Cognitive Impairment and Risk of Stroke
A Systematic Review and Meta-Analysis of Prospective Cohort Studies
Background and Purpose—Cognitive impairment is linked to vascular risk factors and brain vascular pathologies. Several studies have tested whether subjects with cognitive impairment have higher risk for stroke. The aim of this study was to systematically review available evidence on the association between cognitive impairment and risk of stroke to obtain precise effect estimates of the association and to identify which cognitive domains associate most with incident stroke.
Methods—PubMed, EMBASE, and Web of Science were searched from January 1, 1980, to October 1, 2013, without language restriction. Only prospective cohort studies were included. From each study, data on the association between cognitive impairment and stroke estimated with hazard ratios or relative risks with 95% confidence interval (CI) were extracted. For each study, risk of stroke per SD lower performance in various cognitive tests was calculated.
Results—Twelve studies were included, comprising 82 899 participants of whom 3043 had an incident stroke. The pooled relative risk per SD lower global cognitive performance was 1.19 (95% CI, 1.12–1.27). Each SD lower score in executive function or attention was associated with 1.14-fold (95% CI, 1.06–1.24) higher risk of stroke. Lower scores in memory were associated with 1.07-fold (95% CI, 1.02–1.12) higher risk of stroke, and lower scores in language were associated with 1.08-fold (95% CI, 1.02–1.16) higher risk of stroke.
Conclusions—Cognitive impairment is associated with higher risk of stroke. The associations were not significantly different for executive function, memory, and language.
Current evidence indicates that vascular risk factors and disturbance in cerebrovascular hemodynamics are associated with cognitive impairment.1–3 Previous studies have shown that cerebrovascular pathologies enhance the adverse effect of neurodegenerative changes by reducing the threshold for cognitive impairment and increasing the incidence of dementia.4 Consistently, neuroimaging studies have determined that a considerable number of patients with cognitive impairment and dementia have high loads of white matter lesions, infarcts, and cerebral microbleeds.5 Hence, it has been suggested that cognitive impairment is closely linked with clinically unrecognized brain vascular pathologies, which may be associated with a higher risk of stroke.
Several longitudinal studies have investigated whether cognitive impairment is associated with increased risk of stroke.6–17 Nevertheless, interpretation and comparability of their findings are hampered because of methodological differences, application of various cognitive tests, and diverse reporting of the results. The aim of this systematic review was to evaluate available evidence systematically on the association between cognitive function and risk of stroke. In addition, a meta-analysis was performed: (1) to obtain precise and uniform magnitude of the association and (2) to identify specific cognitive domains associated with incident stroke.
PubMed, EMBASE, and Web of Science were searched with the terms cognitive function, cognition disorders, cognitive test, stroke, cerebrovascular disorders, and cerebrovascular accidents (Figure I in the online-only Data Supplement). The design of electronic search strategy was done in consultation with an expert reference librarian, and the period of search was from January 1, 1980, to October 1, 2013, without language restrictions. We used broad inclusion criteria for search strategy, including all types of stroke and cognitive impairment assessed by various cognitive tests and measurements.
Eligibility Criteria/Study Selection
Extracted citations were screened for eligibility by 2 independent reviewers (S.R. and S.M.). Studies were included if they met the following criteria: (1) prospective study design, (2) assessment of cognitive impairment preceding stroke, (3) outcome of interest was stroke, and (4) estimated risk of stroke by valid statistical analysis. The reviewers independently screened titles and abstracts to determine possible eligibility for inclusion (exclusion and inclusion criteria are presented in Table I in the online-only Data Supplement). Differences of opinion between the 2 reviewers were resolved by discussion with the third reviewer (B.S.). Then, full articles with their reference lists were checked for eligibility. To avoid inclusion of studies with similar populations, final studies were checked for similar authors, patient characteristics, and results. Where study populations were overlapping or duplicated, the most comprehensive study was included.
The following information was extracted from the included articles: author name, year of publication, location, number of participants, number of strokes, follow-up (years), male percentage, mean age with SD, cognitive tests and measurements, type of stroke, stroke measures (whether it was self-reported, medical-reported, or death certificates), and exclusion of subjects with stroke at baseline. In each eligible study, data on the association between cognitive impairment and stroke were estimated with hazard ratios or relative risks (RRs) with measures of uncertainty.
