Is hyperglycaemia an independent predictor of poor outcome after acute stroke? Results of a long term follow up studyBMJ 1997; 314 doi: https://doi.org/10.1136/bmj.314.7090.1303 (Published 03 May 1997) Cite this as: BMJ 1997;314:1303
- Christopher J Weir, MRC training fellowa,
- Gordon D Murray, reader in medical statisticsb,
- Alexander G Dyker, lecturer in stroke medicinea,
- Kennedy R Lees, clinical director, acute stroke unita
- a Acute Stroke Unit, University Department of Medicine and Therapeutics, Western Infirmary, Glasgow G11 6NT
- b Robertson Centre for Biostatistics, University of Glasgow, Glasgow G12 8QQ
- Correspondence to: Mr Weir
- Accepted 3 February 1997
Objective: To determine whether raised plasma glucose concentration independently influences outcome after acute stroke or is a stress response reflecting increased stroke severity.
Design: Long term follow up study of patients admitted to an acute stroke unit.
Setting: Western Infirmary, Glasgow.
Subjects: 811 patients with acute stroke confirmed by computed tomography. Analysis was restricted to the 750 non-diabetic patients.
Main outcome measures: Survival time and placement three months after stroke.
Results: 645 patients (86%) had ischaemic stroke and 105 patients (14%) haemorrhagic stroke. Cox's proportional hazards modelling with stratification according to Oxfordshire Community Stroke Project categories identified increased age (relative hazard 1.36 per decade; 95% confidence interval 1.21 to 1.53), haemorrhagic stroke (relative hazard 1.67; 1.22 to 2.28), time to resolution of symptoms >72 hours (relative hazard 2.15; 1.15 to 4.05), and hyperglycaemia (relative hazard 1.87; 1.43 to 2.45) as predictors of mortality. The effect of glucose concentration on survival was greatest in the first month.
Conclusions: Plasma glucose concentration above 8 mmol/l after acute stroke predicts a poor prognosis after correcting for age, stroke severity, and stroke subtype. Raised plasma glucose concentration is therefore unlikely to be solely a stress response and should arguably be treated actively. A randomised trial is warranted.
A plasma glucose concentration above 8 mmol/l after acute stroke predicts poorer chances of survival and independence
This effect of raised glucose concentration persists after adjusting for factors known to affect the outcome of stroke–namely, age, stroke type, and stroke severity
A clinical trial of active control of plasma glucose concentration is warranted
Diabetic patients have worse survival and recovery prospects after acute stroke than non-diabetic patients. In addition, hyperglycaemia in the acute phase of stroke has been established as a predictor of poor outcome in non-diabetic patients. There is dispute, however, whether a raised plasma glucose concentration is independently associated with a poor prognosis. Several studies have suggested that hyperglycaemia in non-diabetic patients after acute stroke is a stress response1 2 3 4 5 6 7 8 reflecting more severe neurological damage. Others have suggested that hyperglycaemia influences outcome independently of stroke severity.9 10 11 If the second was true we should need to investigate whether reversing hyperglycaemia in the acute phase of stroke influenced its adverse effect on survival.
We studied the effect of hyperglycaemia on stroke mortality and morbidity by assessing its effect on outcome after adjusting for known prognostic factors. We describe our findings in a cohort of unselected patients admitted to our acute stroke unit.
Patients and methods
The acute stroke unit serves a catchment population of 220 000. All patients who present within 72 hours of the onset of an acute neurological deficit with no known alternative to a vascular cause are admitted irrespective of age or the severity of the deficit. All patients have clinical data and results of investigations recorded prospectively. A diagnosis of ischaemic or haemorrhagic stroke is established by computed tomography. Magnetic resonance imaging is considered as an additional diagnostic tool, particularly in patients with suspected posterior circulation events. The aim is to complete all investigations within 72 hours of admission. All patients have their stroke subtype categorised on the basis of clinical features according to the Oxfordshire Community Stroke Project classification.12 This classification divides patients into four groups: total anterior circulation syndrome, partial anterior circulation syndrome, posterior circulation syndrome, and lacunar syndrome.
