Association Between Blood Pressure Variability and Cerebral Small-Vessel Disease: A Systematic Review and Meta-Analysis.

Background Research links blood pressure variability (BPV) with stroke; however, the association with cerebral small-vessel disease (CSVD) remains unclear. As BPV and mean blood pressure are interrelated, it remains uncertain whether BPV adds additional information to understanding cerebrovascular morphological characteristics. Methods and Results A systematic review was performed from inception until March 3, 2019. Eligibility criteria included population, adults without stroke (<4 weeks); exposure, BPV quantified by any metric over any duration; comparison, (1) low versus high or mean BPV and (2) people with versus without CSVD; and outcomes, (1) CSVD as subcortical infarct, lacunae, white matter hyperintensities, cerebral microbleeds, or enlarged perivascular spaces; and (2) standardized mean difference in BPV. A total of 27 articles were meta-analyzed, comprising 12 309 unique brain scans. A total of 31 odds ratios (ORs) were pooled, indicating that higher systolic BPV was associated with higher odds for CSVD (OR, 1.27; 95% CI, 1.14-1.42; I2=85%) independent of mean systolic pressure. Likewise, higher diastolic BPV was associated with higher odds for CSVD (OR, 1.30; 95% CI, 1.14-1.48; I2=53%) independent of mean diastolic pressure. There was no evidence of a pairwise interaction between systolic/diastolic and BPV/mean ORs (P=0.47), nor a difference between BPV versus mean pressure ORs (P=0.58). Fifty-four standardized mean differences were pooled and provided similar results for pairwise interaction (P=0.38) and difference between standardized mean differences (P=0.70). Conclusions On the basis of the available studies, BPV was associated with CSVD independent of mean blood pressure. However, more high-quality longitudinal data are required to elucidate whether BPV contributes unique variance to CSVD morphological characteristics.

Clinical Perspective

What Is New?

Blood pressure (BP) variability was associated with cerebral small‐vessel disease independent of mean BP in a pooled meta‐analysis of 27 articles and 12 309 unique brain scans.

There was no evidence of a pairwise interaction between systolic/diastolic and BP variability/mean BP effect sizes.

The strength of evidence was rated as low for all primary outcomes.

What Are the Clinical Implications?

The association between BP variability with cerebral small‐vessel disease, independent of mean BP, may have implications for the monitoring of BP and selection of antihypertensive drugs that minimize BP variability in people with cerebral small‐vessel disease.

The strength of evidence was too low to make definitive clinical recommendations.

Introduction

Epidemiological studies show that cerebral small‐vessel disease (CSVD) subtypes are a common incidental finding from brain imaging among older populations.1, 2, 3, 4, 5, 6 Regularly identified CSVD subtypes include white matter hyperintensities (WMHs), lacunae of presumed vascular origin (LPVO), cerebral microbleeds (CMBs), and enlarged perivascular spaces (EPVSs).7 The heightened risk for stroke, dementia, depression, and mortality attributable to CSVD8, 9, 10 underscores the importance of identifying modifiable vascular risk factors for the development and progression of CSVD to inform vascular risk management among at‐risk populations.11 Principle among vascular risk factors for CSVD, and its clinical sequelae stroke and dementia, is hypertension.12, 13, 14 Randomized controlled trials (RCTs) and cohort studies provide mixed findings as to whether elevated blood pressure (BP) and its management with antihypertensive drugs mitigate CSVD risk and progression of CSVD.15, 16, 17, 18, 19 The inconsistency in findings raise the possibility that factors beyond absolute BP level or treat‐to‐target BP could be important for CSVD development and progression.

A body of empirical work indicates that oscillation in BP between consecutive measures holds additional prognostic significance, alongside mean BP level, to predict subclinical target organ damage, including in the brain.20, 21 Once considered background “noise” or measurement error, intraindividual BP variation, known as BP variability (BPV), holds prognostic value to predict incident and recurrent stroke.22, 23, 24 Higher BPV has indirect effects on the brain, including compromised cerebral autoregulation and transient hypoperfusion.25, 26 Recent interest in BPV's association with brain morphological characteristics27, 28 and dementia,29, 30 and debate surrounding the clinical relevance of BPV to stroke31, 32 and how to define and analyze BPV,33, 34 underscores the topical nature of BPV to brain health.

Previous meta‐analyses on BPV reported associations with acute stroke and transient ischemic attack (TIA),24, 35 headache,36 atrial fibrillation,37 left ventricular mass index,38 mortality,39 cardiovascular outcomes,40 and multiple end points, including, stroke, mortality, and cardiovascular outcomes.41 Other systematic reviews and guidelines have focused on the statistical methods and technical aspects of quantifying BPV.34, 42, 43 This systematic review adds to previous reviews by quantifying the association between BPV with CSVD using meta‐analytic techniques. The objective of this review is to quantify the association between intraindividual BPV and CSVD in populations without recent stroke (subacute <4 weeks). A second objective is to compare the magnitude of association between BPV and CSVD with the effect sizes for mean BP, in studies reporting BPV.

Methods

Search Strategy

This review conforms to the guidelines outlined in the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses statement (Table S1).44 A protocol was registered on PROSPERO (International Prospective Register of Systematic Reviews) (CRD42017081974). A systematic search of the MEDLINE, EMBASE, and SCOPUS electronic databases was performed from database inception without language restrictions; the last search was performed on March 3, 2019 (search string Table S2). Studies from database inception were eligible, as we had no a priori justification to limit studies to a certain era. The authors declare that all supporting data are available within the article and its online supplementary files (Data S1).

Eligibility Criteria

Population

The population of interest included participants aged ≥18 years who had BP quantified using a standardized measure, in which intraindividual BPV (systolic, diastolic, or both) was calculated and participants underwent brain imaging to determine CSVD (defined further below) by computed tomography (CT) or magnetic resonance imaging (MRI). Quantification of mean BP within samples, although desirable for our comparative analysis, was not a prespecified inclusion criterion, nor was the reporting of hypertension or systolic/diastolic load. Likewise, the use of BPV metrics accounting for interdependence with mean BP was not a prespecified inclusion criterion (eg, variance independent of the mean). Studies including people with a past stroke or TIA were eligible if: (1) the measurement of BPV and CSVD was not in the subacute stage (ie, <4 weeks) after a cerebral event requiring hospitalization and (2) the point prevalence of previous stroke or TIA did not exceed 50% of the sample. The rationale was that impaired BP regulation in the acute and subacute stage would likely be a consequence of stroke45 and that the association between BPV with stroke outcomes has already been extensively quantified by meta‐analysis.23, 24

Comparator/control

A bidirectional association was considered between BPV and CSVD. The analyses used comparator groups consisting of people with low BPV versus high BPV for outcome 1 and people without CSVD or with low severity of CSVD for the comparison of BPV means in outcome 2.

