Statin Treatment and Functional Outcome after Ischemic Stroke: Case-control and Meta-analysis

Abstract

Background and Purpose

Multiple studies suggest that statin use prior to acute ischemic stroke (AIS) is associated with improved functional outcome. However, available evidence is conflicting, and several published reports are limited by small sample sizes. We therefore investigated the effect of antecedent use of statins on stroke outcome by performing a meta-analysis of all results from published studies as well as our own unpublished data.

Methods

We performed a systematic literature search and meta-analysis of studies investigating the association between pre-stroke statin use and clinical outcome, and included additional data from 126 pre-stroke statin users and 767 non-users enrolled at our Institution. A total of 12 studies, comprising 2013 statin users and 9682 non- users were meta-analyzed using a random effects model. We also meta-analyzed results for individual TOAST stroke subtypes to determine whether the effect of statin use differed across subtypes, using the Breslow-Day (BD) test.

Results

Meta-analysis of all available data identified an association between pre-stroke statin use and improved functional outcome (Odds Ratio = 1.62, 95% Confidence Interval: 1.39 -1.88), but we uncovered evidence of publication bias. The effect of statin use on functional outcome was found to be larger for small vessel strokes compared to other subtypes (BD p = 0.008).

Conclusions

Antecedent use of statins is associated with improved outcome in AIS patients. This association appears to be stronger in patients with small vessel stroke subtype. However, evidence of publication bias in the existing literature suggests these findings should be interpreted with caution.

Introduction

Despite currently available treatment options for acute ischemic stroke (AIS), patients often face the prospect of substantial post-stroke disability.^1,2 Results from in vitro and animal model studies suggest that statins have beneficial effects on functional recovery after AIS.³ Pleiotropic effects of statins including their influence on angiogenesis, neurogenesis, synaptogenesis, as well as possible regulation of cerebral blood flow and modulation of the coagulation and immune systems,⁴ have been proposed as possible biological mechanisms accounting for these observations.

Evidence from human studies investigating the effect of statins on AIS outcome is conflicting, and the interpretation of published results is often complicated by small sample sizes, possible bias, and confounding.^5-15 In addition, a recent report suggested that statin effect on outcome might vary depending on stroke etiology¹⁴, as determined by Trial of ORG 10172 in Acute Stroke Treatment (TOAST) criteria.¹⁶ We therefore performed a meta-analysis of all available published evidence, as well as our own unpublished data from the Massachusetts General Hospital (MGH). Furthermore, we sought to investigate possible variability in the effect of statins on AIS outcome based on stroke subtype and attempted to replicate and expand on the reported observation in our own cohort.

Methods

MGH Cohort

Patient selection

All AIS subjects were enrolled as part of an ongoing single-center prospective cohort study of consecutive ischemic stroke patients aged ≥18 years admitted to the MGH Stroke Unit after presenting to the Emergency Department (ED) within 24 hours of symptom onset between 2003 and 2009.¹⁷ Ischemic stroke was defined as a clinical syndrome of any duration associated with a radiographically proven acute infarct consistent with a vascular pattern of involvement and without radiographic evidence of a demyelinating or neoplastic disease, or other structural disease including vasculitis, subacute bacterial endocarditis, vasospasm due to subarachnoid hemorrhage or cocaine abuse, or primary intracerebral hemorrhage. Diagnosis of acute cerebral ischemia was confirmed for all AIS subjects in the present study by admission DWI MRI completed within 48 hours following symptom onset. The institutional review board approved all aspects of this study, and informed consent for collection of data was obtained for all subjects or their legal guardians.

Data collection and patient follow-up

All patients were evaluated by a neurologist in the ED, where the National Institute of Health Stroke Scale (NIHSS) score was determined, laboratory values were measured, and clinical information was abstracted prospectively by patient or proxy interview, and/or supplemented through medical chart review. Vascular risk factors including hypertension, diabetes, hyperlipidemia, coronary artery disease, and atrial fibrillation were recorded based on existing international guidelines as previously described.¹⁷ Medication use prior to admission was also ascertained, including antiplatelet, antihypertensive, and oral hypoglycemic agents, as well as warfarin and statins. AIS subtypes were assigned by stroke neurologists according to TOAST criteria.¹⁶ Ischemic stroke infarct size was quantified on DWI MRI according to previously published methods.¹⁸