Biochemical data are obtained routinely from all patients on the day of admission and early next morning. Plasma glucose concentration is measured on both occasions, giving one random and one fasting measurement. In this study we used the random glucose measurement for each patient if it was taken; if not we used the fasting measurement. Glucose concentration was recorded both as a continuous variable and as a binary variable (≤8 mmol/l, normoglycaemic; >8 mmol/l, hyperglycaemic). The upper limit of normal for a fasting plasma glucose concentration is 6.5 mmol/l. As not all glucose measurements in our study were taken fasting, 8 mmol/l was used as the cut point for hyperglycaemia. Other potential prognostic variables considered were age, stroke type (ischaemic or haemorrhagic), admission blood pressure (systolic and diastolic), smoking status (never smoker, former smoker, or current smoker), time to resolution of symptoms (≤72 hours or >72 hours), and Oxfordshire Community Stroke Project category.
The patients in this study presented to our acute stroke unit between June 1990 and December 1993. Patients with previously diagnosed diabetes were included, but the data from these patients were analysed separately, as there is evidence that hyperglycaemia affects outcome differently in diabetic patients.9
Survival and placement follow up were by record linkage13 to the Scottish deaths register and to a national database of hospital discharge records. The method of record linkage was validated in an epidemiological study of hypertension14 and has been used for monitoring end points and adverse events in a large clinical trial.15 Record linkage provides reliable patient follow up; however, admissions to private hospitals or institutions outside Scotland are not detected. Outcome placement was coded as alive at home, alive in care, or dead. This placement information was recorded two, three, six, and 12 months after admission.
Baseline variables in diabetic and non-diabetic patients were compared by χ2 tests for discrete variables and Mann-Whitney tests for continuous variables. Differences in the distributions of potential prognostic variables between placement categories at three months were assessed by χ2 test for discrete variables and Kruskal-Wallis analysis of variance for continuous variables. The main analysis used Cox's proportional hazards regression model16 to estimate the effect of hyperglycaemia on survival after stroke. A separate baseline survival function was fitted for each Oxfordshire Community Stroke Project category, as including the Oxford classification as an explanatory variable was unlikely to fulfil the proportional hazards assumption.
The effect of plasma glucose concentration was determined after entering other significant prognostic variables (selected from age, stroke type, time to resolution of symptoms, smoking status, and systolic and diastolic blood pressure). The assumption of proportional hazards was checked for all variables included in the model. The effect of hyperglycaemia on outcome was further explored by coding outcome at three months as good (alive at home) or poor (alive in care or dead) and then performing a stepwise logistic regression analysis.17 We tested whether hyperglycaemia was independently associated with this outcome after adjusting as necessary for age, time to resolution of symptoms, stroke subtype, Oxford classification category, smoking status, and systolic and diastolic blood pressure. In the proportional hazards and logistic regression analysis a quadratic relation between blood pressure and outcome was permitted.
A total of 811 patients with computed tomography confirmed acute stroke and plasma glucose data were studied. In 624 (77%) cases the plasma glucose concentration was measured on admission, and in 187 (23%) cases it was measured early next morning. The mean times to measurement of glucose concentration were 3.6 hours after admission to the unit and 14.4 hours after stroke onset. Sixty one (8%) patients were diabetic, seven (1%) being insulin dependent. Table 1) compares the characteristics of the diabetic and non-diabetic patients. As expected, the median plasma glucose concentration and proportion of patients with hyperglycaemia were higher in the diabetic group. Our main analysis was restricted to the 750 non-diabetic patients. Fifteen patients were lost to follow up for placement (owing to failure of hospital discharge record linkage) but not for survival. The mean follow up time was 1.65 years.
Table 2) shows the numbers of patients in each outcome category over time. Table 3) shows the distribution of patient variables across the three placement categories. Table 4) gives the results of proportional hazards modelling.
Hyperglycaemia led to higher mortality, even after adjusting for other prognostic variables. Increased systolic and diastolic blood pressure levels were not significant linear or quadratic predictors of poor survival and were not included in the final proportional hazards model. Similarly, smoking status did not predict survival and was excluded from the model. The assumption of proportional hazards held for all variables except plasma glucose concentration (continuous). This variable was therefore removed from the model and plasma glucose concentration considered as a binary variable. Figure 1) shows the Kaplan-Meier survival curves for patients in each Oxford classification category with and without hyperglycaemia.