Outcomes

1. The association between BPV in any metric (independent variable) with the presence of CSVD determined by brain imaging with CT or MRI (dependent variable). CSVD subtypes were considered in the following categories:

Recent small subcortical infarct: ≤20 mm in the territory of one perforating arteriole.

LPVO: 3‐ to 15‐mm round or ovoid, subcortical, fluid‐filled cavity (similar signal as cerebrospinal fluid), consistent with a previous acute small deep brain infarct or hemorrhage in the territory of one perforating arteriole with hyperintense rim.7

Cerebral white matter changes or hyperintensity: hyperintense on T2‐weighted sequences, isointense or hypointense on T1‐weighted sequences,7 or hypodense on CT,46 determined by either visual grading with a recognized scale (eg, Fazekas, Scheltens, or Rezek)46, 47 or volumetry (fully or semiautomated).

CMBs: small hypointense lesions (round or ovoid blooming; ≤10 mm) that are visible on paramagnetic‐sensitive MR sequences, such as T2*‐weighted gradient‐recalled echo or susceptibility‐weighted sequences.7

EPVSs: also known as Virchow‐Robin spaces, a fluid‐filled space ≤2 mm that follows the typical course of a vessel as it goes through gray or white matter, with similar signal intensity to cerebrospinal fluid.7

2. Mean difference in BPV (dependent variable) between people with and without CSVD (independent variable). Mean differences were extracted by either the presence or the absence of CSVD or severity (eg, on Fazekas scale). All BPV metrics derived from ambulatory BP monitoring (ABPM), home BP monitoring, or visit‐to‐visit assessments were eligible on the basis of Rothwell's definition of intraindividual BPV,22 including but not limited to SD, average real variability, and coefficient of variation.34

Study design

Quantitative studies of any design were eligible, including case‐control, cohort, database registry, or RCT. No restriction was placed on sample size.

Exclusion Criteria

Studies were excluded if the study was performed on animal subjects; BPV was quantified during the acute or subacute stage (ie, <4 weeks) of stroke or TIA; BPV was reported in the context of BP instability, stress reactivity, or postural change, such as experimental designs with a stress challenge, orthostatic change, or head‐up tilt‐table testing; the cohort was designed to test the prognosis of coronary revascularization or renal denervation, or the cohort was designed to assess dementia, Parkinson disease, leukodystrophy, or other degenerative neurological disorder prognosis; the report described a case study; effect size data could not be extracted, calculated, or provided by the study authors; or the study reported only brain atrophy or brain volume. Inconsistencies in brain imaging terminology are well documented.7, 48 To optimally harmonize end points, cerebral infarctions >20 mm diameter were excluded from analyses irrespective of terminology (eg, lacunar infarction and lacunar stroke syndrome).

Literature Screening

Three reviewers (P.J.T., K.G., and M.O.) independently screened titles and abstracts of all the retrieved bibliographical records for eligibility using a data abstraction protocol.49 A hand search was performed of the articles selected for full‐text review and of narrative reviews,50, 51, 52, 53 supplementing the electronic search. In the case of title/abstract review disagreements, the study was subjected to full‐text review. After determination of the study eligibility, data extraction was undertaken independently by 3 reviewers (P.D.P., K.G., and E.J.L.L.) and verified by a fourth reviewer (P.J.T.) to reduce reviewer errors, reduce bias, and achieve consensus. The data extracted pertained to study identification (first author, year of publication, country where recruitment took place, and name of study or trial), study design characteristics (sample size and study design) of the population under study (age, sex, use of antihypertensive medication, and prevalence of stroke), BP and BPV exposure (type of measure, BPV metrics, duration of BP measures, and observation period), CSVD adjudication (subtypes and definition and imaging methods), effect size (most adjusted effect size or raw numbers) adjustment for covariates (list of variables), and funding (grant numbers or acknowledgement).

A request for additional published, unpublished, and in progress articles was sent through the Variability in Blood Pressure and Brain Health consortium (members listed in acknowledgements).54 The principal investigators of eligible studies were contacted to clarify published and unpublished data and duplicate articles. Study authors were contacted via e‐mail to request the unreported data. If the initial e‐mail was not replied to within 2 weeks, a second e‐mail was subsequently sent. Confirmation of ineligible data was provided by 1 author,55 eligibility was confirmed for 2 studies,56, 57 and additional unpublished data were provided by 2 studies.58, 59

Risk of Bias

The RTI item bank was used to identify methodological bias for primary outcomes 1 and 2 for all studies.60 Risk of bias assessment was undertaken by 2 reviewers (E.J.L.L. and P.J.T.) with consensus achieved after discussion. Adjudication of the strength of evidence for the primary CSVD outcome and between‐group differences was made according to the Grading of Recommendations Assessment, Development and Evaluation (GRADE) criteria with GRADE Profiler 3.6.1.61

CSVD Grouping and Outcome Adjudication

1. Primary outcome odds for CSVD attributable to BPV: If BPV's association with CSVD was reported for >1 CSVD subtype, we prioritized the analysis with the largest number of CSVD end points, handling CSVD subtypes in stratified analyses. If models were adjusted, we extracted data from the most adjusted model. In instances where different levels of adjustment were made for mean and BPV data, we prioritized data from the same model to ensure equivalence in covariate adjustment. When >1 BPV metric was reported for CSVD subtypes, we prioritized the most adjusted model and models with the largest number of participants. If multiple metrics were available after applying this hierarchy, we extracted both outcomes and reported in sensitivity analyses.