Methodological quality of included studies was assessed independently by the first and second reviewer using previously explained criteria18,19 (details in Table II in the online-only Data Supplement). In total, 18 criteria were considered, and each study that met each criterion received a score of 1, and if not received a score of 0. Therefore, maximum possible score was 18.
To make effect estimates comparable between studies, we decided to express the effect as RR per SD lower performance in cognitive tests for all the studies. Some studies already gave their results in that way.11–13,15,17 One study reported RR per unit lower cognitive score.9 For that study, we extracted data on SD of the cognitive test and multiplied the log RR and its standard error by SD. Other studies categorized the cognitive test scores and provided RRs compared with a reference category.6–8,10,13,14,16 In that case, we assumed that the categories had originated from an underlying continuous cognitive test with a normal distribution. We used the method reported by Hartemink et al20 to calculate the estimate of RR per SD of the underlying normal distribution, assuming a log linear relation on the underlying normal distribution scale between the cognitive test and the risk of stroke. To use this method also, the numbers of cases per category are needed, which were mostly given in the articles. If they were not given, the numbers of cases were calculated from the total number of cases (which was always given) and RRs. For studies that used >1 cognitive test,10–13,15,17 we calculated a pooled RR for each study. After calculation of RR for each study, a meta-analysis was performed by using a random-effects model. The random-effects model was applied because it takes into account the variability between studies. All statistical analyses were performed using STATA version 10. Heterogeneity of RR across studies was evaluated by Q statistic and I2 statistic.21 The possibility of publication bias was examined with Egger test.22
The literature search yielded 2716 articles. After abstract and title screening, 62 studies were identified by the first reviewer and 55 by the second reviewer. After removal of duplicates and full text screening, 14 articles met the entry criteria. However, 2 articles were excluded because of the overlap between study populations,23,24 leaving 12 studies for inclusion in the review. A flow diagram of the search strategy is provided in Figure 1.
The characteristics of 12 included studies are presented in the Table. The total number of participants was 82 899 with 3043 stroke events (3.7%). Follow-up years ranged from 3 to 21. Most studies included both men and women >65 years old. In 2 studies, stroke events were recorded using self-reported questionnaires,9,12 and others reported strokes either by medical report or death certificates. Two studies did not exclude subjects with a history of stroke,6,14 and 2 studies did not specify whether participants had a history of stroke.11,15 From a total of 18 points, the overall quality assessment scores ranged from 10 to 16 points (Table).
Details of the findings from individual studies are provided in Table IV in the online-only Data Supplement. After calculation of RR per SD lower performance in cognitive tests, 10 of 12 studies showed a significant association. As presented in Figure 2, pooling the estimates from individual studies showed that each SD lower performance in cognitive test was associated with 15% higher risk of stroke (RR, 1.15; 95% confidence interval [CI], 1.10–1.21). Significant heterogeneity between studies was present (I2=70.4%; P<0.001). The funnel plot was asymmetrical, and the analysis of publication bias confirmed a possibility for publication bias (P=0.003). We performed a sensitivity analysis to test whether the exclusion of studies with smallest sample size (<1000 participants)6,13,15,16 influenced the findings. The exclusion of these studies did not materially alter the results; pooled RR per SD lower cognitive performance was 1.15 (95% CI, 1.09–1.21; P for heterogeneity, 0.036; I2=40%). Another sensitivity analysis tested whether the exclusion of studies that included participants with a history of stroke6,14 or did not have report on that11,15 changed the association. The exclusion of these studies did not essentially change the results; pooled RR per SD lower cognitive performance was 1.21 (95% CI, 1.11–1.32; P for heterogeneity 0.003; I2=52.0%). We performed a sensitivity analysis and excluded studies that assessed cognitive function only using Mini-Mental State Examination (MMSE).8,9,14 This sensitivity analysis showed similar association; pooled RR per SD lower cognitive performance was 1.13 (95% CI, 1.07–1.19; P for heterogeneity, 0.001; I2=68.7%).
