Impact of pre-stroke sulphonylurea and metformin use on mortality of intracerebral haemorrhage

Abstract

Introduction

Few proven therapies for intracerebral haemorrhage exist. Preliminary observational evidence suggests that sulphonylurea and metformin may be protective in ischaemic stroke. We assessed the association of pre-intracerebral haemorrhage sulphonylurea and metformin use on outcome in diabetic patients.

Methods

We merged datasets from the consecutive single-centre Helsinki ICH Study, the intracerebral haemorrhage arm of the Virtual International Stroke Trials Archive (VISTA-ICH) and the Royal Melbourne Hospital ICH Study. Logistic regression adjusting for known predictors of intracerebral haemorrhage outcome (age, sex, baseline Glasgow Coma Scale, National Institutes of Health Stroke Scale, intracerebral haemorrhage volume, infratentorial location, intraventricular extension, and pre-intracerebral haemorrhage warfarin use) estimated the association of metformin and sulphonylurea with all-cause 90-day mortality.

Results

From a dataset of 2404 consecutive intracerebral haemorrhage patients, we included 374 (16%) patients with diabetes. Of these, 113 (30%) died by 90 days. Metformin was used in 148 (40%) patients and sulphonylurea in 115 (31%) patients at intracerebral haemorrhage onset. After adjusting for baseline characteristics, metformin use was associated with lower 90-day mortality (OR 0.51; 95% CI 0.26–0.97; p = 0.041) irrespective of whether the drug was continued or not during the admission, while sulphonylurea use was not associated with mortality (OR 0.96; 95% CI 0.49–1.88; p = 0.906). Haematoma location or evacuation did not modify the association between metformin and mortality; neither did adding insulin use, baseline glucose and serum creatinine into the model (OR 0.50; 95% CI 0.25–0.99; p = 0.047).

Conclusion

Pre-intracerebral haemorrhage metformin use was associated with improved outcome in diabetic intracerebral haemorrhage patients. Our results generate hypotheses which after further validation could be tested in clinical trials.

Introduction

The only currently proven interventions for intracerebral haemorrhage (ICH) are organised stroke unit care and blood pressure control.¹ In ischaemic stroke, observational studies have reported both neutral and beneficial results of pre-stroke sulphonylurea^2–6 and metformin use.^5,6 A recently completed phase II randomised controlled trial demonstrated less midline shift in patients with large territory ischaemic stroke treated with intravenous glyburide when compared to placebo.⁷

Sulphonylurea inhibits the SUR-1-TRPM4 ion channel, which is up-regulated within several hours of ischaemic stroke and also within the peri-haematomal region after ICH.^8,9 Up-regulated SUR-1-TRPM4 can lead to an unchecked intracellular influx of sodium ion accompanied by chloride and water, resulting in cell swelling and cytotoxic oedema. In a rat subarachnoid haemorrhage model, sulphonylurea reduced inflammation and improved cognitive recovery.⁸ In a mouse ischaemic stroke model, chronic pre-stroke metformin administration reduced lactate production and infarct volume.¹⁰

There are no published reports assessing the association of sulphonylurea or metformin with ICH outcome. We hypothesised that pre-stroke sulphonylurea or metformin use is associated with better 90-day outcome independent of known predictors.

Methods

Study design

We combined datasets from the single-centre Helsinki ICH study,¹¹ the Royal Melbourne Hospital (RMH) ICH study,¹² and the ICH arm of the pooled repository of trials Virtual International Stroke Trials Archive (VISTA).¹³ The Helsinki ICH study is a retrospective analysis of consecutive ICH patients admitted to Helsinki University Central Hospital between January 2005 and March 2010. The RMH ICH study retrospectively analysed the impact of warfarin on location and haematoma volume in ICH patients admitted between October 2007 and January 2012. VISTA is an international academic collaborative repository for stroke clinical trials and collects the data in an anonymised manner to allow for novel exploratory analyses. All included patients had non-traumatic spontaneous ICH. Approval for this analysis was granted by the Melbourne Health Human Research Ethics Committee.

