Development and validation of models for risk of death in patients with systemic lupus erythematosus admitted to the intensive care unit: a retrospective study.

Abstract

For patients with systemic lupus erythematosus (SLE) in the intensive care unit (ICU), early detection of the mortality risk is a significant factor in improving outcomes. In this study, we developed and validated predictive models for the mortality risk. Using MIMIC-III v1.4 and MIMIC-IV v0.4, we collected data of patients with SLE who were admitted to the ICU. The patients were divided into two groups based on death or survival within 30 days. Two prediction models were built for each group: a traditional logistic regression model and a linear discriminant analysis (LDA) model constructed by the random subspace method (RSM) (RSM-LDA model). The performance of the two models was analyzed using the area under the receiver operating characteristic curve (AUC). MIMIC-III and MIMIC-IV were used to establish and validate the models. This study involved 383 patients with SLE, 65 of whom died. They were divided into two groups according to whether they died within 30 days. The predictive factors were the type of admission to the ICU, SLE-associated interstitial pneumonia, lupus nephritis, immunoglobulin G level, and cardiolipin antibody level. A logistic regression model and an RSM-LDA model were established. The AUCs of the two models were 0.87 (95% confidence interval, 0.86-0.90) and 0.91 (95% confidence interval, 0.88-0.93), respectively. The RSM-LDA model can predict the risk of death in patients with SLE admitted to the ICU at an early stage. Key Points • Compared with traditional prediction models, RSM-LDA model has a better ability to predict the risk of death inpatients with systemic lupus erythematosus. • Compared with traditional prediction models, the more input variables (mortality related risk factors), the better the prediction results of RSM-LDA model.