamlit.app/) greatly improved risk prediction for SLE-ILD and was implemented effectively.