We obtained anonymised patient data for baseline demographics, National Institutes of Health Stroke Scale (NIHSS), Glasgow Coma Scale (GCS), imaging, medications, and 90-day outcome. In the VISTA-ICH database extensive demographic, clinical and outcome information was collected prospectively as part of the individual trials. Anatomical therapeutic chemical (ATC) classification system was used to determine use of metformin (A10BA) and sulphonylurea (A10BB) in the VISTA-ICH dataset. In the Helsinki ICH study, data collection was performed retrospectively by chart review with reconstruction of NIHSS where it was not recorded at time of stroke assessment.¹¹ In the RMH ICH study baseline patient information was recorded as part of the prospective stroke database.¹² Chart review was performed by TYW to reconstruct baseline NIHSS where not pre-recorded, obtain medication information, baseline glucose and creatinine and to assess for 90-day outcome. In both the Helsinki ICH and RMH ICH cohorts, anti-diabetic medication use was determined by documented use of generic or trade drug names in medication charts. Medication discontinuation was defined as a pause of >24 h during the first two days of ICH.

We included all consecutive diabetic patients irrespective of treatment. Non-diabetic patients were excluded from analysis due to confounding by indication for anti-diabetic use. Initial pooling was performed between Helsinki and VISTA-ICH datasets and RMH patients were subsequently added to increase the number of unselected real-life patients and the generalisability of the findings. Diabetes was defined as a pre-existing diagnosis prior to the ICH and not on basis of diabetic medication use. The primary outcome was 90-day mortality, adjusted for pre-defined covariates known to influence ICH outcome: age, gender, NIHSS, GCS, pre-stroke warfarin use, baseline ICH volume, infratentorial location of haemorrhage, and ventricular extension. Patients with missing data on any of these variables were excluded.

Statistical analysis

Data are reported as median with interquartile range or n (%) and analysed with Mann–Whitney U Test, Pearson χ² or Fisher’s exact tests as appropriate. Crude unadjusted survival is presented as Kaplan–Meier curves. The primary outcome of 90-day mortality was analysed with a logistic regression model with all the pre-defined covariates entered into the model. Test for multi-collinearity for the variables included in the logistic regression model demonstrated variance inflation factor of <3 indicating no significant multi-collinearity. Receiver operator characteristic area under the curve (AUC) analysis was performed to assess the regression model fit. Several post-hoc analyses assessing potential confounding factors impacting the regression model were performed including propensity score matching 1:1 to the nearest neighbour with caliper set at 0.2 standard deviations of the logit of the propensity scores. The primary analysis was duplicated in the propensity matched population. We used SPSS 22 (IBM, Armonk, NY, USA). A two-sided p < 0.05 was considered statistically significant. As the analysis is hypothesis generating, we did not adjust for multiple testing.

Results

From a dataset of 2404 consecutive ICH patients, the proportion of diabetic patients was 148/1013 (15%) in Helsinki, 87/404 (22%) in RMH, and 173/987 (18%) in VISTA-ICH. After excluding 34 cases with missing pre-defined covariates (Helsinki n = 10, RMH n = 2, VISTA-ICH n = 22), we included 374 diabetic patients (Helsinki n = 138, RMH n = 85, VISTA-ICH n = 151) in the analysis. Excluded patients had similar 90-day mortality (33% vs. 30%, p = 0.721) and baseline characteristics with less metformin (28% vs. 40%, p = 0.202) and sulphonylurea (24% vs 31%, p = 0.455) use. Most (90%) patients presented within 24 h of symptom onset.

Metformin use was recorded in 148 (40%) patients while 115 (31%) patients used sulphonylurea at ICH onset. Metformin users compared to non-users had smaller ICH, milder symptoms, and lower mortality (24% vs. 34% p = 0.045) in univariate analysis (Table 1 and Figure 1). There was no difference in the baseline oedema volume between users and non-users of metformin (7.8 vs 8.7 mL, p = 0.206) or sulphonylurea (8.8 vs 7.9 mL, p = 0.617) (Table 1). Metformin users had lower mortality rates across all strata of NIHSS and baseline ICH volumes (Figure 2 and Supplementary Table). Sulphonylurea users were older with more concurrent insulin use than non-users, but the other characteristics and mortality (30% vs. 31%; p = 0.860) were similar. Metformin was paused or stopped in 105 (71%) and sulphonylurea in 68 (59%) of patients in the first two days.

