Neutrophil-to-albumin ratio as a predictor of sepsis-associated encephalopathy in patients with rheumatoid arthritis: an exploratory machine learning study.

Background and purpose: Patients with rheumatoid arthritis (RA) face an elevated risk of sepsis and its associated neurological complications, most notably sepsis-associated encephalopathy (SAE). Nonetheless, early identification of SAE in this population remains a substantial clinical challenge. To address this gap, we aimed to develop and validate a predictive model that incorporates the neutrophil-to-albumin ratio (NAR) to estimate SAE risk in RA patients with sepsis. Methods: This retrospective multicenter cohort study included a derivation cohort of 89 patients with RA and sepsis from two centers and an independent external validation cohort of 37 patients from a third center. Patients in the derivation cohort were classified into SAE and non-SAE groups. Three machine learning algorithms (LASSO, random forest, and XGBoost) were applied for feature selection, and the optimal model was selected based on area under the receiver operating characteristic curve (AUC). Model performance was evaluated using bootstrap resampling, calibration curves, and decision curve analysis. The final XGBoost model was subsequently evaluated in the external validation cohort. A web-based dynamic prediction tool was developed for clinical application. Kaplan-Meier analysis and Cox regression were performed to evaluate 28-day survival. Results: SAE occurred in 23.6% of patients and was associated with significantly higher 28-day mortality (61.9% vs. 33.8%, Conclusion: The NAR-integrated XGBoost model provides a robust and interpretable tool for early SAE prediction in RA patients with sepsis, showing potential clinical utility for bedside risk stratification.