Higher fasting glucose in hospital tied to worse recovery in young stroke patients

Young adults who arrive at hospital with an ischaemic stroke and show higher fasting blood glucose on routine early testing face substantially worse odds of independent recovery three months later, according to a prospective multicentre cohort study published in Frontiers in Neurology. Among 1,127 Chinese patients aged 18 to 49 years enrolled in the CASTOR registry, each 1 mmol/L increase in the first in-hospital fasting glucose value was associated with 30% higher adjusted odds of an unfavourable 90-day outcome, defined as a modified Rankin Scale score of 3 or higher, meaning moderate disability requiring some help with daily activities.

The study, led by Yanfei Pan and colleagues at Peking University First Hospital on behalf of the CASTOR Investigators, analysed patients admitted to 80 hospitals across China between May 2015 and October 2017. Participants were eligible if they had acute ischaemic stroke and were admitted within seven days of symptom onset, or within the thrombolytic time window if they received clot-dissolving treatment. The cohort was overwhelmingly male at 80.0%, with a median age of 45.01 years. Most strokes were mild: the median National Institutes of Health Stroke Scale score, assessed within 24 hours of admission, was just 2 on a scale where higher numbers indicate more severe deficits. Even so, 283 patients, or 25.1%, had an unfavourable outcome at 90 days.

Fasting blood glucose was not measured under a research protocol. It was the first fasting venous glucose value drawn as part of routine post-admission biochemistry, typically on the first morning after admission following an overnight fast, although fasting status was determined by routine clinical care rather than verified by the study. Crucially, the registry recorded the glucose value but not the clock time of sampling, so the interval from stroke onset to measurement could not be calculated. Because patients who did not receive thrombolysis could be enrolled up to seven days after onset, this timing varied considerably between patients, and in-hospital glucose-lowering treatment may have preceded the blood draw in some cases. The authors therefore describe the exposure as an early in-hospital fasting glucose rather than a standardised measurement at a fixed interval after stroke.

After adjusting for stroke severity, level of consciousness, prior diabetes, triglycerides, intensive care admission, pre-stroke aspirin use, post-stroke treatment with human urinary kallidinogenase and acute thrombolysis, the association between glucose and outcome remained robust. The odds ratio of 1.30 per 1 mmol/L carried a 95% confidence interval of 1.15 to 1.46. A four-knot restricted cubic spline analysis found no departure from linearity, with a pooled Wald test p value of 0.992, meaning the risk rose steadily across the observed glucose range without any threshold or inflection point. Median fasting glucose was higher among patients with unfavourable outcomes, 5.72 mmol/L versus 5.30 mmol/L in those who recovered well.

A key question was whether pre-existing diabetes changed the picture. It did not, at least not detectably. An interaction model including a glucose-by-diabetes term produced an interaction odds ratio of 0.87 with a 95% confidence interval of 0.69 to 1.09 and a p value of 0.220. However, the authors caution that only 200 patients had prior diabetes and only 57 of them experienced an unfavourable outcome, so the test had limited statistical power. The confidence interval remains compatible with a moderate weakening of the glucose-outcome link in diabetic patients, and clinically relevant effect modification cannot be excluded. Complicating matters further, 109 of the 200 patients with diabetes had no recorded fasting glucose value at all, so the subgroup estimate rests heavily on imputed data and should be treated as exploratory.

Missing data posed a substantial challenge. Fasting glucose was unavailable in 28.5% of patients, the highest missingness of any laboratory variable, and its absence was not random. Patients without a recorded value were far more likely to have prior diabetes and to have received post-stroke glucose-lowering therapy, plausibly because their glucose was monitored by point-of-care or non-fasting methods. Importantly, availability was unrelated to stroke severity, consciousness, intensive care admission or the 90-day outcome itself. The researchers used multiple imputation by chained equations across five datasets, and the association held up in sensitivity analyses including a 30-dataset re-imputation, a complete-case analysis of 730 patients, and alternative adjustment sets, with odds ratios ranging from 1.29 to 1.38.

Whether the finding translates into practical risk prediction is a different matter. Adding fasting glucose to a model already containing stroke severity, consciousness level and the other prespecified covariates raised the area under the receiver operating characteristic curve by a median of only 0.024, from 0.727 to 0.748, with a median DeLong p value of 0.025. The continuous net reclassification improvement of 0.306 looked more impressive on the surface but was driven entirely by the non-event component of 0.334; the event component was actually negative at -0.035, meaning the model ranked predicted risk slightly worse among patients who went on to have unfavourable outcomes once glucose was added. The authors conclude that fasting glucose is a readily available prognostic correlate but not a validated tool for individual risk stratification, and they call for external validation with standardised sampling times.

The observational design also limits what can be inferred about treatment. The authors explicitly warn that the absence of non-linearity should not be read as evidence that lowering glucose will improve recovery. Randomised trials, including SHINE and GIST-UK, have not demonstrated improved functional outcomes with intensive glucose control after ischaemic stroke, and current American Heart Association guidance treats inpatient glucose management as supportive rather than outcome-modifying. Higher glucose may reflect acute metabolic stress, stroke severity, impaired tissue metabolism and underlying vulnerability rather than a direct causal driver of disability, and residual confounding cannot be excluded, particularly because smoking, alcohol use and body mass index could not be adequately incorporated into the analysis.

Internal validation strengthened confidence in the association itself. Across 100 repeated stratified 70/30 train-validation splits, the median validation odds ratio for fasting glucose was 1.30, and the odds ratio exceeded 1 in 99.6% of validation models. Twenty repetitions of five-fold cross-validation and 500 bootstrap resamples per imputed dataset produced similarly stable estimates, with all bootstrap odds ratios greater than 1. Calibration of the full model was adequate, with a median Hosmer-Lemeshow p value of 0.773.

For young stroke survivors, the message is nuanced. Early-onset ischaemic stroke is increasingly recognised as a growing clinical and public health problem, and its risk factor profile only partially overlaps with that of stroke in later life. This study extends observations from older and unselected stroke populations to a younger group in whom prognostic data on acute glycaemia have been limited and heterogeneous. A routine morning blood test already performed on nearly every hospitalised stroke patient appears to carry independent prognostic information, but its modest and inconsistent incremental predictive value means it does not yet justify changes to individual risk assessment. The findings generalise with caution beyond hospitalised Chinese adults aged 18 to 49 treated in comparable routine stroke-care settings, and the authors emphasise that external validation with standardised sampling times is required before any clinical application.

Subject of Research: Prognostic association between early in-hospital fasting blood glucose and 90-day functional outcome in young adults with acute ischaemic stroke.

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

Article References: Pan, Y., Peng, Q., Wei, L., Sun, W., Liu, R., Jin, H., Sun, W., Xing, H., Huang, Y., & on behalf of the CASTOR Investigators (2026). Early in-hospital fasting blood glucose and 90-day functional outcome in early-onset ischaemic stroke: a multicentre prospective cohort study. Frontiers in Neurology, 17, Article 1895654. https://doi.org/10.3389/fneur.2026.1895654

Image Credits: AI Generated

DOI: 10.3389/fneur.2026.1895654

Keywords: early-onset ischaemic stroke, fasting blood glucose, functional outcome, young adults, prognosis, CASTOR registry, modified Rankin Scale, diabetes, risk prediction, multicentre cohort study, hyperglycaemia, stroke recovery