Cerebral Small Vessel Disease and Renal Function: Systematic Review and Meta-Analysis

Background The small vessel disease (SVD) that appears in the brain may be part of a multisystem disorder affecting other vascular beds such as the kidney and retina. Because renal failure is associated with both stroke and white matter hyperintensities we hypothesised that small vessel (lacunar) stroke would be more strongly associated with renal failure than cortical stroke. Therefore, we performed a systematic review and meta-analysis to establish first if lacunar stroke was associated with the renal function, and second, if cerebral small vessel disease seen on the MRI of patients without stroke was more common in patients with renal failure. Methods We searched Medline and EMBASE for studies in adults with cerebral SVD (lacunar stroke or white matter hyper intensities (WMH) on Magnetic Resonance Imaging (MRI)), in which renal function was assessed (estimated glomerular filtration rate (eGFR) or proteinuria). We extracted data on SVD diagnosis, renal function, demographics and comorbidities. We performed two meta-analyses: first, we calculated the odds of renal impairment in lacunar (small vessel) ischaemic stroke compared to other ischaemic stroke subtypes (non-small vessel disease); and second, we calculated the odds of renal impairment in non-stroke individuals with WMH on MRI compared to individuals without WMH. We then performed a sensitivity analysis by excluding studies with certain characteristics and repeating the meta-analysis calculation. Results After screening 11,001 potentially suitable titles, we included 37 papers reporting 32 studies of 20,379 subjects: 15 of stroke patients and 17 of SVD features in non-stroke patients. To diagnose lacunar stroke, 13/15 of the studies used risk factor-based classification (none used diffusion-weighted MRI). 394/1,119 (35%) of patients with lacunar stroke had renal impairment compared with 1,443/4,217 (34%) of patients with non-lacunar stroke, OR 0.88, (95% CI 0.6-1.30). In individuals without stroke the presence of SVD was associated with an increased risk of renal impairment (whether proteinuria or reduced eGFR) OR 2.33 (95% CI 1.80-3.01), when compared to those without SVD. After adjustment for age and hypertension, 15/21 studies still reported a significant association between renal impairment and SVD. Conclusion We found no specific association between renal impairment and lacunar stroke, but we did find that in individuals who had not had a stroke, having more SVD features on imaging was associated with a worse renal function, which remained significant after controlling for hypertension. However, this finding does not exclude a powerful co-associate effect of age or vascular risk factor exposure. Future research should subtype lacunar stroke sensitively and control for major risk factors.


Introduction
Impaired kidney function is associated with an increased risk of stroke [1] , and reduced estimated glomerular filtration rate (eGFR) is associated with an increased risk of cerebral small vessel disease (SVD) such as white matter hyperintensities (WMH) [2] , and lacunes. It has been proposed that SVD is a manifestation of an underlying multi-system endothelial disorder affecting the small vessels of the kidney, brain, heart, and retina [3] , possibly mediated through inflammation [4,5] .
All studies of renal function and SVD are faced with the challenge of disentangling whether these disorders are a common consequence from shared risk factors or represent a causative relationship. One way of doing this is to compare patients with different stroke subtypes, that is, to compare patients with lacunar (small vessel) stroke to patients with other stroke subtypes, or to compare the prevalence of renal disease in individuals with imaging-determined SVD features. We hypothesised that: patients presenting with a symptomatic recent ischaemic lacunar stroke may have a greater risk of renal impairment than patients with the other ischaemic stroke subtypes that are associated with embolism or large vessel disease. Additionally, individuals without symptomatic stroke may be at increased risk of SVD on imaging if they have renal impairment, after risk factor adjustment.
We performed a systematic review of the literature to establish first the risk of renal impairment in patients with lacunar stroke compared to patients in other ischaemic stroke subtypes, and second, the risk of renal impairment in non-stroke participants with SVD features on imaging (e.g. WMH, lacunes) compared to those without SVD.