Patients and their caregivers were interviewed by telephone at 3 months post-AIS by certified research staff to assess functional outcome using the modified Rankin Scale (mRS) score. The use or discontinuation of medications after discharge was specifically assessed in this interview. Favorable outcome was defined as mRS ≤ 2 at 90 days, in agreement with previously published studies.^5-15

Statistical analysis

All statistical analyses were performed using STATA 10.0. Continuous numerical variables were expressed as mean ± standard deviation (SD) with the exception of DWI volume which was expressed as the median ± inter-quartile range (IQR). Cases and controls were compared in univariate analyses using t-test, Wilcoxon rank sum, chi-square, or Fisher's exact test as appropriate.

Independent predictors of post AIS functional outcome were identified using multivariate logistic regression analysis. Candidate covariates for multivariate modeling included all variables showing a trend in association with outcome in univariate analysis (p < 0·20), as well as variables showing trends towards differential distribution in statin users vs. non users (p < 0·20). Backward elimination of non-significant variables (p > 0·05) was subsequently used to generate a minimal model. We separately analyzed effects of statins on outcome in each TOAST stroke subtype group, and compared them using the Breslow-Day test. Significance threshold was set at p < 0·05 (two-tailed) for all analyses.

Meta-analysis

Literature search and study inclusion criteria

A literature search was performed independently by A.B. and W.J.D. in order to identify human studies in English language exploring the effect of statin use on functional outcome after AIS. Search terms are listed in the Supplementary Appendix (S1) and literature search work-flow is detailed in Supplementary Figure (S2)(please see http://stroke.ahajournals.org). Queried databases included PubMed, Medline, Embase and Ovid. Manual review of references in articles matching searching criteria was conducted to identify potential additional reports. Data were retrieved and meta-analyzed independently by A.B. and W.J.D., and results compared for consistency.

Statistical methods

Results from multivariate regression analyses in individual studies were meta-analyzed using a conservative inverse variance-based random effects pooling method (DerSimonian-Laird).¹⁹ Cochran's Q test was used to estimate heterogeneity, followed by calculation of I² (percentage of effect size attributable to heterogeneity). Effect size heterogeneity was considered significant for heterogeneity p-values < 0.10 or I² > 0.20.²⁰ Potential study characteristics predicting heterogeneity in effect size determination were identified using a meta-regression approach.²¹

Publication bias was quantified by inspection of funnel plots and computation of Egger's and Begg's tests p-values.²¹ Further quantification of the impact of publication bias on meta-analysis results was obtained by applying the “trim and fill” method to estimate the percentage of unpublished negative studies.²²

Results

MGH

Of 893 AIS patients, there were 126 subjects with antecedent statin use and 767 subjects without (Table 1). Compared to non statin users, AIS patients who used statins were more likely to have a history of hyperlipidemia and ischemic heart disease, and to be prescribed antihypertensive and glycemic control agents (p<0.003). Ischemic stroke characteristics (including NIHSS and TOAST subtype breakdown) did not differ between groups. Favorable functional outcome at 90-days (mRS 0-2) was not associated with antecedent statin use in univariate analysis (p = 0.10), nor in multivariate analysis (Odds Ratio (OR) = 1.51, 95% Confidence Interval (CI): 0.94 – 2.44), with evidence of a non-significant trend only towards association between statin use and favorable functional outcome (p = 0.08).

Table 1. MGH Cohort Characteristics.

