Early in-hospital fasting blood glucose and 90-day functional outcome in early-onset ischaemic stroke: a multicentre prospective cohort study

Background The prognostic value of fasting blood glucose (FBG) in early-onset ischaemic stroke and its variation by prior diabetes status remain uncertain. Methods We analysed 1,127 patients aged 18–49 years from the prospective multicentre CASTOR registry. For study eligibility, acute ischaemic stroke was defined as ischaemic stroke with admission within 7 days of symptom onset; thrombolysed patients were admitted within the thrombolytic time window. FBG was the first fasting venous glucose value from routine post-admission biochemistry; sampling time was unavailable. Unfavourable outcome was defined as a 90-day modified Rankin Scale score ≥3. Missing laboratory data underwent multiple imputation by chained equations. Multivariable logistic regression assessed associations, restricted cubic splines evaluated non-linearity, and interaction with prior diabetes was tested. Incremental predictive value was assessed by comparing models with and without FBG. Internal validation used repeated stratified 70/30 splitting, repeated five-fold cross-validation and bootstrap optimism correction. Results Median age was 45.01 years (IQR 40.91–47.91); 902 patients (80.0%) were male. The median National Institutes of Health Stroke Scale score assessed within 24 h of admission was 2 (IQR 1–4). Unfavourable outcomes occurred in 283 patients (25.1%). Each 1 mmol/L higher FBG was associated with higher adjusted odds of an unfavourable outcome (OR 1.30, 95% CI 1.15–1.46). There was no evidence of non-linearity (pooled Wald p = 0.992) or interaction with prior diabetes (interaction OR 0.87, 95% CI 0.69–1.09; p = 0.220), although only 200 patients had prior diabetes (57 events), limiting power to detect interaction. The FBG association was consistent across sensitivity analyses and internal validation. Adding FBG increased the area under the receiver operating characteristic curve by a median 0.024 (median DeLong p = 0.025) across five imputed datasets, but the event component of continuous net reclassification improvement was negative (−0.035). Conclusion Higher baseline FBG was independently associated with worse 90-day outcome, but its modest, inconsistent incremental predictive value does not establish utility for individual risk stratification. External validation with standardised sampling times is required.