Methods
We followed the 'Preferred Reporting Items for Systematic Reviews and Meta-Analyses' (PRISMA) guidelines [6] . We searched MEDLINE (1966-present) and EMBASE (1981-present) using OVID (version OvidSP_UI03.08.01.105), using the search terms in the Supplementary Information -last search was conducted on April 2013. SM and FABC independently reviewed the titles to identify the relevant papers and extracted the data. We resolved the disagreements through mutual discussion and consultation with JMW. We hand-searched the past editions of the journal 'Stroke', and the abstracts of presentations at European Stroke Conferences from 2006-2013 (published in Cerebrovascular Diseases), and the reference lists of relevant review papers. We deemed a paper to be potentially relevant if it included a reference to the measurement of renal function or mentioned SVD in adult humans in the title, and then went on to read the whole paper.
We included studies that measured the renal function in living humans with either symptomatic lacunar stroke or imaging features of SVD. We included studies that described WMH, lacunes, or 'silent cerebral infarcts', as these terms are commonly used to refer to SVD on imaging [7,8] , but excluded atrophy as this was inconsistently reported. We excluded studies that were performed on animals or post mortem, that measured renal disease but did not investigate the renal function (e.g. studies of renal biopsy findings), and studies that only included participants with renal impairment.
We extracted data on study population; location (community or hospital); inclusion and exclusion criteria; diagnosis of stroke subtype; the details of any imaging and image analysis methods; the definition used of SVD; how renal function was measured and defined; blinding; the differences in risk factors between the participants with and without stroke/SVD; the numbers of participants with and without stroke/imaging features of SVD who had renal impairment; and any adjusted or unadjusted summary statistics such as Odds Ratios (OR). We contacted the authors if it was apparent that the data had been acquired but was not reported in the paper.
We considered renal impairment to be either a reduced eGFR (<60 ml/min, Stage 3 Chronic Kidney Disease) or albuminuriaeither micro (30-300 ml/l) or macro (>300 ml/l). For the purpose of the meta-analysis, we used the definitions of renal impairment, lacunar stroke, and SVD from the individual studies within our overall definition.
We carried out two meta-analyses. First, we compared the risk of renal impairment in patients with lacunar stroke to patients in other ischaemic stroke sub-groups. Second, we compared the risk of renal impairment in non-stroke participants with SVD on imaging to those without SVD. The meta-analysis included all the studies in that had dichotomised participants into those with and without renal impairment. If a study gave the mean and standard deviation (SD) of the eGFR, we assumed a normal distribution and calculated the number of patients with an eGFR below 60. We used a random effects model to account for differences in underlying study methodology. We performed analyses using Stats Direct (StatsDirect statistical software version 2.7.9 http://www. stats direct.com. England: StatsDirect Ltd., 2008) and RevMan (Version 5, Cochrane Collaboration).
Each meta-analysis followed the same procedure. We calculated the summary OR of all studies using the Mantel-Haenszel random effects model and assessed heterogeneity using the I 2 statistic. We used a funnel plot to examine for publication bias. To assess the causes of heterogeneity and the risk of bias in individual studies we performed a sensitivity analysis by excluding certain studies with various characteristics and repeating the meta-analysis. Lastly, we examined the summary statistics of the individual studies that had carried out a multivariable analysis accounting for age and hypertension. No protocol was published externally.

Results
We identified 11,001 potentially suitable titles ( fig. 1 ). Of these, we excluded 1,0676 titles because they did not refer to either renal function, or SVD in adults, and read the abstract or full paper of the remaining 325 references. We excluded 246 studies because they did not measure both renal function and SVD, two because they were published only in abstract, 20 because they only included patients with established kidney disease, 14 because they did not report the renal function of patients by stroke subtype and a further two because they only included one particular stroke sub-type. Full details of excluded studies are available on request.
Most studies were from developed countries and varied with respect to inpatient and outpatient recruitment. One study [20] (n = 958) included only young patients with a stroke but the rest included all ages (overall the mean age was 67). All studies measured the renal function as soon as possible after stroke, with the exception of one study of 96 patients which assessed renal function 6-8 weeks post-stroke [14] .

Characteristics of the 13 Studies that Subtyped Stroke
Of the 13 studies that subtyped the stroke, only one [16] used the Oxfordshire Community Stroke Project (OCSP) [44] classification (8 of studies, 7% of sub-typed stroke patients); all other studies used the risk-factor based Trial of Org 10,172 in Acute Stroke Treatment (TOAST) classification [45] . No studies used diffusion weighted MRI (DWI-MRI) in the acute phase. Four studies measured proteinuria [13,14,19,22] (31 of studies, 9% of sub-typed stroke patients), and nine [9, 15-18,
Ten studies [24-26, 28, 29, 32, 34, 36, 38, 39] reported the blood pressure of participants with and without SVD (either as the percent of subjects previously diagnosed with hypertension or as the mean systolic blood pressure on examination), and two studies [24,36] reported little difference between the participants with and without SVD (a difference of less than 2% in the proportion of participants or a difference of less than 5 mm Hg in systolic blood pressure between groups).