Variable	Statin (n, %)	Non-Statin (n, %)	p-value
No. of Subjects	126 (100.0)	767 (100.0)
Age (Mean ± Standard Deviation)	65.7 ± 13.0	66.1 ± 15.1	0.60
Gender (% female)	42 (0.33)	316 (0.41)	0.10
Caucasian	115 (0.91)	686 (0.90)	0.75
Hypertension	89 (0.71)	486 (0.63)	0.13
Type 2 Diabetes	28 (0.22)	159 (0.21)	0.72
Coronary Artery Disease	34 (0.27)	142 (0.19)	0.03
Atrial Fibrillation	14 (0.11)	126 (0.16)	0.15
Previous TIA	9 (0.07)	56 (0.07)	0.99
Previous Ischemic Stroke	18 (0.14)	117 (0.15)	0.89
Hyperlipidemia	78 (0.62)	295 (0.38)	< 0.001
Previous History of ICH	0 (0.00)	4 (0.05)	0.99
Antihypertensive Agent Use	63 (0.50)	214 (0.28)	< 0.001
Glycemic Control Agent Use	24 (0.19)	73 (0.10)	0.003
Ever Smoker	80 (0.65)	392 (0.52)	0.009
Ever Alcohol User (> 1 drink / week)	81 (0.64)	392 (0.51)	0.007
NIHSS at Admission > 8	21 (0.17)	143 (0.21)	0.39
Systolic Blood Pressure at Admission (Mean, Standard Deviation in mmHg)	153 ± 28	152 ± 29	0.61
Diastolic Blood Pressure at Admission (Mean, Standard Deviation in mmHg)	80 ± 15	79 ± 16	0.65
TOAST Ischemic Stroke Subtype			0.16
Cardioembolic	36 (0.29)	249 (0.32)
Large Artery	34 (0.27)	189 (0.25)
Small Vessel	10 (0.08)	92 (0.12)
Other Determined Etiology	11 (0.09)	29 (0.04)
Unknown Etiology	35 (0.27)	208 (0.27)
90-day Favorable Outcome (mRS ≤ 2)	68 (0.54)	317 (0.41)	0.10
90-day Mortality	8 (0.06)	84 (0.11)	0.12

In the TOAST subtype subset analysis we identified an association between statin use and favorable outcome among 102 individuals (10 of whom were statin users) with small vessel ischemic stroke: OR = 1.97, 95% CI: 1.02 – 3.79, p = 0.04. The size of the effect of statin on outcome of small vessel strokes was found to be larger than in any other stroke subtype (p = 0.008).

Meta-Analysis

A total of 2013 AIS statin users and 9682 statin non-users were included in meta-analysis of data from our cohort and 11 previously published studies (Table 2). Meta-analysis identified an association between antecedent statin use and favorable outcome: OR = 1.62, 95% CI: 1.39 – 1.88, p < 0.001 (Figure 1). TOAST subtype meta-analysis showed association between statins use and AIS outcome in both large artery (OR = 2.01, 95% CI: 1.14 – 3.54, p = 0.016) and small vessel strokes (OR = 2.11, 95% CI: 1.32 – 3.39, p = 0.002).

Table 2. Individual Studies Included in Meta-analysis.

Study	Reference	Endpoint		Study Setting	Subjects			Matched Controls	Stroke Subtypes	Cholesterol Levels
		Definition	Time		All	Statin	Non Statin
Reeves	5	mRS :0 - 4	Discharge	Multi-center	1360	309	1051	Yes	No	No
Greisenegger	6	mRS: 0 - 4	7 days	Multi-center	1691	152	1539	No	Yes	Yes
Yu	7	mRS: 0 - 3	10 days	Single Center	339	74	265	No	No	No
Yoon	8	mRS: 0 - 2	Discharge	Single Center	393	161	232	No	No	No
Arboix	9	mRS: 0 - 2	Discharge	Single Center	2082	381	1701	No	No	No
Marti-Fabregas	10	Barthel Index: 95-100	3 Months	Multi-center	167	30	137	No	Yes	Yes
Moonis	11	mRS: 0 - 2	3 Months	Multi-center	729	129	600	No	Yes	No
Elkind	12	other	Discharge	Population Based	570	48	522	No	No	No
Jonsson	13	other	7 Days	Population Based	275	125	150	Yes	No	No
Martinez-Sanchez	14	mRS: 0 - 2	Discharge	Single Center	2742	281	2461	No	Yes	No
Goldstein	15	mRS: 0 - 2	3 Months	Multi-center (Clinical Trial)	454	197	257	Yes	Yes	Yes
MGH (Present Study)	--	mRS: 0 - 2	3 Months	Single Center	893	126	767	No	Yes	Yes

Figure 1. Meta-Analysis Forest Plot.

