Development and Validation of Nosocomial Bacterial Infection Prediction Models for Patients With Systemic Lupus Erythematosus

Abstract

Background Hospital-acquired bacterial infections are associated with high morbidity and mortality rates in patients with systemic lupus erythematosus (SLE). This study aimed to develop and validate predictive models for the risk of hospital-acquired bacterial infections in patients with SLE. Methods A historical cohort study was designed for development, and another bidirectional cohort study was used for external validation. The risk of bacterial infection was assessed upon admission and after 5 days of hospitalization. Predictor selection employed the least absolute shrinkage and selection operator (LASSO) techniques. Multiple imputations were used to handle missing data. Logistic regression models were applied, and the properties of discrimination, calibration, and decision curve analysis were evaluated. Results The development cohort comprised 1686 patients and 237 events (14.1%) from 3 tertiary hospitals. The external validation cohort included 531 patients and 84 infection outcomes (15.8%) from 10 hospital centers in Colombia (secondary and tertiary level). The models applied at admission and after 120 hours of stay exhibited good discrimination (AUC > 0.74). External validation demonstrated good performance among patients from the same tertiary institutions where the models were developed. However, geographic validation at other institutions has been suboptimal. Conclusions Two predictive models for nosocomial bacterial infections in patients with SLE are presented. All infection prevention recommendations should be maximized in patients at moderate/high risk. Further validation studies in diverse contexts, as well as clinical impact trials, are necessary before potential applications in research and clinical care.