2. Difference between groups in CSVD: If BPV mean±SD was reported in different CSVD pathological conditions or severities, the subgroup with the highest N was used, prioritizing statistical power over severity,62 ensuring one effect size was analyzed per study rather than collapsing CSVD groups.63 For the mean difference between CSVD groups, we prioritized the most common method of BPV in primary analyses (SD method) and reported other metrics in sensitivity analyses.

1 and 2. BPV data were extracted and prioritized in order, as follows: 24‐hour ABPM, awake ABPM measures (versus sleep), office/clinic/casual (commonly referred to as visit to visit), and home BP monitoring.

Mean BP

Mean BP data were extracted for eligible studies. Odds ratios (ORs) and risk ratios for CSVD were standardized to 10/5–mm Hg increase, multiplying the log of the summary statistic by 10/5, depending on whether systolic BP or diastolic BP was being modeled. It was not possible to standardize the OR effect sizes per SD increase because of inconsistent reporting in the original studies. Between CSVD group differences in BP were modeled with standardized mean differences (SMDs); thus, no standardization was necessary.

Statistical Analysis

Primary outcome 1 was analyzed in R, version 3.5.2, using the metafor package. A multilevel meta‐analysis was performed.64 A key advantage of a multilevel meta‐analysis, compared with a traditional meta‐analysis, is that a single study contributes multiple effect sizes (eg, systolic BP, systolic BPV, diastolic BP, and diastolic BPV). Moreover, a multilevel meta‐analysis accounts for the dependence in effect sizes within a study (ie, between BPV and mean BP, from which BPV is often calculated; and dependence of systolic and diastolic BP). A mixed effects model, with a random intercept per study, tested fixed effect moderators for BP type (diastolic versus systolic) and measure (mean versus variability) and their pairwise interaction. The heterogeneous CSVD subtypes were pooled together in the first instance as each is presumed to have a predominant ischemic origin7 and CSVD is considered a whole‐brain disease.48 Likewise, ABPM, home, and visit‐to‐visit variability data were pooled together, consistent with a previous review,41 as short‐ and longer‐term BPVs are correlated65 and share some postulated mechanisms, such as arterial reflex and compliance, and dosing/titration of antihypertensive drugs.21 Original data pertaining to the likelihood of CSVD were reported as OR and risk ratio with 95% CIs. The estimates were pooled together in a meta‐analysis because they are presumed to measure the same underlying effect,66 and consensus exists that these are approximately equivalent for effect sizes <2.5 and follow‐up <20 years.67 Primary outcome 1 was analyzed stratified by CSVD subtype, BPV metric, day versus night ABPM measures, oscillometric versus auscultatory method of BP measurement, ABPM versus home or visit‐to‐visit BPV, visit‐to‐visit interval <6 and >6 months, use of MRI versus CT with or without MRI, and Asian versus non‐Asian populations in ancillary analyses with RevMan, version 5.3.68 The latter stratification is because of the purported higher morning surge component of BPV among Asian populations.69

The association between BPV and WMHs was modeled with RevMan, version 5.3,68 for d family effect sizes and Comprehensive Meta‐Analysis software70 for r family effect sizes. To demonstrate the association between BPV and the CSVD subtype of WMH, the study effect sizes was reported as r (continuous outcome but no grouping by BPV) or d family (continuous or dichotomized, with grouping by BPV). Because no BPV subgroups were reported in the r family studies, the effect sizes were ineligible for combined pooling and were, thus, pooled separately. The r family effect sizes were pooled with Comprehensive Meta‐Analysis (CMA) software, which converts r values to Fisher Z distribution. In the studies where standardized β values were reported,28, 63, 71, 72 these values were converted to r using the formula outlined by Peterson and Brown73: r=0.98β+0.05λ, where λ is an indicator variable that equals 1 when β is nonnegative and 0 when β is negative. The formula applies to β values between −0.50 and 0.50.

The SMD in BPV between people with and without CSVD was pooled in R, version 3.5.2, using the metafor package, testing mixed effects models with a random intercept per study and fixed effect moderators for BP type (diastolic versus systolic) and measure (mean versus variability) and their pairwise interaction. The effect sizes (Hedges g) were interpreted according to Cohen's74 criteria: 0.2 represents a small effect; 0.5, a moderate effect; and 0.8, a large effect. In studies that reported BPV mean and no SD or SE (eg, interquartile range), we calculated the t‐statistic using the P value with RevMan and then calculated the SMD using the methods outlined by Hedges and Olkin.75 SMD analyses were stratified by BPV metric, day versus night ABPM measures, oscillometric versus auscultatory method of BP measurement, ABPM versus home or visit‐to‐visit BPV, visit‐to‐visit interval <6 and >6 months, use of MRI versus CT, and Asian versus non‐Asian populations in ancillary analyses using RevMan.

Random effects models (inverse‐variance method) were used regardless of statistical heterogeneity under the assumption of high sampling variability between studies, different BPV metrics, and CSVD outcomes.34 Statistical heterogeneity was evaluated with the I2 statistic: I2=0%–60% (not important to moderate), and I2 >60% (substantial statistical heterogeneity).76 Methodological heterogeneity was explored with meta‐regression in relation to the primary outcomes, examining mean or median age, proportion of women, antihypertensive treatment or hypertension, prevalence of previous stroke or TIA, body mass index (kg/m2), and tesla size of MRI (coded 0 for CT). It was not possible to perform a meta‐regression with BP dipping because of inconsistent reporting within studies. The presence of publication bias was evaluated with the test of Egger et al,77 the test of Begg and Mazumdar,78 and Duval and Tweedie's trim‐and‐fill funnel plot.79

Results

Study Selection and Characteristics

The search yielded 8304 citations, from which 51 were reviewed in detail and 27 articles were retained (Table S3 and Figure S1). One study was close to meeting the inclusion threshold80; however, in this study, BPV was reported in relation to white matter integrity and brain health, quantified with fractional anisotropy. The 27 included articles reported on 26 unique samples, comprising a pooled sample of 12 309 unique brain scans, a mean sample size of 473 participants per study, a mean age of 73.0±5.2 years, 48.8±17.6% women, a mean 55.8±27.1% with hypertension or using antihypertensive treatment, a 5.1% stroke prevalence at baseline, and a mean body mass index of 26.2±1.9 kg/m2 (study characteristics in Tables S4 and S5). The included studies originated primarily from Asia (10 studies), Europe (9 studies), and North America (7 studies), with 1 of these recruiting Americans with Japanese ancestry.81 Studies were mostly cohort designs (13 studies), cross‐sectional (6 studies), or case‐control (5 studies). One sample was reported as both cross‐sectional and cohort, and 1 study was RCT design. Most studies used MRI, 3 studies used CT, and only 1 study used MRI and CT. In 1 study, the imaging protocol was unclear. The studies using CT only reported data for analysis of primary outcome 2, SMDs in BPV. Most studies quantified WMH (n=20) or LPVO (n=7), with 6 studies including multiple CSVD outcomes, and few studies quantified CMBs or EPVSs (Table S6).