Meta-analysis Heterogeneity

We identified only minor between-study heterogeneity of effect sizes: heterogeneity p-value was 0.36, with I² = 0.17 (95% CI: 0.00 – 0.57). To further explore meta-analysis heterogeneity, we performed meta-regression to identify associations between study characteristics and reported effect size (Table 3). We identified a significant association between small sample size for AIS statin users and larger reported effect size (p = 0.032), as well as a similar trend for non-statin users sample size (p = 0.078). Conversely, lack of adjustment for TOAST subtypes showed a trend towards association with smaller effect sizes (p = 0.09). Overall, predictors evaluated in meta-regression explained approximately 92% of identified heterogeneity.

Table 3. Meta-regression Results.

Study Characteristic	Effect Size (log Odds Ratio) Modification	95% CI Effect Size Modification	p
No. of Statin AIS (for a 100 subjects decrease)	+0.22	+0.02 / +0.43	0.032
No. of Non-Statin AIS (for a 100 subjects decrease)	+0.21	-0.03 / +0.45	0.078
Outcome Captured before 3 months post-stroke	-0.94	-2.54 / +0.66	0.18
Outcome not defined by mRS	+0.90	-0.52 / +2.30	0.15
Single Center vs. Multi-center / Population-Based	+0.84	-0.48 / +2.16	0.15
TOAST Subtypes Not Available	-0.81	-1.85 / +0.23	0.09
Cholesterol Levels Not Available	+0.77	-0.51 / +2.05	0.17

Small Study Effect and Publication Bias

As suggested by the meta-regression analysis, we identified evidence of small study effect (i.e. association between smaller sample size and larger reported effect) in our analysis, based on p-values for the Egger test (p < 10^-3) and the Begg test (p = 0.003). This finding is most likely reflects publication bias, the presence of which is suggested by the meta-analysis funnel plot (Figure 2). Using the “trim and fill” method to more accurately quantify the impact of publication bias, we estimated at least four studies (33% of published reports) to be unpublished and missing in our meta-analysis. Upon performing meta-analysis including both published and the hypothesized missing studies, the association between antecedent statin use and favorable outcome remained significant: OR = 1.52, 95% CI: 1.28 – 1.82, p < 0.0001 (Figure 3). However, heterogeneity was still apparent in this simulated meta-analysis (p = 0.08, I² = 0.35).

Figure 2. Meta-analysis Funnel Plot.

Figure 3. Forest Plot for the “Trim and Fill” Meta-analysis.

Discussion

Results from this meta-analysis suggest an association between pre-stroke statin use and improved functional outcome at 90 days. These findings suggest that statin exposure antecedent to acute cerebral ischemia may reduce its severity or improve immediate recovery from it.

Following-up on results from a previous report, we investigated potential differences in statin-related effect on functional outcome based on stroke etiology, as defined by TOAST criteria.¹⁴ Meta-analysis of subtype results yielded a beneficial effect of statins in both large artery and small vessel ischemic strokes; however, this meta-analysis of results for TOAST subtypes was limited to two studies. Furthermore, available sample size for the small vessel subtype analysis in our MGH case-control study was relatively small. Further exploration of any subtype-specific effects is necessary to elucidate possible associations with underlying biology of disease.²³

Our study has limitations. Most results included in meta-analysis were derived from observational studies. These studies are exposed to several potential sources of bias, including severity bias, recall bias, bias due to statins prescription practices, and propensity to treatment. However, results from the clinical trial included in meta-analysis are in agreement with observational studies. We also found minor effect size heterogeneity across studies; meta-regression results confirmed this finding and indentified small studies as consistently reporting larger effect sizes. It is therefore not surprising that evidence of small study effect is present in our meta-analysis, most likely due to publication bias. We attempted to correct for publication bias using the “trim and fill” method, i.e. by adding missing studies to the meta-analysis, and observed that the updated meta-analysis did not change significantly. However, introduction of missing unpublished studies generated substantial meta-analysis heterogeneity, suggesting caution in interpreting our findings.

Conclusion

Pre-treatment with statins is associated with improved outcome in patients with acute ischemic stroke. This association may be strongest in small vessel stroke patients. However, evidence of publication bias in the existing literature suggests these findings should be interpreted with caution. More clinical trial and experimental model data will be required before the biological mechanisms linking statin use to improved outcome after ischemic stroke are fully elucidated.