Table 1.

Baseline characteristics and 90-day mortality.

	Diabetic (n = 374)	Metformin (n = 148)	Non-metformin (n = 226)	p	Sulphonylurea (n = 115)	Non-SU (n = 259)	p
Age (year)	68 (60–76)	68 (62–75)	70 (59–76)	0.984	71 (62–78)	67 (59–75)	0.010
Male	258 (69.0%)	101 (68.2%)	157 (69.5%)	0.802	82 (71.3%)	176 (68.0%)	0.518
NIHSS	12 (6–19)	11 (4–19)	13 (8–19)	0.024	11 (4–18)	13 (7–19)	0.056
GCS	14 (12–15)	14 (12–15)	14 (12–15)	0.584	15 (13–15)	14 (11–15)	0.127
ICH volume (mL)	12.6 (4.9–29.5)	10.6 (3.4–26.8)	14.3 (5.8–30.8)	0.024	11.1 (4.3–30.0)	13.2 (4.9–28.7)	0.621
Oedema volume (mL)	8.2 (4.0–20.4)	7.8 (3.1–21.3)	8.7 (4.4–19.6)	0.206	8.8 (3.9–19.6)	7.9 (4.2–20.8)	0.617
Infratentorial location	48 (12.8%)	23 (15.5%)	25 (11.1%)	0.205	14 (12.2%)	34 (13.1%)	0.799
Intraventricular extension	141 (37.7%)	64 (43.2%)	77 (34.1%)	0.073	51 (44.3%)	90 (34.7%)	0.077
Warfarin use	65 (17.4%)	32 (21.6%)	33 (14.6%)	0.080	26 (22.6%)	39 (15.1%)	0.075
90-day mortality	113 (30.2%)	36 (24.3%)	77 (34.1%)	0.045	34 (29.6%)	79 (30.5%)	0.856
Insulin use	87 (23.3%)	29 (19.6%)	58 (25.8%)	0.167	14 (12.2%)	73 (28.3%)	0.001
Admission glucose (mmol/L)	9.9 (7.8–13.1)	10.1 (7.9–12.4)	9.9 (7.7–13.5)	0.972	10.4 (8.4–13.7)	9.6 (7.8–12.6)	0.145
Serum creatinine (mmol/L)	80 (62–97)	72 (60–93)	81 (63–102)	0.003	80 (64–93)	80 (62–100)	0.406

Note: All values are median (interquartile range) or n (%). NIHSS, National Institutes of Health Stroke Scale; GCS, Glasgow Coma Scale; ICH, intracerebral haemorrhage; SU, sulphonylurea.

Figure 1.

Kaplan Meier curve for all-cause mortality at 90 days for users (n = 148) and non-users (n = 226) of metformin.

Figure 2.

Pre-stroke metformin treatment status and 90-day mortality rate stratified by baseline intracerebral haemorrhage volume (a) and baseline National Institutes of Health Stroke Scale (b).

In the multivariable model, there was no association with sulphonylurea (OR 0.96; 95% CI 0.49–1.88; p = 0.906) on outcome. Metformin use prior to ICH was associated with lower 90-day mortality (OR 0.51; 95% CI 0.26–0.97; p = 0.041) (Table 2). The association of metformin and 90-day mortality remained significant when analysis was limited to real life consecutive Helsinki and RMH ICH patients. There was no treatment interaction by any of the three datasets (p = 0.334 for interaction). Discontinuation of metformin on admission did not influence the association (p = 0.471 for interaction). The regression model fit was good (AUC 0.88; 95% CI 0.85–0.92).

Table 2.

Multivariable logistic regression on factors associated with 90-day mortality in diabetic ICH patients (n = 374).

