Arterial stiffness as a cause of cognitive decline and dementia: a systematic review and meta-analysis

Authors


  • Funding: MPP and NAG are each funded by an Australian Postgraduate Award.

  • Conflict of interest: MFO is a founding director of AtCor Medical, manufacturer of systems for analysing the arterial pressure pulse.

Matthew P. Pase, Centre for Human Psychopharmacology, Swinburne University of Technology, 400 Burwood Road, Hawthorn, Melbourne, Vic. 3122, Australia. Email: matthewpase@gmail.com

Abstract

Background:  Although arterial stiffness has recently been confirmed as a predictor of cardiovascular disease, the association between arterial stiffness and cognitive decline is less clear.

Aim:  We performed a systematic review and meta-analysis to examine the evidence for large artery stiffness as a cause of cognitive decline and dementia.

Method:  Electronic databases were systematically searched until September 2011 for studies reporting on the longitudinal relationship between any validated measure of large artery stiffness and cognitive decline or dementia. Meta-analysis was performed on four studies investigating the association between aortic pulse wave velocity and a decline in Mini-Mental State Examination scores.

Results:  Six relevant longitudinal studies were located, conducted over an average of 5 years follow up. Arterial stiffness was predictive of cognitive decline in five/six studies. In meta-analysis, higher aortic stiffness predicted lower Mini-Mental State Examination scores within the sample (β=−0.03, 95% confidence interval (CI): −0.06 to 0.01, n= 3947), although studies were not all homogeneous, and statistical heterogeneity was present (I2= 71.9%, P= 0.01). Removal of one study with a relatively younger cohort and lower median aortic stiffness found higher aortic stiffness to significantly predict cognitive decline (β=−0.04, 95% CI: −0.07 to −0.01, n= 3687) without evidence of heterogeneity (I2= 9.5%, P= 0.33). There was little research investigating the effects of aortic stiffness on the development of dementia.

Conclusion:  Aortic stiffness was found to predict cognitive decline in both qualitative review and quantitative analysis.

Introduction

Ageing is associated with a decline in fluid intelligence and slowing of cognitive abilities. Although specific domains of cognitive performance are known to decline with increasing age, the underlying causes require further investigation. Uncovering the mechanisms underpinning cognitive decline may help guide and tailor future interventions with the aim of preserving cognitive performance across the lifespan.

Normal ageing is associated with marked increase in stiffness of the body's central arteries due to changes in arterial structure1,2 and function.1,3 Increased stiffness of central arteries causes an increase in the speed of arterial wave propagation so that reflected waves arrive back at the aorta earlier, during the systolic part of the cardiac cycle.1,2 As a consequence, the systolic blood pressure (BP) is augmented, whilst the diastolic BP is reduced, creating an increase in arterial pulse pressure. As a result of increased arterial stiffness, the cerebral vasculature is exposed to the potentially damaging forces of augmented arterial pulse pressures.4

Many non-invasive methods have been developed to quantify arterial stiffness. Pulse wave velocity (PWV) measures the speed of a pulse wave travelling between two selected sites of measurement; this increases as arterial stiffness increases. The measurement of PWV between the carotid and femoral sites (cf-PWV) reflects arterial stiffness along the aorta, the primary site of age-associated arterial stiffening. This approach is considered to be the gold standard in the non-invasive assessment of large artery stiffness.1,5,6 Measurement of PWV between other sites does not measure aortic stiffness (as in the case of carotid–radial or femoral–tibial PWV) and has not been associated with cardiovascular events.7 The brachial–ankle PWV (ba-PWV) includes the stiffness of the muscular arteries in the arm and leg, which do not increase in stiffness with age.1 However, in limited studies, ba-PWV has been shown to be predictive of mortality.8

An increase in aortic PWV augments the aortic systolic pressure at the expense of the aortic diastolic pressure. Assessment of aortic pressures therefore provides indirect estimates of arterial stiffness through measures, such as aortic pulse pressure (systolic minus diastolic pressure), augmentation index (augmented pressure due to wave reflection divided by the pulse pressure expressed as a percentage) and amplification (peripheral divided by central pulse pressure). Central haemodynamic indices have been confirmed as predictive of cardiovascular events and mortality through meta-analysis.9