Assessment of Risk of Bias

The adjudication of risk of bias and precision is provided in Table S7. In 10 studies, inclusion criteria were insufficiently described. Seven studies were deemed underpowered to detect an effect size for the primary outcomes. Most studies (n=21) reported blinded outcome adjudication. Generally, BP was assessed with reliable methods, such as ABPM or clinic visit to visit with standardized protocols; 3 studies were rated as low for description of exposure.

Primary Outcome 1: Association Between BPV and CSVD

Systolic and diastolic BPV and risk for CSVD

A total of 31 estimated ORs were reported: 13 for systolic BPV, 9 for mean systolic BP, 6 for diastolic BPV, and 3 for diastolic BP. Higher systolic BPV was associated with higher odds for CSVD (OR, 1.27; 95% CI, 1.14–1.42; I2=85%), independent of mean systolic BP per 10–mm Hg increase (OR, 1.17; 95% CI, 1. 09–1.25; I2=52%) (Figure 1). Likewise, higher diastolic BPV was significantly associated with higher odds for CSVD (OR, 1.30; 95% CI, 1.14–1.48; I2=53%), independent of mean diastolic BP per 5–mm Hg increase (OR, 1.14; 95% CI, 1.10–1.19; I2=0%). There was no evidence of a pairwise interaction between BP type (diastolic versus systolic) and measure (mean versus variability; P=0.47; Akaike information criterion=−1.5), nor was there evidence of a difference between measures (mean versus variability; P=0.58; Akaike information criterion=−6.6).

Figure 1

Forest plot showing the association between blood pressure variability (BPV) and mean blood pressure (BP) with odds for cerebral small‐vessel disease (CSVD). Data are presented numerically (log odds ratio [OR] and SE) and graphically (forest plot), showing the CSVD risk in each of the included studies and the pooled effect size, stratified by systolic BPV and mean systolic BP per 10–mm Hg increase and diastolic BPV and mean diastolic BP per 5–mm Hg increase. Effect sizes are presented graphically (OR and 95% CI), and those to the right of the vertical line (OR=1) indicate higher odds for CSVD. Data were pooled together using the inverse variance method with random effects. IV indicates inverse variance.

Primary Outcome 2: Between CSVD Group Comparison on Mean BPV

Systolic and diastolic BPV and mean BPV

Fifty‐four SMDs were reported between individuals with and without CSVD: 15 for systolic BPV, 13 for mean systolic BP, 14 for diastolic BPV, and 12 for diastolic mean BP. Systolic BPV was significantly higher in CSVD populations compared with their non‐CSVD counterparts (g=0.21; 95% CI, 0.13–0.28; I2=28%), as was mean systolic BP (g=0.28; 95% CI, 0.18–0.38; I2=44%); however, the effect sizes were small (Figure 2). Likewise, diastolic BPV was higher in people with CSVD compared with their non‐CSVD counterparts (g=0.13; 95% CI, 0.08–0.19; I2=0%), as was mean diastolic BP (g=0.12; 95% CI, −0.00 to 0.24; I2=66%), although effect sizes were small. There was no evidence of a pairwise interaction between BP type (diastolic versus systolic) and measure (mean versus variability; P=0.38; Akaike information criterion=−13.4), nor was there evidence of a difference between measures (mean versus variability; P=0.70; Akaike information criterion=−18.4).

Figure 2

Forest plot showing the standardized mean difference (SMD) between cerebral small‐vessel disease (CSVD) groups on blood pressure variability (BPV) and mean blood pressure (BP). Data are presented numerically (SMD as Hedges g and SE) and graphically (forest plot), showing the between CSVD group difference in each of the included studies and the pooled effect size, stratified by systolic BPV and mean systolic BP and diastolic BPV and mean diastolic BP. Effect sizes are presented graphically (Hedges g and 95% CI), and those to the right of the vertical line (SMD=0) indicate higher systolic BPV or BP in populations with CSVD. Data were pooled together using the inverse variance method with random effects. IV indicates inverse variance.

Meta‐Regression

The meta‐regression results on the primary outcomes are presented in Table S8. Odds for CSVD attributable to systolic BPV were associated with increased age, nonuse of antihypertensives or hypertension, stroke prevalence, body mass index, and increased tesla of MRI. Odds for CSVD attributable to diastolic BPV were associated with younger age and stroke prevalence. Difference between people with and without CSVD in systolic BPV was associated with younger age, whereas difference between CSVD groups in diastolic BPV was associated with female sex; the effect for MRI tesla was marginal (P=0.061).

Publication Bias and GRADE Rating of Primary Outcomes

Evidence of publication bias was suspected for odds of CSVD attributable to systolic and diastolic BPV, evidenced by the test of Egger et al77 and the funnel plot trim‐and‐fill method, although only 6 diastolic BPV studies were retrieved by our systematic review (Table S9 and Figures S2 through S5). GRADE rating of the quality of evidence was low for all outcomes across systolic and diastolic BPV metrics (Table S10).