Meta-Analysis of the Risk of Renal Impairment in Lacunar Stroke Versus Other Stroke Subtypes
First, we performed a meta-analysis of the studies reporting the numbers of lacunar and non-lacunar stroke patients with renal impairment (defined as proteinuria or an eGFR below 60 ml/min): 12 studies of 5,338 patients [9,10,[13][14][15][16][17][18][20][21][22][23] . We excluded a study of patients who suffered deterioration in renal function after stroke [19] and two studies of stroke patients that measured WMH volume, not stroke subtype [11,12].
Overall there was no specific association between renal function and stroke subtype -lacunar versus non-lacunar: 394/1,119 (35%) of patients with lacunar stroke had renal impairment, compared with 1,443/4,217 (34%) of patients with non-lacunar stroke ( fig. 2 ) OR 0.88, 95% confidence interval (CI) 0.61-1.28. There was a high degree of heterogeneity (inconsistency) with an I 2 of 76%. When comparing patients with lacunar and non-lacunar stroke there was no statistically significant difference in the odds of proteinuria, OR 0.79 (95% CI 0. 38 1 All studies excluded patients unable to have MRI; 2 Calculated from mean and SD assuming a normal distribution. UTI = Urinary tract infection; TOAST = trial of org 10,172 in acute stroke treatment; CKD = chronic kidney disease; CT = computerised topography; MRI = magnetic resonance imaging; OCSP = oxfordshire community stroke project; eGFR = estimated glomular filtration rate; TPA = tissue plasminogen antigen; WMH = white matter hyperintensities; SD = standard deviation; IHD = ischaemic heart disease; TIA = transient ischaemic attack; CSF = cerebrospinal fluid; OR = odds ratio; ACR = albumin creatinine ratio.  Renal impairment more common in non-lacunar stroke Renal impairment more common in lacunar stroke Fig. 2. A forest plot demonstrating the results of the meta-analysis of studies of renal function in patients with lacunar and cortical stroke. Fig. 3. The Odds Ratio of renal impairment in patients with lacunar stroke compared to other stroke sub-type for different subgroups of studies. 2.44-8.82); whereas in studies of patients with a mean age of 70 or greater there was no significant difference between subtypes. In Asian studies (Japan [9,15,17,[21][22][23] , and Bangladesh [10] ) patients with lacunar stroke had a reduced risk of renal impairment OR 0.65 (95% CI 0.49-0.85). Neither the method of stroke sub-typing (OCSP or TOAST), nor a study that recruited only inpatients, affected the lack of association between lacunar stroke subtype and renal impairment.

Meta-Analysis of the Risk of Renal Impairment in Non-Stroke Participants
We included 12 studies [24-26, 32, 34-36, 38-40, 42, 46] of 11,269 participants in this analysis. We excluded studies that only measured Cystatin C [41] or serum creatinine [29] ; those that used microalbuminuria as the dependent variable [33] ; or those that did not dichotomise renal impairment or SVD [30,31,37,47] . For the two studies [35,36] that did not report a total WMH score but instead reported the findings for deep and periventricular WMH separately, we included the data for periventricular lesions in the meta-analysis, as these are more prevalent. Two studies reported the results of eGFR and proteinuria in separate papers [32,39,40,49] ; therefore, we ensured that each participant only contributed once to each calculation.

Unadjusted Odds of Renal Impairment in Participants with and without Imaging Features of Small Vessel Disease
The OR of renal impairment (either eGFR or proteinuria) in participants with SVD compared to those without was 2.33 (95% CI 1.80-3.01), with an I 2 of 78.2% ( fig. 4 )  Studies that recorded 'silent brain infarcts' had a higher degree of heterogeneity (I 2 84%), which may represent the range of different lesions described as 'silent brain infarcts'. Further meta-analysis of the unadjusted data revealed that participants with SVD were twice as likely to have proteinuria compared with participants without SVD: OR 2.00 (95% CI 1.44-2.78) with a moderate degree of heterogeneity (I 2 54.1%); and almost three times as likely to have an eGFR <60, OR 2.82 (95% CI 1.94-4.10), but with a high degree of heterogeneity: I 2 84.4%.
We investigated whether the relationship between SVD and renal function varied in studies of particular groups of participants ( fig. 5 ). For studies of younger patients (average age 50-60) there was a stronger relationship between impaired renal function and SVD (OR 3.19, 95% CI 1.69-6.01) in comparison with studies of patients over 70 (OR 1.53, 95% CI 1.53-1.79). Other study factors had little consistent effect on the relationship between renal function and SVD. A funnel plot (online suppl. fig. B) revealed little evidence of publication bias.