Although arterial stiffness is an established determinant of cardiovascular health,1,10 in relative terms, arterial stiffness is only an emergent predictor of cognitive performance, cognitive decline11 and even dementia.12 However, plausible mechanisms, now accepted to explain microvascular lesions in malignant hypertension,4,13,14 have been proposed to explain microvascular cerebral lesions in the elderly, and these are closely related to indices of aortic stiffening.15 Although many studies have been conducted in the area, to our knowledge, a systematic review of the human clinical data on arterial stiffness and cognitive decline has not been conducted. The aim of the current systematic review was to summarise the relevant studies and to examine the evidence for large artery stiffness as a cause of cognitive decline and dementia. Consequently, we have conducted a systematic review and meta-analysis of studies exploring the longitudinal relationship between the stiffness of large arteries and cognitive performance. No limits were set on the nature of the sample population.

Methods

Literature search

MEDLINE (PubMed), Web of Science and PsycINFO were searched until September 2011 using the following terms; ‘cognitive’ or ‘cognition’ or ‘memory’ or ‘dementia’ or ‘mini mental state examination’ and eight terms relating to arterial stiffness and related central pressures and wave reflections (‘arterial stiffness’ OR ‘pulse pressure’ OR ‘wave reflections’ OR ‘augmentation index’ OR ‘pulse wave velocity’ OR ‘aortic stiffness’ OR ‘central pressure’ OR ‘arterial compliance’). Forward searchers were performed on the studies meeting the inclusion criteria using SCOPUS. Furthermore, major journals were scrutinised for articles dealing with related issues, they, and their references were also included in the search.

Study eligibility

To be considered appropriate for review, each study must have included a validated measure of large artery stiffness, central pressure, wave reflections or large artery compliance. Each study had to report on the effects of arterial stiffness on a valid assessment of cognitive decline or dementia. As the current study aimed to assess the causality between arterial stiffness and cognitive decline, cross-sectional studies were not considered appropriate for review. No limits were imposed on the characteristics of the sample population. Articles of all languages were considered appropriate for review.

Data extraction and statistical analysis

Literature searching and the assessment of inclusion criteria were completed separately by two researchers (MPP and NAG), with results later compared and combined according to consensus of the research group. For the purpose of qualitative review, the data extracted consisted of study characteristics as well as regression coefficients and their reported significance, pertaining to the effect of arterial stiffness on each cognitive outcome. A meta-analysis was performed to estimate the predictive value of aortic stiffness (cf-PWV) on cognitive decline, as measured by Mini-Mental State Examination (MMSE) scores obtained at follow up. For this purpose, data were extracted by one author (AH) and verified by another (MPP), according to the protocol later.

The corresponding author of each study considered appropriate for meta-analysis was contacted by email or telephone and asked to provide individual patient data for the purpose of the current meta-analysis. For those studies where individual patient data were available, regression analyses were conducted to obtain the regression coefficients for baseline PWV measurements predicting final MMSE scores. Where a study administered the MMSE at multiple time points, the most recent MMSE score was used in meta-analysis, as this represents the longest duration in the study and therefore the most information on the strength of the association between baseline PWV and cognitive decline. The advantage of conducting regressions on the individual patient data was that all regression coefficients could be adjusted for the same confounding variables, including age, sex, mean arterial pressure, history of dementia, baseline MMSE scores and the study duration.

For those studies where individual patient data were not provided upon request, the regression coefficient reported in the original paper was extracted along with the associated confidence interval (CI). Where multiple regression models were reported each with different covariates, the regression coefficient and CI was taken from the model that controlled for the most covariates. This approach provided a more conservative estimate into the effect of PWV on outcome. One study did not report the CIs required for meta-analysis.11 In this case, the authors were contacted by email, and the requested statistics were provided.