Ancillary Analyses

The ancillary analyses stratified by CSVD subtype are presented online (Tables S11 and S12). Ancillary analyses by BPV metric, day versus night ABPM measures, ABPM versus home or visit‐to‐visit BPV, visit‐to‐visit interval <6 and >6 months, use of MRI versus CT with or without MRI, and Asian versus non‐Asian populations are reported in Tables 1, 82, 83, 84, 85, 86, 87, 88, 89, 90 and 2.91, 92, 93, 94, 95, 96, 97, 98 The ancillary analyses supported an association between systolic BPV with total and periventricular WMH, but not other subtypes. The association between systolic and diastolic BPV in primary outcomes 1 and 2 was generally replicated and consistent when stratified by different methodological factors. Specifically, primary outcome 1 demonstrated comparable magnitude of ORs when stratified by day versus night and 24‐hour ABPM versus home BP monitoring and visit‐to‐visit BPV. There was, however, evidence of heterogeneity between Asian and non‐Asian systolic BPV effect sizes for primary outcome 1. Likewise, for primary outcome 2, comparable magnitude of SMDs was evident when analyses were stratified by day versus night and 24‐hour ABPM versus home BP monitoring and visit‐to‐visit BPV, as well as Asian versus non‐Asian populations.

Table 1

Ancillary Analysis Results for Odds of CSVD, Stratified by BPV Metric, BP Measurement, and CSVD Assessment

BP Measure	Type of BPV Metric	No. of StudiesReferences	OR	95% CI	I2
Systolic	SD	627, 57, 83, 88, 89, 90	1.28a	1.19–1.39a	0
	Weighted SD	289, 90	1.75a	1.38–2.22a	0
	Coefficient of variation	627, 58, 86, 88, 89, 90	1.36a	1.12–1.67a	88
	Variance independent of the mean	188	1.23	1.05–1.44	···
	Successive variation	188	1.00	0.91–1.10	···
	Average real variability	282, 89	1.18 a	1.03–1.35a	2
	Variance in residuals	281, 84	1.36	0.66–2.79	71
	Oscillometric	1027, 57, 58, 82, 83, 85, 86, 87, 89, 90	1.29a	1.13–1.48a	88
	Auscultatory	381, 84, 88	1.18	0.95–1.46	62
	Dayb	427, 82, 83, 90	1.25a	1.08–1.46a	72
	Nightb	427, 82, 83, 90	1.19a	1.05–1.34a	63
	24‐h ABPM	727, 57, 82, 83, 86, 89, 90	1.26a	1.09–1.47a	82
	HBPM and visit to visit	658, 81, 84, 85, 87, 88	1.27a	1.11–1.45a	70
	Visit‐to‐visit BP interval <6 mo	287, 88	1.24a	1.06–1.44a	0
	Visit‐to‐visit BP interval >6 mo	358, 81, 84	1.21	0.96–1.51	71
	MRI only	1327, 57, 58, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90	1.27a	1.14–1.42a	85
	CT with or without MRI	···	···	···	···
	Primarily Asian	727, 57, 62, 81, 88, 89, 90	1.42a	1.27–1.58a	36
	Non‐Asian	658, 82, 84, 86, 87, 97	1.13 a	1.02–1.26a	73
Diastolic	SD	288, 89	1.24a	1.03–1.50a	0
	Weighted SD	189	1.97	1.06–3.66	···
	Coefficient of variation	427, 58, 88, 89	1.33a	1.10–1.60a	63
	Variance independent of the mean	188	1.20	0.98–1.47	···
	Successive variation	188	1.08	0.95–1.23	···
	Average real variability	189	1.56	0.89–2.73
	Variance in residuals	···	···	···	···
	Dayb	227, 90	1.46a	1.23–1.74a	0
	Nightb	227, 90	1.27a	1.11–1.47a	23
	Oscillometric	427, 58, 87, 89	1.56a	1.19–2.04a	56
	Auscultatory	284, 88	1.15	1.05–1.27	0
	24‐h ABPM	227, 89	1.86 a	1.36–2.55a	0
	HBPM and visit to visit	458, 84, 87, 88	1.22a	1.14–1.31a	13
	Visit‐to‐visit BP interval <6 mo	287, 88	1.35	0.95–1.92	48
	Visit‐to‐visit BP interval >6 mo	258, 84	1.19a	1.09–1.29a	15
	MRI only	627, 58, 84, 87, 88, 89	1.30a	1.14–1.48a	53
	CT with or without MRI	···	···	···	···
	Primarily Asian	327, 88, 89	1.53 a	1.08–2.16a	63
	Non‐Asian	358, 84, 87	1.21a	1.08–1.37a	42

ABPM indicates ambulatory BP monitoring; BP, blood pressure; BPV, BP variability; CSVD, cerebral small‐vessel disease; CT, computed tomography; HBPM, home BP monitoring; MRI, magnetic resonance imaging; OR, odds ratio.

Significant values (when ≥2 studies).

Any metric of BPV recorded from ABPM and subdivided by day and night measures.

Table 2

Ancillary Analysis Results for SMDs in BPV, Stratified by BPV Metric, BP Measurement, and CSVD Assessment

