Hyperglycemia, Ischemic Lesions, and Functional Outcomes After Intracerebral Hemorrhage

Abstract

Background Ischemic lesions observed on diffusion‐weighted imaging (DWI) magnetic resonance imaging are associated with poor outcomes after intracerebral hemorrhage (ICH). We evaluated the association between hyperglycemia, ischemic lesions, and functional outcomes after ICH. Methods and Results This was a retrospective observational analysis of 1167 patients who received magnetic resonance imaging in the ERICH (Ethnic and Racial Variations in Intracerebral Hemorrhage) study. A machine learning strategy using the elastic net regularization and selection procedure was used to perform automated variable selection to identify final multivariable logistic regression models. Sensitivity analyses with alternative model development strategies were performed, and predictive performance was compared. After covariate adjustment, white matter hyperintensity score, leukocyte count on admission, and non‐Hispanic Black race (compared with non‐Hispanic White race) were associated with the presence of DWI lesions. History of ICH and ischemic stroke, presence of DWI lesions, deep ICH location (versus lobar), ICH volume, age, lower Glasgow Coma Score on admission, and medical history of diabetes were associated with poor 6‐month modified Rankin Scale outcome (4–6) after covariate adjustment. Inclusion of interactions between race and ethnicity and variables included in the final multivariable model for functional outcome improved model performance; a significant interaction between race and ethnicity and medical history of diabetes and serum blood glucose on admission was observed. Conclusions No measure of hyperglycemia or diabetes was associated with presence of DWI lesions. However, both medical history of diabetes and presence of DWI lesions were independently associated with poor functional outcomes after ICH.

Keywords

• diabetes
• diffusion‐weighted imaging
• functional outcome
• hyperglycemia
• intracranial hemorrhage
• predictive models