To assess whether impairment of various cognitive domains had similar associations with risk of stroke, we categorized the cognitive tests in groups of (1) global cognitive function, (2) memory function, (3) executive function or attention, and (4) language (detailed information regarding tests used for specific cognitive domains provided in Table III in the online-only Data Supplement). This subgroup analysis showed that each SD lower score in global cognitive function was associated with 1.19-fold (95% CI, 1.12–1.27) higher risk of stroke; lower scores in executive function or attention were associated with 1.14-fold (95% CI, 1.06–1.24) higher risk of stroke; lower scores in memory were associated with 1.07-fold (95% CI, 1.02–1.12) higher risk of stroke; and lower scores in language were associated with 1.08-fold (95% CI, 1.02–1.16) higher risk of stroke (Figure 3). In addition, we calculated the pooled RR using unadjusted or minimally adjusted models that were reported in included studies. This analysis showed that each SD lower cognitive performance was associated with 25% higher risk of stroke (RR, 1.25; 95% CI, 1.14–1.37) when individual studies did not account for cardiovascular factors.
Our systematic review and meta-analysis show that cognitive impairment is associated with higher risk of stroke. The associations were similar for executive function, memory, and language.
Previous studies have shown that ≈10% of patients with first-time stroke have pre-existing dementia, and that poststroke dementia substantially increases the risk of recurrent stroke.25 There are different explanations for the association between cognitive impairment and risk of stroke. First, the association might be explained by shared risk factors between cognitive impairment and stroke. Cardiovascular risk factors and diseases such as hypertension, diabetes mellitus, and arterial fibrillation associate with higher risk of both cognitive impairment and stroke.10,14,26–28 However, most of the studies in our systematic review showed that the association between cognitive impairment and stroke was independent of cardiovascular risk factors. Another explanation could be that cognitive impairment might reflect microvascular injuries and silent cerebrovascular events, which eventually develop into a clinical stroke over time.8 However, only 1 study reported brain MRI data, which were limited to brain structural volumes and white matter hyperintensities rather than silent cerebrovascular events. Providing these data would allow us to further assess how pre-existing silent stroke events contribute to the relationship between cognitive impairment and developing clinically evident strokes. A third explanation could be that patients with cognitive impairment are possibly more exposed to unhealthy lifestyles, which indirectly increase the risk of stroke.9,29–31In this systematic review, we observed that the majority of studies did not consider lifestyle factors, such as diet and physical activity, in their analyses, which need to be considered in future studies.
Some studies have suggested an age-dependent association between cerebrovascular pathologies and cognitive impairment. For instance, postmortem studies have shown that, with an increase in age, vascular pathologies become the main pathological findings in the brain of subjects with dementia, whereas cognitive impairment in younger subjects is more related to typical neuropathologic changes commonly observed in Alzheimer disease.32,33 In other words, cognitive impairment in old age might be a better reflection of brain vascular pathologies, which predispose subjects to future cerebrovascular events. Another explanation could be that patients with cognitive impairment might have a worse medical adherence or they receive less aggressive treatments compared with cognitively healthy people.
Most of the included studies assessed cognitive function just once at baseline.6,8–11,13–17 Three studies measured cognitive function more than once.7,12,15 DeFries et al12 showed that the stability of cognitive functioning also plays a role in the prediction of stroke. They showed that participants with stable cognitive performance had lower probability to develop stroke in comparison with those who had a considerable decline in cognitive functioning. Likewise, Ferrucci et al7 showed that subjects with stable cognitive status or those who improved over time were at a lower risk of stroke. This association was independent of baseline cognitive function. In contrast, Batterham et al15 reported that a decline in cognitive performance was not associated with a higher risk of stroke-related mortality. Another aspect that needs to be highlighted is the quality of studies retrieved from our systematic review. Despite a relatively high overall quality scores, some of the included studies assessed cognitive function applying only 1 type of cognitive test and did not provide data on the characteristics of nonresponders or those who did not complete the study. To improve generalizability of the findings, future studies need to address these issues.