Extracted regression coefficients and CIs were then pooled in a meta-analysis along with the regression coefficients and CIs calculated from those studies where individual patient data were available. Studies were each weighted by their inverse variance. Estimates were calculated using a fixed effects model except where there was significant evidence of heterogeneity. In this case, a random effects model was reported. The heterogeneity of studies was investigated using Higgins I2 statistic.16 Quantitative analysis was performed using STATA IC version 10 (StataCorp LP, College Station, TX, USA).

Results

Qualitative summary

Of the 133 located studies, six relevant studies met inclusion criteria and were included in review. The systematic review flow chart is shown in Figure 1.

Figure 1.

Systematic review flow chart. Cf-PWV, carotid–femoral pulse wave velocity; MMSE, Mini-Mental State Examination.

As detailed in Table 1, the most widely used measure of aortic stiffness was cf-PWV, which was implemented in all but one of the reviewed studies. Other measures of stiffness included the cardio–ankle vascular index, which was predictive of cognitive decline over a follow-up period of 4 years in a single study,17 as well as carotid distensibility, which was not predictive of cognitive decline or the development of dementia in the Rotterdam cohort.18

Table 1.  Summary of the studies investigating the relationship between arterial stiffness and cognitive decline or dementia
First author/year (ref)Stiffness measureCognitive outcomesAge (years)Men (%)Sample sizeParticipantsResult
  1. *P < 0.05, **P < 0.01, ***P < 0.001. ABC, Ageing, and Body Composition; AD, Alzheimer's disease; AS, arterial stiffness; CAVI, cardio–ankle vascular index; CD, carotid dispensability; cf-PWV, pulse wave velocity between the carotid and femoral sites; HDS-R, Hasegawa Dementia Scale – Revised; I-M-C, Information-Memory-Concentration; MLR, multiple linear regression; MMSE, Mini-Mental State Examination; OR, odds ratio; VaD, vascular dementia.

Benetos 201119cf-PWVMMSE8721873PARTAGE cohort. Participants living in nursing homes were followed up after 1 year.AS predicted decline in MMSE scores (r=−0.095**)
Poels 200718cf-PWV, CDMMSE, executive function, diagnosis of dementia71422767Rotterdam population-based cohort. Average of 5 years follow upIncreased PWV predicted MMSE, stroop and word fluency in MLR. AS did not predict cognitive decline or dementia.
Scuteri 200720cf-PWVMMSE7930102Subjective memory complaints Median follow up of 1 yearAS predicted cognitive decline (β=−0.74***) in MLR
Waldstein 200811cf-PWVNeuropsychological test battery, MMSE, Blessed I-M-C test5444582Baltimore Longitudinal Study of Ageing cohort. Non-demented, stroke free. Maximum of 11 years follow upAS predicted prospective decline on verbal learning (β=−0.0002*) and memory (β=−0.0008*), delayed recall (β=−0.0010**), non-verbal memory (β= 0.002***) and Blessed I-M-C test (β= 0.0004*). MMSE, simple attention, perceptuo–motor speed, executive functions and language were not predicted.
Watson 201121cf-PWVNeuropsychological test battery7348406The Health ABC study (Cognitive Vitality Sub-study). 6-year follow upAS predicted decline in psychomotor speed (OR = 1.42) but not verbal memory, global function or perceptual speed.
Yamamoto 200917CAVIMMSE, HDS-R793195Japanese without AD, VaD or symptomatic cerebral or myocardial infarction. Follow up at 4 yearsHigh AS at baseline predicted decline in the HDS-R (β= 0.24*) but not MMSE.