	OR 95% CI	p value
Age (per year)	1.06 (1.03–1.09)	<0.001
Male	2.87 (1.39–5.93)	0.004
NIHSS	1.11 (1.05–1.17)	<0.001
GCS (per point)	0.91 (0.79–1.05)	0.199
ICH volume (per mL)	1.03 (1.01–1.05)	<0.001
Infratentorial location	3.10 (1.11–8.68)	0.031
Intraventricular extension	3.02 (1.63–5.60)	<0.001
Warfarin use	1.06 (0.48–2.34)	0.890
Sulphonylurea	0.96 (0.49–1.88)	0.906
Metformin	0.51 (0.26–0.97)	0.041

NIHSS, National Institutes of Health Stroke Scale; GCS, Glasgow Coma Scale; ICH, intracerebral haemorrhage.

Post-hoc analyses

The association of metformin and outcome remained significant after including pre-ICH insulin use, baseline glucose and creatinine into the model (OR 0.50; 95% CI 0.25–0.99; p = 0.047). Prior insulin use, (OR 0.82; 95% CI 0.35–1.89; p = 0.635), baseline glucose (OR 0.97; 95% CI 0.90–1.05; p = 0.486) or serum creatinine (OR 1.00; 95% CI 0.99–1.01; p = 0.781) were not independently associated with 90-day mortality.

After excluding 24 (6.4%) patients who underwent surgical haematoma evacuation, the association with metformin and 90-day mortality remained (OR 0.44; 95% CI 0.20–0.88, p = 0.020).

There were 159 (43%) lobar and 215 non-lobar (57%) haemorrhages by location. The location did not influence the association of metformin and 90-day mortality (p = 0.502 for interaction).

Propensity score matching reduced the sample size to 254 patients. The logistic regression analysis performed following matching demonstrated a trend towards reduced 90-day mortality in patients with pre-ICH metformin use (OR 0.55; 95% CI 0.24–1.24; p = 0.15).

Discussion

To our knowledge, this is the first study to investigate the association of pre-ICH metformin or sulphonylurea use on outcome. We demonstrated that pre-ICH metformin use was associated with halved odds of death by 90 days in diabetic ICH patients. In contrast to observational studies in ischaemic stroke,^2–4 pre-ICH use of sulphonylurea was not associated with outcome.

Metformin induces neuronal adenosine monophosphate-activated protein kinase (AMPK), an important mediator of cellular energy homeostasis.¹⁴ In rodent models of ischaemic stroke, metformin administration beginning at 24 h after stroke did not attenuate infarct growth at 72 h from stroke onset.¹⁵ In contrast, chronic pre-stroke metformin administration reduced lactate production and infarct volume¹⁰ while chronic post-stroke use starting at 24 h and continued for three weeks improved neurological recovery with enhanced angiogenesis and reduced glial scarring in an AMPK dependent mechanism.¹⁵ The effect of metformin pre-conditioning may persist for at least 96 h after drug cessation.¹⁴ The survival benefit observed in the present study may be mediated by a similar mechanism of reduced metabolic stress and improved glycaemic control.

Hyperglycaemia has been associated with mortality after ICH,^16,17 likely mediated through increased risk of infection and cardiac complications. In the present study, baseline glucose was not associated with metformin or sulphonylurea use, nor with outcome. In a rat model of ICH, hyperglycaemia significantly increased oedema volume and peri-haematomal cell death.¹⁸ Although limited human data suggest no significant effect of glucose on oedema growth^19–21 and we observed no difference in baseline oedema volume between drug groups, it is plausible that metformin pre-conditioning resulted in benefits through other pathways, such as decreased lactate production via AMPK and fewer medical complications through improved glycaemic control. Despite a high drug discontinuation rate and the negative statistical interaction of drug cessation on outcome, it is likely that continuation of metformin will influence ICH outcome through these mechanisms.

The results of our analysis were derived from multiple datasets and the different management approaches to ICH may potentially influence the results. There was, however, no difference in the association of metformin with outcome by dataset (p = 0.334 for interaction), which adds to the external validity of our findings.