BP Measure	Type of BPV Metric	No. of StudiesReferences	Hedges g	95% CI	I2
Systolic	SD	1227, 56, 57, 72, 82, 88, 89, 90, 91, 93, 94, 95	0.17a	0.09 to 0.26a	9
	Weighted SD	457, 82, 89, 90	0.17a	0.09 to 0.26a	0
	Coefficient of variation	927, 58, 82, 88, 89, 90, 94, 95, 96	0.21a	0.12 to 0.30a	14
	Variance independent of the mean	188	0.13	−0.17 to 0.43	···
	Successive variation	188	0.15	−0.15 to 0.45	···
	Average real variability	282, 89	0.35a	0.18 to 0.53a	0
	Variance in residuals	···	···	···	···
	Dayb	627, 57, 82, 90, 91, 94	0.13a	0.03 to 0.23a	17
	Nightb	627, 57, 82, 90, 91, 94	0.21a	0.08 to 0.35a	35
	Oscillometric	1327, 56, 57, 58, 72, 82, 89, 90, 91, 92, 93, 94, 95	0.22a	0.14 to 0.30a	34
	Auscultatory	288, 96	0.09	−0.14 to 0.32	0
	24‐h ABPM	1227, 56, 57, 72, 82, 90, 91, 92, 93, 94, 95	0.22a	0.12 to 0.31a	38
	HBPM and visit to visit	358, 88, 96	0.21a	0.11 to 0.31a	0
	Visit‐to‐visit BP interval <6 mo	188	0.16	−0.14 to 0.47	···
	Visit‐to‐visit BP interval >6 mo	258, 96	0.18	−0.02 to 0.39	38
	MRI only	1127, 56, 57, 58, 72, 82, 88, 89, 90, 91, 92	0.21a	0.13 to 0.29a	27
	CT with or without MRI	493, 94, 95, 96	0.21	−0.12 to 0.53	48
	Primarily Asian	827, 57, 88, 89, 90, 95, 96, 98	0.20a	0.11 to 0.29a	16
	Non‐Asian	756, 58, 72, 82, 91, 93, 94	0.21a	0.06 to 0.36 a	46
Diastolic	SD	1127, 56, 57, 72, 82, 88, 89, 90, 91, 93, 94	0.07a	0.00 to 0.14a	0
	Weighted SD	357, 89, 90	0.14	−0.03 to 0.31	42
	Coefficient of variation	827, 58, 82, 88, 89, 90, 94, 96	0.20a	0.13 to 0.28a	0
	Variance independent of the mean	188	0.21	−0.09 to 0.51	···
	Successive variation	188	0.29	−0.01 to 0.59	···
	Average real variability	282, 89	0.19a	0.01 to 0.36a	0
	Variance in residuals	···	···	···	···
	Dayb	627, 57, 82, 90, 91, 94	0.06	−0.02 to 0.14	0
	Nightb	627, 57, 82, 90, 91, 94	0.09a	0.01 to 0.18a	0
	Oscillometric	1227, 56, 57, 58, 72, 82, 89, 90, 91, 92, 93, 94	0.14a	0.08 to 0.20a	0
	Auscultatory	288, 96	0.08	−0.20 to 0.36	30
	24‐h ABPM	1127, 56, 57, 72, 82, 89, 90, 91, 92, 93, 94	0.10 a	0.03 to 0.16a	0
	HBPM and visit to visit	358, 88, 96	0.19 a	0.02 to 0.35a	34
	Visit‐to‐visit BP interval <6 mo	188	0.21	−0.09 to 0.51	···
	Visit‐to‐visit BP interval >6 mo	258, 96	0.13	−0.18 to 0.45	67
	MRI only	1127, 56, 57, 58, 72, 82, 88, 89, 90, 91, 92	0.14a	0.08 to 0.20a	1
	CT with or without MRI	393, 94, 96	−0.01	−0.24 to 0.21	0
	Primarily Asian	727, 57, 88, 89, 90, 96, 98	0.10a	0.02 to 0.17a	0
	Non‐Asian	756, 58, 72, 82, 91, 93, 94	0.20a	0.10 to 0.29a	0

ABPM indicates ambulatory BP monitoring; BP, blood pressure; BPV, BP variability; CSVD, cerebral small‐vessel disease; CT, computed tomography; HBPM, home BP monitoring; MRI, magnetic resonance imaging; SMD, standardized mean difference.

Significant values (when ≥2 studies).

Any metric of BPV recorded from ABPM and subdivided by day and night measures.

Discussion

This systematic review and meta‐analysis indicated a higher odds for CSVD attributable to systolic BPV, independent of mean BP. Published data were significant but sparse for diastolic BPV and CSVD risk. When SMDs in BPV were compared between people with and without CSVD, higher systolic and diastolic BPV means were evident in CSVD populations, independent of mean BP. There was no evidence of a pairwise interaction between BP type (diastolic versus systolic) and measure (mean versus variability), nor was there evidence of a difference between measures (mean versus variability) for primary outcomes 1 and 2. Collectively, these findings suggest that BPV is associated with CSVD, independent of mean BP, in the retained articles. The findings were derived from mainly cross‐sectional articles, with moderate to high heterogeneity and suspicion of publications bias; therefore, the GRADE rating of evidence was low for all primary outcomes.

The association between BPV with CSVD reported herein sits alongside previous BPV systematic reviews relating to neurological outcomes, including acute stroke or TIA24 and headache.36 Rothwell and colleagues23 demonstrated the predictive utility of BPV in relation to incident and recurrent strokes by investigating ratio of variances and variance independent of the mean among antihypertensive drug versus placebo RCTs. By contrast, our review included only included 1 RCT, which might explain why the use of antihypertensive drug or hypertension prevalence at baseline was largely unrelated to CSVD effect sizes in the meta‐regression. Differential effects of antihypertensive drug classes on BPV over the short‐ and long‐term are documented, implicating calcium channel blockade and thiazide diuretics in optimal BPV control,24, 99 which might modulate the effects of BPV on CSVD.

In this review and previous systematic reviews,24, 36, 37, 38, 39, 40, 41 the effect sizes for BPV were pooled together concurrently with mean BP effect sizes. However, our meta‐analysis used multilevel modeling to account for within‐study dependence between mean and BPV, as well as systolic and diastolic effect sizes.64 There was generally no evidence for an interaction or heterogeneity between BPV and mean BP pooled effect sizes herein, suggesting that each is associated with CSVD and offers unique variance, and potentially that the magnitude of association with CSVD is comparable between mean and BPV effect sizes. One exception to this finding from ancillary analyses was that mean systolic BP but not systolic BPV increased risk for incident CSVD. Our pooled analyses are, however, limited by only modeling mean BP in the studies investigating BPV. It is likely that including all CSVD studies reporting mean BP, but not necessarily BPV, would quantify heterogeneity between effect sizes more robustly. This could be important given that mean and other indexes of BP (eg, pulse pressure, mean arterial pressure, and morning surge) are associated with CSVD.98, 100

There were sparse published data for CSVD subtypes apart from WMH. WMH data were inconsistently reported and thus pooled separately for r and d family effect sizes. When total as well as deep and periventricular WMH volumes were pooled, the significant association with total WMH was largely attributable to periventricular and not deep WMH volumes. This finding contrasts to the conclusion by Kim and colleagues101 that irregular periventricular WMHs are uniquely associated with chronic hemodynamic insufficiency or hypoperfusion caused by carotid atherosclerosis, whereas deep WMHs are more causatively related to hypertension and CSVD. Further investigation of BPV and regional WMH volumes could be an important avenue of future work, given that periventricular WMH burden is particularly implicated in cognitive decline in processing speed and executive function domains.102, 103 Recent studies indicate that BPV is associated with dementia and cognitive decline,29, 30, 104, 105, 106 raising the possibility that BPV's impact on brain health might be through periventricular white matter pathways susceptible to chronic hemodynamic insufficiency or hypoperfusion.101 As most published BPV data pooled herein relate to WMH, it is important to explore other pathways that lead to the development or progression of CSVD subtypes, such as LPVO,57, 58, 86, 87 CMBs,87 and EPVSs.27