Of the studies to use cf-PWV, the average duration of follow up was 5 years, participants tended to be of old age and predominantly female. All but the Rotterdam trial18 reported that cf-PWV was a significant predictor of cognitive decline.11,19–21 Specifically, two studies found cf-PWV to predict a decline in MMSE scores over the course of a single year in elderly subjects.19,20 Waldstein and colleagues11 reported a relationship between cf-PWV and a decline in learning and memory performance but not MMSE scores over a maximum of 11 years follow up. Watson et al. found that cf-PWV predicted decline in psychomotor speed but not verbal memory, global function or perceptual speed over 6 years.21 The Rotterdam trial did not find aortic stiffness to be predictive of dementia, but significant associations between arterial stiffness and cognitive performance were reported at baseline.18

Quantitative summary

Across studies, there were sufficient data to examine the association between aortic stiffness (cf-PWV) and cognitive decline (end-point MMSE scores) in meta-analysis. Of the four relevant studies, the raw individual patient data were obtained upon request from both Scuteri et al.20 and Benetos et al.19 Individual patient data were not obtained for the Baltimore11 or Rotterdam18 studies. Aggregate statistics were extracted from these studies, as per a common meta-analysis. The characteristics of those studies included in meta-analysis are presented in Table 2. With the exception of the Baltimore study,11 the other three relevant studies were relatively well matched across most demographical variables of interest. The Baltimore study reported a markedly lower median PWV score and a much lower median age of the cohort. Furthermore, the duration of follow up in the Baltimore study tended to be longer. Although the median MMSE score of the Baltimore cohort is not disclosed, the sample was described as high functioning, which is in contrast to the samples studied by Scuteri et al. and Benetos et al. Given the basic difference between the Baltimore and other three studies included in meta-analysis, the pooled effect size statistics are presented with and without the inclusion of the Baltimore study.

Table 2.  Characteristics of those studies included in meta-analysis
 Scuteri et al.Benetos et al.Poels et al.Waldstein et al.
  1. BMI, body mass index; DBP, diastolic blood pressure; HDL, high-density lipoprotein; Hx, history of; Index; IQR, interquartile range; MAP, mean arterial pressure; Max, maximum; Min, minimum; MMSE, Mini-Mental State Examination; PP, pulse pressure; PWV, pulse wave velocity; —, data not available.

Sample size (n)548662767260
Median duration (months)13 (IQR = 9 to 21)12 (IQR = 12 to 12)Min = 26, Max = 96Max = 132
Female (%)69.879.458.155.7
Median age (years)79 (IQR = 75 to 83)86.8 (IQR = 84 to 91)70.754.3
Education (%)Median years = 5 (IQR = 5 to 10)Primary = 51.5%At least primary = 27.0%Median years of education = 16.8
Secondary = 31.1%
Superior = 11.5%
Median BMI (kg/m2)24.9 (IQR = 23 to 28)26.9
Current smoker (%)9.43.615.120.8
Median total cholesterol (mg/dL)220 (IQR = 206 to 238)227.8203.6
Median HDL cholesterol (mg/dL)58 (IQR = 46 to 69)54.1
Hx Stroke (%)1.914.92.70
Hx Dementia (%)79.200
Median PWV (m/s)13.1 (IQR = 12 to 15)13.4 (IQR = 11 to 17)13.27.1
Median MAP (mmHg)Using DBP + 0.4*PP = 101.8 (93 to 112)Using DBP + 0.4*PP = 96.8 (89 to 107)Using DBP + 0.3*PP = 106.1Using DBP + 0.3*PP = 95.5
Median MMSE24 (IQR = 20 to 28)25 (IQR = 21 to 28)28

Results of the meta-analysis are presented in Figure 2. A pooled estimate was firstly calculated for the two studies, whereby individual patient data were available. Of the original 102 patients in the study by Scuteri et al.,20 it was possible to have individual data for 56 patients, of which, sufficient data for the regression were available for 54 of these. For the PARTAGE19 study, data were provided for all 1128 patients who took part, of which 866 provided enough data to be included in the regression. Analysis for each of these studies was adjusted for age, sex, education, mean arterial pressure and MMSE scores at baseline. The regression coefficient obtained from Scuteri's study was further adjusted for a history of dementia and the exact duration of follow up for each participant. The pooled effect size for these two studies was −0.09 (95% CI: −0.15 to −0.02), indicating that higher baseline aortic stiffness predicted lower MMSE scores at follow up. This pooled estimate was then extended to include that of the Rotterdam study.

Figure 2.