Risk Factor Adjusted Odds of Renal Impairment in Participants with and without Imaging Features of Small Vessel Disease
Nineteen studies [24-32, 34-37, 39, 40, 42, 46, 47, 49] calculated odds ratios adjusted for age, hypertension and a variety of other risk factors; but we could not carry out a meta-analysis of the adjusted ORs as all studies adjusted for slightly different parameters.
After adjusting for (at least) age and hypertension, nine studies [24-27, 31, 35, 42, 46, 47] reported a significant association between renal function and SVD, but with a smaller OR than the unadjusted statistic. However, three studies [28,34,37] reported no significant link between renal function and SVD after adjustment for age and hypertension; one [49] found a significant link between proteinuria and SVD, but not between eGFR and SVD; one [30] found a significant link between renal impairment for participants with periventricular, but not deep WMH, whereas another [36] found that there was a significant link between renal impairment and deep (but not periventricular) WMH.

Discussion
Participants with cerebral SVD features on imaging were found to be at increased risk of renal impairment compared to participants without SVD, but patients with a symptomatic lacunar stroke were at no more risk of renal impairment than patients with a non-lacunar stroke.
An association between lacunar sub-type and renal impairment could have been missed by the studies of stroke patients because the sub-typing (largely based on clinical and CT diagnosis) was not sufficiently precise to distinguish lacunar from non-lacunar stroke [50] . No studies used a gold-standard means of stroke sub-typing, namely risk-factor free clinical sub-typing aided by DWI-MRI in the acute phase. As some studies [51] have suggested that lacunar stroke affects patients at a younger age than non-lacunar stroke, the lack of adjustment for age in the analysis of patients with symptomatic stroke may have masked an association between lacunar stroke subtype and impaired renal function. In a study of younger patients [20] there was a stronger association between impaired renal function and lacunar stroke than with other stroke subtypes, which is interesting as an association between renal function and lacunar stroke may not be present across all age groups as it could be diluted by older patients having more heterogeneous risk factors.
We did not investigate whether the different non-lacunar sub-types were associated with renal impairment as we were investigating the association between small vessel disease and renal impairment.
There was a high level of heterogeneity throughout the literature with different methods of measuring SVD, stroke sub-type, proteinuria, and eGFR. No studies reported differences in risk factors between patients with and without lacunar stroke, which limited our investigation of covariates such as hypertension. Almost all studies measured renal impairment in the acute phase, leading to potential confounding by dehydration, which is common after stroke [52] .
This work has been hampered by the lack of a standardised definition of SVD, with various studies using definitions such as silent brain infarcts, and 'lacunes' to represent similar imaging findings. This problem was compounded by various definitions of proteinuria/albuminuria. Future studies should use the recently published standardised imaging definitions of SVD [7] .
The strengths of this review include a comprehensive literature search incorporating studies from America, Europe and Asia with no language exclusions. In addition to the relationship between reduced eGFR and silent lesions investigated by Vogels et al. [2] we have included patients with proteinuria, symptomatic stroke, and a meta-analysis. Weaknesses comprise the inclusion of only dichotomised studies in the meta-analysis: some more recent studies investigated the continuous relationship between WMH volume and renal impairment. We were not able to fully investigate the effects of powerful confounding factors (e.g. age and hypertension), because it was not possible to carry out a meta-analysis of the adjusted ORs from multivariate analysis as they had all corrected for different confounders.
The apparently strong link between 'silent' SVD and renal impairment in studies of stroke-free patients was not seen in studies of symptomatic stroke. This calls into question the hypothesis that cerebral and renal SVD are directly associated as part of the same multi-system disease rather than representing end organ damage from shared risk factors particularly of hypertension. However, it is difficult to draw firm conclusions due to a high degree of heterogeneity and imprecise stroke sub-typing.
Studies of non-stroke participants should use agematched controls and carry out multivariate analysis of confounding factors. As over 10,000 participants have already undergone MRI and the measurement of renal function, it should be possible to achieve this by re-anal-ysis of the existing data using a well-resourced individual patient data meta-analysis. Future studies of stroke patients should perform careful sub-typing using risk factor-free clinical classification (i.e. OCSP [44] aided by early DWI-MRI), measure proteinuria and eGFR outside the acute phase, and compare with age-matched non-lacunar stroke controls accounting for variations in risk factors.