A lack of consensus on BPV methods contributes to substantial heterogeneity between studies with 36 different variability metrics (class, type, and timing) identified in a systematic review of ABPM studies by Taylor and colleagues.34 Moreover, there is no consensus on the optimal metrics to study BPV independent of mean BP and the number of measurements required to quantify visit‐to‐visit BPV for risk stratification.99 Because of this uncertainty in BPV measurement, we pooled different BPV metrics separately in ancillary analyses and found generally consistent results for the most common metrics (SD, coefficient of variation, and average real variability) for primary outcomes 1 and 2. Most studies reported either ABPM or visit‐to‐visit variability. The magnitude of association between systolic BPV and CSVD was comparable in studies adopting ABPM versus home or visit‐to‐visit variability for both primary outcomes. This finding suggests that both short‐ and long‐term systolic BPVs are related to CSVD or that they share similar mechanisms insofar as the relationship with CSVD is concerned. Although the mechanisms underlying BPV are poorly understood, short‐ and longer‐term BPVs are hypothesized to both reflect arterial reflex and compliance and dosing/titration of antihypertensive medications.21 Previously meta‐analytic findings indicate that long‐term BPV is associated with cardiovascular outcomes and mortality, although ABPM evidence was sparse.41 Further direct comparisons of ABPM and long‐term BPV in CSVD are important given that casual cuff BPV measures are less predictive of stroke than those obtained from ABPM107 and the mechanisms remain uncertain.21

This review found mixed evidence for discrete differences in BPV between Asian and non‐Asian populations. The risk for CSVD attributable to systolic BPV was higher among Asian populations; however, there were no observed differences in SMDs of either systolic or diastolic BPV. Kario and Wang69 posited that Asian populations exhibit higher morning surge in BP, a pressor component of BPV. The observed higher odds for CSVD in Asian populations was unlikely caused by method given that a comparable number of studies used 24‐hour ABPM in the primarily Asian and non‐Asian studies (4 versus 3). Potentially, the data derived from new investigations in Asian subpopulations could identify the extent to which BPV and its pressor components differ among people with and without CSVD.108, 109

The strengths of this study include the large pooled sample, meta‐regression, and ancillary analyses by CSVD subtypes, BPV metric, and BP measurement intervals. Several limitations temper the results of this review, including suspicion of publication bias. The included studies were predominantly case‐control or cohort studies and thereby are prone to inherent biases, such as selection bias, attrition, and unmeasured confounding. Examples of unmeasured confounding include genetic risk factors; secondary causes, such as primary aldosteronism, renal artery stenosis, and sleep apnea; and lifestyle factors, such as physical fitness, sodium, potassium, and alcohol intake.110 Moreover, adjustment for antihypertensive medications controlling BP was inconsistent, as was inclusion and adjustment for clinical strokes. The absence of sufficient published data for incident CSVD and CSVD progression indicates that the direction of the association between BPV and CSVD cannot be determined. More specifically, it is possible that higher BPV is merely a consequence of CSVD, rather than a causative risk factor, as others have noted.111 In support, mean systolic and diastolic BP, but not BPV, was associated with incident CSVD herein, although only ≤3 studies analyzed incident CSVD.

Another limitation of this systematic review is that the reporting in the retrieved studies precluded an adequately powered investigation of CSVD subtypes apart from WMH. Less information was available for other CSVD subtypes suspected to result from hypertensive vasculopathy, including deep and infratentorial CMBs112 and EPVSs in the basal ganglia.113 Moreover, we did not include atrophy or brain volume because of their diffuse and sometimes nonischemic causative pathways that are inconsistently reported in epidemiological studies.7 The sparse evidence for CSVD subtypes is partly related to discrepant terminology and brain image interpretation, acquisition, and reporting in the literature.7 Moreover, evidence suggests that WMH and LPVO increase the risk for CMBs pointing to overlap in CSVD burden and potential shared risk factors.1, 2 Close investigation of BPV in relation to CSVD subtypes could be an important area of investigation as a prior meta‐analysis suggests a cumulative effect for increasing number of CSVD markers on stroke outcome.8 Moreover, the meta‐regression pointed to an association between tesla size and CSVD, and studies adopting CT are unable to detect CMBs and EPVSs. We opted to include these predominantly case‐control studies (in primary outcome 2) as their exclusion would lead to upward bias in the pooled results for diastolic BPV. Moreover, access to and use of MRI technology is far from universal, affected by factors such as rural location, patient health insurance, MRI contraindications, and the preference for cheaper neuroimaging techniques.114, 115

In conclusion, higher systolic BPV was associated with higher odds for CSVD, and the BPV means were generally higher among populations with CSVD compared with their counterparts without CSVD. Collectively, the extant BPV literature suggests that the magnitude of association between systolic BPV and CSVD is similar to that reported for mean BP, indicating BPV may contribute independent information and variance to the association with CSVD. Evidence was sparse in relation to diastolic BPV and CSVD subtypes apart from WMH, where it was found that BPV was associated with periventricular WMH especially. Given that these findings cannot rule out that high BPV is a consequence of CSVD, additional large RCTs as well as prospective cohort studies with ABPM, investigating CSVD subtypes with 3‐T MRI, would strengthen the evidence base on whether BPV increases CSVD risk. Given the heterogeneity and suspicion of publication bias in primary outcome 1, an individual participant data meta‐analysis of BP data sets might help answer important questions on hypertensive vasculopathy attributable to BPV in specific CSVD subtypes that are lesser reported, including infratentorial CMBs112 and EPVSs in the basal ganglia.113