Hyperglycaemia also predicted poor outcome at three months. This variable significantly (P=0.0003) improved prediction of outcome at three months (alive at home versus in care or dead) by logistic regression after adjusting for age, time to resolution of symptoms, stroke subtype, and Oxford classification category. Levels of systolic and diastolic blood pressure were not included in the logistic regression model, as they were neither linear nor quadratic predictors of outcome. Smoking status did not predict outcome and was excluded from the model.
Our results show that hyperglycaemia predicts higher mortality and morbidity after acute stroke independently of other adverse prognostic factors, such as older age, type and severity of stroke, and non-reversibility of the neurological deficit. The effect of hyperglycaemia on mortality is large. The estimated relative hazard of 1.87 was greater than that for haemorrhagic versus ischaemic stroke and equivalent to adding more than 20 years to a patient's age.
The results suggest that hyperglycaemia is not solely a stress response to neurological insult, as it predicts outcome after taking other prognostic factors into account. Indeed, the relative risk conferred by hyperglycaemia is greatest in patients with lacunar stroke. Previous studies which concluded that hyperglycaemia was a stress response, based on a correlation between stroke severity and plasma glucose concentration,6 7 did not consider whether hyperglycaemia independently predicted outcome after adjusting for stroke severity. Van Kooten et al showed that noradrenaline concentrations were associated with stroke severity but could not find significant relations between catecholamine and plasma glucose concentration or between glucose concentration and stroke severity.10 They concluded that raised plasma glucose concentration in non-diabetic stroke patients could not be explained by a stress response. Jorgensen et al showed a correlation between glucose concentration and stroke severity but found that glucose concentration independently predicted outcome after adjusting for stroke severity.9
We sought to correct for the blood pressure on admission in our modelling of survival, as raised blood pressure after admission for stroke may be due to both the mental stress of hospitalisation18 19 and the physical stress of the neurological damage. However, neither systolic nor diastolic blood pressure was associated with outcome. In addition, diastolic blood pressure was not significantly correlated with plasma glucose concentration (Spearman's rank correlation coefficient (rs) 0.053; P=0.0819) and systolic blood pressure was only weakly correlated with plasma glucose concentration (rs=0.131; P=0.0003). This further indicates that raised plasma glucose concentration is not due to a stress response after acute stroke.
Duration of hyperglycaemia
Probably in many hyperglycaemic patients in our study the raised plasma glucose concentration was of long standing. Other workers investigated this by measuring haemoglobin A1c concentration and inferred that raised values indicated a long prestroke history of hyperglycaemia.23911 Haemoglobin A1c concentration is not routinely monitored in our unit, so we could not estimate the prevalence of previously undiagnosed diabetes.
The mechanism by which hyperglycaemia might influence stroke outcome is uncertain. Both acute and chronic hyperglycaemia are associated with increased oedema and infarct size20 and with reduced cerebral blood flow and cerebrovascular reserve.21 Ischaemia leads to a slowing of the oxidative glucose metabolism and an increase in anaerobic glycolysis. The concentration of lactic acid increases locally as a result. Hence intracellular pH is lowered and cells die or become dysfunctional.22 Hyperglycaemia exacerbates these changes.23 24 Experimental evidence suggests that hyperglycaemia may increase lactate production in two ways–either directly in severely ischaemic brains by increasing available glucose, or indirectly in the case of incomplete cerebral ischaemia by inhibiting mitochondrial respiration and glucose oxidation.24 Such increased lactate production in the ischaemic penumbra may lead to poorer outcome. The above mechanisms may also cause a worse outcome in hyperglycaemic primary intracerebral haemorrhage, the excess lactate production occurring in the area of ischaemia around the site of the haemorrhage.
Our results suggest that a randomised trial of glucose control is warranted in patients with stroke complicated by hyperglycaemia. Randomisation should be soon enough after stroke onset to allow treatment during the “window of opportunity” for pharmacological intervention. Recent studies suggest that this time window lasts for up to three25 or even 1226 hours after stroke onset.
We thank Chris Povey, of Scottish Record Linkage, NHS Information and Statistics Division, Edinburgh, for record linkage to death and hospital discharge records.
Funding: CJW is supported by a Wellcome Trust prize studentship.
Conflict of interest: None.