We acknowledge that our results are based on a relatively small sample and the association is at risk from type I error. However, our analysis was defined a priori and the association remained significant in a number of post-hoc analyses including adjustment for baseline glucose, creatinine and removing the potential impact of haematoma evacuation. Our results require further validation in other consecutive ICH cohorts as following propensity score matching the association was not significant, accounted for by further reduction of power (n = 254). However, metformin is inexpensive and has been demonstrated in the Carotid Atherosclerosis: Metformin for Insulin Resistance (CAMERA) trial to be safe in non-diabetic subjects²² making it a suitable agent for assessment in a randomised controlled trial.

There are possible explanations for the neutral results observed with sulphonylurea. SUR-1-TRPM4 is activated by adenosine triphosphate depletion²³ and peri-haematomal hypoperfusion²⁴ may not reach thresholds that trigger physiologically significant SUR-1-TRPM4 activation. Also, the relative contribution of SUR-1-TRPM4 in secondary injury after ICH is unknown and has not been assessed in non-subarachnoid ICH models. Effective blockade of SUR-1-TRPM4 may not mitigate subsequent injury due to iron-mediated neuro-toxicity and blood brain barrier dysfunction.²⁵ In addition, discontinuation of sulphonylurea would reduce SUR-1-TRPM4 binding and the high discontinuation rates observed make our neutral results with sulphonylurea more difficult to interpret.

Limitations

There are limitations to our study. First, due to the retrospective nature, it is subject to a number of possible biases. We tried to reduce these by pre-defining our analysis prior to pooling of the data. Also, only 8% of consecutive patients had to be excluded due to missing data. Second, in an observational study of patients with diabetes, the use of metformin or sulphonylurea is inherently not random. It is possible that our results are confounded by indication and biased by uncontrolled for confounders. We tried to control for baseline glucose, prior insulin use, renal function, impact of surgical evacuation and influence of haematoma location in our post-hoc analyses and our results remained robust. Also, the overall model fit with AUC of 0.88 was high.

Third, we do not have information on oedema or haematoma growth, an important cause of neuronal injury^9,26–29 and a possible mediator of the observed associations. We also do not have detailed glucose profiles beyond admission, HbA1c, or record of medical complications following admission, which could have provided insight into a potential mediator of improved outcome exhibited by metformin. While mortality is a robust outcome measure in ICH, data on functional outcomes would have been preferable but not available across the datasets. Finally, our neutral results for sulphonylurea use are based on a relatively small sample size and thus only rule out a major mortality benefit, still allowing for the possibility of a modest effect.

Conclusion

In conclusion, we have observed an association between metformin use and improved outcome in ICH. Metformin has a known safety profile and may provide an economical and accessible therapeutic option in ICH. Our results generate hypotheses which after further validation could be tested in clinical trials.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: TT reports receiving grants from Boehringer-Ingelheim, Sanofi Aventis, Brains Gate, Bayer, Pfizer, and personal fees from Bayer and Pfizer, outside the submitted work. SMD has received consulting fees from Boehringer Ingelheim, BMS, Pfizer, Ever Neuropharma and Medtronic. AM reports speaker’s honoraria and travel cost from Siemens and speaker’s honoraria and consulting fees from Stryker and Boehringer Ingelheim outside the submitted work. All other authors report no relevant disclosures.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: TYW is supported by grants from the Neurological Foundation of New Zealand (grant number 1313-CF) and Royal Melbourne Hospital Neuroscience Foundation; BCVC receives research support from the National Health and Medical Research Council of Australia (GNT1043242, GNT1111972, GNT1113352), National Heart Foundation (Future Leaders Fellowship 100782); TT is supported by the Helsinki University Central Hospital and Sahlgrenska University Hospital grants for ICH research; SMD is supported by grants from the National Health and Medical Research Council (Australia). NY is supported by the National Health and Medical Research Council (Australia) and Australian Research Council. AM is supported by grants from National Health and Medical Research Council (Australia), Academy of Finland and the Finnish Medical Foundation.

Ethical approval

This study was approved by the Melbourne Health Human Research Ethics Committee, reference number QA2015049.

Informed consent

This observational registry study was approved with no formal patient consent required.

Guarantor

AM.

Contributorship

TYW and AM conceived and designed the study and performed statistical analysis. TYW drafted the manuscript. TYW, DS, JP, and AM collected data. All authors interpreted the data and revised the manuscript for intellectual content.