Appendix

Variability in Blood Pressure and Brain Consortium Members

Kaarin J. Anstey, PhD, Neuroscience Research Australia, University of New South Wales, Sydney, Australia. Nigel Beckett, MD, PhD, Guy's and St Thomas’ National Health Service (NHS) Trust, London, United Kingdom; Imperial College London, London, United Kingdom. Alexa S. Beiser, PhD, Department of Neurology, Boston University School of Medicine, Boston, MA. Jonathan Birns, BSc, MBBS, PhD, Department of Ageing and Health, Guy's and St Thomas’ Hospital, London, United Kingdom; School of Medicine, Health Education England, London, United Kingdom. Adam M. Brickman, PhD, Department of Neurology, Taub Institute for Research on Alzheimer's Disease and the Aging Brain, College of Physicians and Surgeons, Columbia University, New York, NY. Nicholas R. Burns, PhD, School of Psychology, The University of Adelaide, Adelaide, Australia. Mahir Cengiz, MD, Department of Internal Medicine, Biruni University Faculty of Medicine, Istanbul, Turkey. Suzanne Cosh, PhD, School of Psychology and Behavioural Science, University of New England, Armidale, Australia. Rianne A. A. de Heus, MSc, Department of Geriatric Medicine, Radboud Alzheimer Center, Donders Institute for Brain, Cognition and Behavior, Radboud University Medical Center, Nijmegen, the Netherlands. Peter W. de Leeuw, MD, Subdivision Vascular Medicine, Division of General Internal Medicine, Department of Internal Medicine, Maastricht University Medical Centre, Maastricht University, Maastricht, the Netherlands. Diana Dorstyn, M.Psych(Clin), PhD, School of Psychology, The University of Adelaide, Adelaide, Australia. Merrill F. Elias, PhD, Department of Psychology and Graduate School of Biomedical Science and Engineering, The University of Maine, Orono, ME. J. Wouter Jukema, MD, Department of Cardiology, Leiden University Medical Center, Leiden, the Netherlands. Masahiro Kikuya, MD, PhD, Department of Hygiene and Public Health, Teikyo University School of Medicine, Tokyo, Japan. Abraham A. Kroon, MD, Subdivision Vascular Medicine, Division of General Internal Medicine, Department of Internal Medicine, Maastricht University Medical Centre, Maastricht University, Maastricht, the Netherlands. Rajiv Mahajan, MD, PhD, Centre for Heart Rhythm Disorders, South Australian Health and Medical Research Institute, University of Adelaide and Lyell McEwin and Royal Adelaide Hospitals, Adelaide, Australia. Emer R. McGrath, MD, Department of Neurology, Massachusetts General Hospital; Harvard Medical School, Boston, MA. Eric P. Moll van Charante, MD, Department of General Practice, Department of Social Medicine, Amsterdam Public Health Research Institute, Amsterdam University Medical Center, University of Amsterdam, Amsterdam, the Netherlands. Toshiharu Ninomiya, MD, PhD, Department of Epidemiology and Public Health, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan. Tomoyuki Ohara, MD, PhD, Department of Neuropsychiatry, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan. Takayoshi Ohkubo, MD, PhD, Department of Hygiene and Public Health, Teikyo University School of Medicine, Tokyo, Japan. Emi Oishi, MD, Department of Epidemiology and Public Health, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan. Ruth Peters, PhD, Neuroscience Research Australia, University of New South Wales, Sydney, Australia. Edo Richard, MD, Department of Neurology, Academic Medical Center, Amsterdam, the Netherlands; Department of Neurology, Donders Institute for Brain, Cognition and Behavior, Radboud University Medical Center, Nijmegen, the Netherlands. Michihiro Satoh, PhD, Division of Public Health, Hygiene and Epidemiology, Faculty of Medicine, Tohoku Medical and Pharmaceutical University, Sendai, Japan. Joseph Selvayanagam, MD, Cardiovascular Medicine, Flinders University, and Department of Cardiology, Flinders Medical Centre, Adelaide, Australia, where he is the Director of Non‐Invasive Cardiac Imaging. Sudha Seshadri, MD, Glenn Biggs Institute for Alzheimer's and Neurodegenerative Diseases, University of Texas Health Sciences Center, San Antonio, TX; Neurology, Boston University School of Medicine, Boston, MA; Senior Investigator, the FHS (Framingham Heart Study), Boston, MA. David J. Stott, MD, Institute of Cardiovascular and Medical Sciences, University of Glasgow, Glasgow, United Kingdom. Stella Trompet, PhD, Section of Gerontology and Geriatrics, Department of Internal Medicine, Leiden University Medical Center, Leiden, the Netherlands. Willem A. van Gool, MD, Department of Neurology, Academic Medical Center, Amsterdam, the Netherlands. Tessa van Middelaar, MD, Department of Neurology, Academic Medical Center, Amsterdam, the Netherlands; Department of Neurology, Donders Institute for Brain, Cognition and Behavior, Radboud University Medical Center, Nijmegen, the Netherlands. Deborah A. Turnbull, PhD, School of Psychology, The University of Adelaide, Adelaide, Australia.

Sources of Funding

This work was supported by the Alzheimer's Drug Discovery Foundation (grant RC‐201711‐2014067) and the National Health and Medical Research Council of Australia (grant APP1053578). Funding sources played no role in the writing of the manuscript or decision to submit it for publication.

Disclosures

None.

Data S1.

Table S1. PRISMA Checklist

Table S2. Electronic Search String by Database

Table S3. Reason for Study Exclusion After Full Text Review

Table S4. Descriptive Characteristics of Included Studies

Table S5. Blood Pressure Monitoring Characteristics of Included Studies

Table S6. Imaging Methods for CSVD in the Included Studies

Table S7. RTI Scale Adjudication of Included Studies for Items 1 to 14

Table S8. Meta‐Regression Results for Primary Outcomes

Table S9. Table Showing Tests of Publication Bias for the Primary Outcomes

Table S10. Assessment of Strength of Recommendations Using the GRADE Criteria

Table S11. Ancillary Analysis Results for Association Between BPV and WMH

Table S12. Ancillary Analysis Results for Odds of CSVD Stratified by Sub‐Type

Figure S1. Flow chart of literature search and article selection process.

Figure S2. Funnel plot of odds for CSVD attributable to systolic BPV (trim and fill method).

Figure S3. Funnel plot of odds for CSVD attributable to diastolic BPV (trim and fill method).

Figure S4. Funnel plot of standardized mean differences in systolic BPV between persons with and without CSVD (trim and fill method).

Figure S5. Funnel plot of standardized mean differences in diastolic BPV between persons with and without CSVD (trim and fill method).

Click here for additional data file.