Data for this retrospective cohort study were obtained from the Medical Information Mart for Intensive Care (MIMIC) database from 2001 to 2019. MIMIC is a large, single-center database of de-identified hospitalization-related information for patients admitted to the intensive care unit (ICU) at Beth Israel Deaconess Medical Center [15, 16]. MIMIC database contains patient demographics, laboratory test results, vital sign measurements, procedures, medications, medical history, and mortality data. The inclusion criteria for patients were as follows: (1) patients aged ≥ 18 years old; (2) patients already diagnosed with AIS on ICU admission; (3) patients hospitalized in an ICU for at least 24 h; and (4) patients with repeated glucose measurements (≥ 2) within 24 h of ICU admission. Patients with missing survival information were excluded. AIS was determined from the International Classification of Diseases, ninth/tenth revision (ICD-9/10) codes [ICD-9: 43301, 43311, 43321, 43331, 43381, 43391, 43401, 43411, 43491; ICD-10: I63xxx) in the MIMIC database. For patients with multiple hospitalization records, data were collected only for the patient’s first ICU admission. The requirement of ethical approval for this was waived by the Institutional Review Board of Shanxi Provincial People’s Hospital, because the data was accessed from MIMIC database (a publicly available database). The need for written informed consent was waived by the Institutional Review Board of Shanxi Provincial People’s Hospital due to retrospective nature of the study. All methods were performed in accordance with the relevant guidelines and regulations.

The outcome of this study was 30-day mortality, which occurred within 30 days of the patient’s admission to the ICU. The follow-up period was from the time the patient was admitted to the ICU to the subsequent 30 days or mortality during this period.

The exposures in this study were blood glucose levels at ICU admission and the blood glucose level trajectories within 24 h of ICU admission. Current stroke management guidelines categorized patients’ blood glucose levels into 3 groups: normoglycemia (< 140 mg/dL), moderate hyperglycemia (140–180 mg/dL), and severe hyperglycemia (≥ 180 mg/dL) [17]. Therefore, when blood glucose levels were analyzed as a categorical variable, the blood glucose levels in this study were classified into 3 categories (< 140 mg/dL, 140–180 mg/dL, and ≥ 180 mg/dL). The latent growth mixture modeling (LGMM) was utilized to classify blood glucose level trajectories. The LGMM assumes that the population consists of multiple potential categories, each with similar trajectories and characteristics [18]. A key factor in generating LGMM is determining the number of potential categories. The number of suitable LGMM categories should satisfy that the Akaike Information Criterion (AIC) and Bayesian Information Criteria (BIC) are as small as possible [19], the Entropy needs to be greater than 0.7, the minimum share of each category should not be less than 1%, and the average value of the posterior probability in each category needs to be greater than 70%. After screening, 4 categories of blood glucose level trajectories were the most suitable in this study (Supplement Tables 1 and 2).

The selection of covariates was based primarily on previous studies of ischemic stroke patients admitted to the ICU [20, 21]. Patient characteristics were collected including age, gender (female, male), race (White, Black, other, unknown), admission type (neuro ICU, cardiac ICU, surgical ICU, others), heart rate, systolic blood pressure (SBP), diastolic blood pressure (DBP), mean arterial pressure (MAP), respiratory rate, temperature, sepsis (no, yes), cardiogenic shock (no, yes), diabetes (no, yes), anemia (no, yes), atrial fibrillation (no, yes), Simplified Acute Physiology Score II (SAPS II), Sequential Organ Failure Assessment (SOFA), Glasgow Coma Scale (GCS) (< 15, ≥ 15), Charlson comorbidity index (CCI), oxyhemoglobin saturation (SPO2), white blood cells (WBC), platelet, hemoglobin, red blood cell distribution width (RDW), hematocrit, blood creatinine, blood urea nitrogen (BUN), magnesium levels, international normalized ration (INR), prothrombin time (PT), anion gap, urine output, ventilation (no, yes), vasopressor (no, yes), anticoagulants (no, yes), antiplatelet agents (no, yes), statins (no, yes), insulin (no, yes), thrombectomy (no, yes), and thrombolysis (no, yes). Urine output was defined as the sum of urinary output within 24 h of admission to the ICU.

Skewness and kurtosis methods were used to assess the normality of continuous variables. Continuous variables were described as the mean ± standard deviation (SD) or median and quartile [M (Q1, Q3)], and categorical variables were described as numbers and percentages [n (%)]. The ANOVA or Welch ANOVA test or Kruskal-Wallis H test was used for comparison between groups of continuous variables, and the Chi-square test or Fisher’s exact test was used for comparison between groups of categorical variables. Variables with more than 10% of missing values were excluded (Supplement Table 3), and missing values for the remaining variables were imputed using the multiple imputation method (Supplement Table 4).

Univariable Cox regression analysis was applied to screen for confounders related to 30-day mortality, and variables with P < 0.1 were adjusted in multivariable Cox regression analysis. After screening, multivariable Cox regression analysis adjusted for age, gender, race, admission type, respiratory rate, temperature, SOFA, CCI, platelet, anemia, RDW, BUN, anion gap, urine output, anticoagulants, statins, thrombectomy, atrial fibrillation, and thrombolysis (Supplement Table 5). Because of the important effect of anemia and thrombolysis on AIS [21, 22], anemia and thrombolysis were adjusted in the multivariable model in addition to variables with P < 0.1. Univariable and multivariable Cox regression analyses were applied to examine the relationship between blood glucose levels at ICU admission and blood glucose level trajectories and the risk of 30-day mortality in patients with AIS. Hazard ratio (HR) and 95% confidence interval (CI) were used to report relationships. Subgroup analysis was performed based on age (< 65, ≥ 65 years), gender (female, male), diabetes (no, yes), and insulin use (no, yes). Statistical analyses were performed using R 4.2.3 software (Institute for Statistics and Mathematics, Vienna, Austria), and P < 0.05 was considered statistically significant.