Multivariate regression model and clinical scoring systems for predicting mortality at ICU admission in patients with ischemic stroke

Patients with stroke comprise 7.7% among those with prolonged and chronic critical illness (PCI/CCI). Prediction of outcomes in this cohort remains highly relevant. Available prognostic tools for patients with acute critical illness may be not so effective for this specific patient population. Objective. To compare prognostic performance of clinical scoring systems and binary logistic regression model based on data within 48 hours after ICU admission in adults with acute ischemic stroke and PCI/CCI. Material and methods. A retrospective single-center study was based on real-world data from the Russian Intensive Care Dataset (RICD) v2.0. A total of 697 adults in acute phase of ischemic stroke (days 4–28) were included. Predictive performance of the model and clinical scores was assessed using the area under the receiver operating characteristic curve (AUROC) and the area under the precision–recall curve (AUPRC) with paired comparisons. Results. ICU mortality was 15.5%. The model showed good predictive performance (AUROC 0.819; AUPRC 0.530) in the overall cohort and significantly outperformed all evaluated clinical scores in terms of AUPRC and most scores in terms of AUROC. The model included predictors reflecting organ dysfunction, systemic inflammation, and coagulopathy. Conclusion. Multivariate binary logistic regression model based on data within 48 hours after ICU admission predicts mortality in patients with ischemic stroke and PCI/CCI more accurately than clinical scoring systems.