Current evidence indicates that certain domains of cognition might be more vulnerable to the adverse effects of cardiovascular risk factors.34 Impaired executive function and attention is a major clinical presentation in patients with vascular cognitive impairment, whereas memory deficit is mainly associated with Alzheimer type of dementia.35 In line with this evidence, we found that estimates of the association of lower performance in executive function and attention with stroke were slightly larger compared with the estimates on the association of lower memory performance with stroke. This finding might suggest that subjects with impaired executive function might carry higher loads of covert brain vascular pathologies, which put them at a higher risk for future clinically evident cerebrovascular events.
Several reports have shown that strategies to identify subjects at high risk of stroke can reduce the burden of stroke.36–38 In this setting, different prediction tools have been developed to predict stroke in high-risk individuals.39 Our findings suggest that cognitive assessment, particularly in older subjects, might be an easily accessible tool in daily practice for better identification of subjects at higher risk of stroke. Feasibility, acceptability, and additive value of cognitive assessment in early identification of subjects at high risk of stroke need to be further investigated in future studies.
This systematic review has certain strengths and limitations. As strength, we included long-term large prospective studies from Europe, United States, and Australia, the regions in which aging of the population and consequently increasing prevalence of age-related disorders, such as dementia and stroke, have been major public health issues. In addition, we searched 3 major databases without language restriction to find the most relevant articles. The included studies used different types of cognitive tests, and this gave us a chance to review the association of different cognitive domains with stroke. However, this variability could be considered as a limitation because it is difficult to compare the findings among different studies. Another limitation was an increased probability of publication bias because our search strategy did not include unpublished data. It is possible that studies with negative findings had lesser chance of being published. However, the restriction to larger studies showed that the influence of small studies on the association between cognitive impairment and stroke was minimal. Some of the included studies assessed cognitive function by using MMSE as a single test. Despite the advantages of MMSE, such as being used worldwide and can be administered with minimal expertise, MMSE score may not reliably reflect the degree of cognitive impairment. It is worth pointing out that when we excluded studies that assessed cognitive function only with MMSE, the findings did not essentially change. As a major limitation, 4 of the included studies did not exclude subjects with a history of stroke or did not report about that. However, a sensitivity analysis with exclusion of those studies did not essentially change the results.
In conclusion, our systematic review and meta-analyses indicate that impaired cognitive performance is significantly associated with higher risk of stroke, with no clear distinction between cognitive domains. Cognitive assessment possibly in combination with other existing diagnostic methods might be considered as a strategy for better identification of subjects at higher risk of stroke.
We thank Karin van der Hoorn from the Walaeus Library, Leiden University Medical Centre, for her help in literature search. S.R. and S.M. contributed in design of search strategy, performing the literature search, data extraction, and drafting of the article. B.S. contributed in formulating research question, design of search strategy, interpretation of data, and revising the article. T.S. contributed in statistical analyses of the data and revising the article. A.J.M.d.C. contributed in design of search strategy, interpretation of data, and revising the article.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.114.004658/-/DC1.
- Received February 19, 2014.
- Revision received February 19, 2014.
- Accepted February 24, 2014.
- © 2014 American Heart Association, Inc.
- Gale CR,
- Martyn CN,
- Cooper C
- O’Donnell M,
- Teo K,
- Gao P,
- Anderson C,
- Sleight P,
- Dans A,
- et al
- Sabayan B,
- Gussekloo J,
- de Ruijter W,
- Westendorp RG,
- de Craen AJ
- Weinstein G,
- Beiser AS,
- Decarli C,
- Au R,
- Wolf PA,
- Seshadri S
- Tooth L,
- Ware R,
- Bain C,
- Purdie DM,
- Dobson A
- Hartemink N,
- Boshuizen HC,
- Nagelkerke NJ,
- Jacobs MA,
- van Houwelingen HC
- Shipley BA,
- Der G,
- Taylor MD,
- Deary IJ
- Whincup PH,
- Gilg JA,
- Emberson JR,
- Jarvis MJ,
- Feyerabend C,
- Bryant A,
- et al
- Gorelick PB,
- Scuteri A,
- Black SE,
- Decarli C,
- Greenberg SM,
- Iadecola C,
- et al
- Liu L,
- Xue F,
- Ma J,
- Ma M,
- Long Y,
- Newschaffer CJ