Forest plot showing the longitudinal association between carotid–femoral pulse wave velocity and Mini-Mental State Examination scores. CI, confidence interval; IPD, individual patient data.

The standardised beta coefficient reported in the Rotterdam study18 was originally calculated controlling for age, sex, education, mean arterial pressure, heart rate, current smoking status, diabetes status, body mass index, total cholesterol, high-density lipoprotein cholesterol and intima-media thickness. When the results of the Rotterdam study were added to the meta-analysis, higher aortic stiffness continued to predict lower MMSE scores at follow up (β=−0.04, 95% CI: −0.07 to −0.01, n= 3687). There was no evidence of heterogeneity (I2= 9.5%, P= 0.33).

The standardised beta coefficient reported in the Baltimore Longitudinal Study of Ageing11 was originally calculated controlling for age, sex, education, depression, mean arterial pressure, antihypertensive medications, cardiovascular co-morbidities, heart rate, smoking status, alcohol use, body mass index and total cholesterol. Adding the Baltimore study coefficient to the meta-analysis created significant heterogeneity in the overall effect (I2= 71.9%, P= 0.01), and thus a random effects model was reported. Across all four studies, higher aortic stiffness was found to predict lower MMSE scores within the sample (β=−0.03, 95% CI: −0.06 to 0.01, n= 3947).

Discussion

Of the six relevant longitudinal studies included in review, all but one reported arterial stiffness to be a significant predictor of cognitive decline. Across four studies, there was sufficient data to examine the effects of aortic stiffness on cognitive decline in meta-analysis. Pooled analysis found higher aortic stiffness (cf-PWV) to predict a decline in MMSE scores.

Participant characteristics differed markedly between the Baltimore11 and the other studies included in meta-analysis in terms of both the low median age and PWV score of the cohort. With a median age under 55, participants in the Baltimore study were likely to be performing at ceiling on the MMSE, especially given the education level of the subjects.22 Indeed, this cohort has previously been described as high functioning.11 These differences were highlighted by the presence of statistical heterogeneity when the Baltimore study was added to the pooled analysis. When meta-analysis was completed with the Baltimore study excluded, aortic stiffness was found to predict MMSE scores without evidence of statistical heterogeneity. This association was evident even with the inclusion of the Rotterdam study,18 which was both the largest and the only reviewed study to report no association between arterial stiffness and cognitive decline.

Although the Baltimore study did not find aortic stiffness to predict MMSE scores, significant associations were reported between PWV and a decline in verbal learning and memory performance.11 This suggests that the MMSE lacks the sensitivity to assess overall cognitive changes in this high functioning middle-aged sample. This hypothesis is supported by population norms for the MMSE, whereby middle-aged adults are reported to perform at or close to ceiling.22 The biggest study effect sizes observed in our meta-analysis were for those conducted in older cohorts and with lower MMSE scores at baseline. Our overall pooled effect size suggests that each unit increase in aortic PWV is associated with a decrease of up to 0.06 on the MMSE. Given the insensitivity of the MMSE to detect cognitive changes in younger high functioning subjects, our meta-analysis is likely to underestimate the effects of aortic stiffness on cognitive decline. Tests which target specific cognitive domains appear to be more sensitive than global cognitive measures in assessing the relationship between aortic stiffness and cognitive decline.11,21

Only one of the reviewed studies examined the association between arterial stiffness and the development of dementia.18 In this study, PWV was not found to predict the onset of dementia, although very few participants (n= 11) were diagnosed with vascular dementia during the course of the study. Nevertheless, this result is in conflict with cross-sectional findings where those with both vascular dementia and Alzheimer's disease appear to have higher aortic stiffness than age-matched controls.12 Thus, at present, there is insufficient evidence to state that aortic stiffness is involved in the aetiology of either vascular or non-vascular dementia.

It is widely hypothesised that arterial stiffness causes cognitive deficits because augmented pressure pulses, caused by aortic stiffness, penetrate and damage small cerebral vessels which are not protected against pulsatile blood flow.4 Indeed, research conducted by our own group confirmed that augmented pressures caused by arterial stiffness (measured as aortic pulse pressure and augmentation index) independently predicted cognitive performance.23 Further supporting the earlier hypothesis, arterial stiffness has been associated with cerebral pathology, including stroke,24 white matter hyperintensities and lacunar infarction.25

Although aortic PWV was found to predict cognitive decline across studies, in clinical practice the usefulness of measuring aortic stiffness as a predictor of cognitive decline will ultimately depend on two factors: (i) the availability of efficacious treatments to reduce aortic stiffness and (ii) whether reductions in aortic stiffness, following treatment, translate into improved cognitive outcomes.

With respect to the first factor, numerous therapies have been shown to reduce aortic stiffness, albeit indirectly through decreasing wave reflections and/or aortic systolic pressure. Antihypertensive therapies, such as olmesartan, perindopril and lercanidipine have been shown to reduce central haemodynamic indices, while treatment with atenolol lacks efficacy.26–28 In addition to antihypertensive medications, effects of early wave reflections caused by arterial stiffening1,15,29 may be reduced with lifestyle modification,30 aerobic exercise,31 pharmaceutical medications1,29 and even dietary supplementation.32,33

With respect to the second factor, reducing arterial stiffness will likely downgrade pulsatile pressure through the cerebral vasculature. However, it is unclear whether reductions in aortic stiffness can mitigate or ameliorate the effects of cognitive ageing as this has not been the subject of direct investigation. A recent Cochrane review concluded that there was insufficient evidence to suggest antihypertensive therapy in later life reduces the incidence of cognitive impairment.34 However, this finding should be tempered by the fact that, across these trials, a large proportion of control subjects received antihypertensive treatment, and many patients were lost to follow up.34 Furthermore, although different classes of antihypertensive agents alter brachial BP to a similar degree, their effects on central haemodynamics are varied.26 Thus, when evaluating these trials, it cannot be assumed that central haemodynamic indices were similarly reduced by antihypertensive therapy despite reductions in brachial BP.29 Consequently, future research is required to investigate directly how reductions in aortic stiffness and central haemodynamics impact upon cognitive functioning.

The main limitation of the current review was that our overall meta-analysis is likely to underestimate the effects of aortic stiffness on cognitive decline. The MMSE is insensitive to small changes in cognitive functioning, especially in cognitively intact and young individuals. In line with Waldstein et al.,11 who reported arterial stiffness as a significant predictor of declining neuropsychological function but not MMSE scores, future research in this area should endeavour to implement cognitive tests that are sensitive to subtle differences in cognitive functioning. A further limitation was that individual patient data were not available for two studies included in meta-analysis. For these two studies, the regression coefficients were adjusted for different sets of covariates, meaning that not all regression coefficients were homogeneous. To overcome this limitation, we have presented pooled estimates for all studies together, those studies considered relatively homogeneous as well as the individual patient data estimates only. The current meta-analysis examined the effects of aortic PWV on cognitive decline. We recommend that future studies seek to validate central haemodynamic indices as longitudinal predictors of cognitive decline, given that these indices have emerged as useful predictors of cardiovascular disease and mortality.9 This review also highlights how future research is required to further clarify the impact of arterial stiffening in the aetiology of both vascular and non-vascular dementia.

Conclusion

Arterial stiffness was found to predict cognitive decline in both qualitative review and quantitative analysis. When measuring the cognitive changes associated with aortic stiffening, the MMSE appears to be sensitive in older but not younger high functioning cohorts. Thus, future research in this area is encouraged to implement neuropsychological tests sensitive to subtle cognitive changes. Due to a lack of research in the area, there is insufficient evidence at this time to implicate arterial stiffness in the aetiology of either vascular or non-vascular dementia.

Acknowledgements

The authors would like to thank Professor Scuteri as well as Professors Benetos and Labat of the PARTAGE study for providing individual patient data for the purpose of the current meta-analysis. We also extend thanks to Dr Waldstein and Dr Wendell of the Baltimore Longitudinal Study of Ageing for providing the required summary statistics upon our request.

Ancillary