Association of herpesviruses and stroke: Systematic review and meta-analysis

Background Herpesviruses induce a range of inflammatory effects potentially contributing to an increased risk of stroke. Objectives To investigate whether patients with infection, or reactivation of, human herpesviruses are at increased stroke risk, compared to those without human herpesviruses. Data sources Six medical databases and grey literature sources from inception to January 2017. Study eligibility criteria Studies where the exposure was any human herpesvirus and the outcome was stroke. We included randomised controlled trials, cohort, case-control, case-crossover and self-controlled case series designs. Methods Meta-analyses when sufficiently homogeneous studies were available. Quality of evidence across studies was assessed. Results We identified 5012 publications; 41 met the eligibility criteria. Across cohort and self-controlled case series studies, there was moderate quality evidence that varicella infection in children was associated with a short-term increased stroke risk. Zoster was associated with a 1.5-fold increased stroke risk four weeks following onset (summary estimate: 1.55, 95%CI 1.46–1.65), which resolved after one year. Subgroup analyses suggested post-zoster stroke risk was greater among ophthalmic zoster patients, younger individuals and those not prescribed antivirals. Recent infection/reactivation of cytomegalovirus and herpes simplex viruses, but not past infection, was associated with increased stroke risk; however the evidence across studies was mainly derived from small, very low quality case-control studies. Conclusions Our review shows an increased stroke risk following zoster and suggests that recent infection or reactivation of other herpesviruses increases stroke risk, although better evidence is needed. Herpesviruses are common and potentially preventable; these findings may have implications for reducing stroke burden.


Introduction
Globally, stroke is the second most frequent cause of death. [1] There is a growing literature indicating that infections, particularly acute respiratory and urinary infections, may play a role in triggering vascular events. [2] Herpesviruses are a family of common viruses persisting latently after primary infection and reactivating periodically. The viruses induce a range of inflammatory effects, [2] potentially contributing to thrombogenesis, atherosclerosis, vasculopathy and platelet activation and thus an increased risk of stroke.
Six previous reviews support an association between herpes zoster (caused by the reactivation of varicella zoster virus (VZV)) and stroke. [3][4][5][6][7][8] One reported a risk ratio of 1.36 (95%CI 1.10-1.67) for the association between zoster and stroke pooled across six cohort studies, [4] whilst the other reviews found around 2-fold increased risk shortly after zoster, which decreased over the following year. [3,[5][6][7] Cytomegalovirus (CMV) is also hypothesised to modulate stroke risk, especially among immunocompromised populations [9] and a recent systematic review concluded that cytomegalovirus infection is associated with an increased risk of cardiovascular disease. [10] Although these reviews have made a significant contribution, there are certain limitations, such as; exclusion of self-controlled case series (SCCS), [4] exclusion of studies among children, [3][4][5][6][7][8] limited subgroup analyses (only one study assessed whether antiviral therapy modified stroke risk) [7] and restricted scope by looking exclusively at clinically apparent zoster and stroke risk. Studies assessing any of the eight herpesviruses known to infect humans and utilising laboratory tests and serological analysis, as well as clinical diagnoses, could also help elucidate the role of latent, sub-clinical or clinical infection and stroke risk.
The primary objective of the systematic review was therefore to investigate whether patients with infection, or reactivation of, human herpesviruses are at increased risk of stroke, cross-sectional studies, ecological studies, case-series, case-reports and reviews. Studies were required to report an effect estimate or the data that allow its calculation. We placed no restrictions on time period, publication status, language, geographical setting or healthcare setting.

Participants
Eligible studies included human participants. No restrictions were placed participants' on age or immunosuppression status.

Exposure
The exposures of interest were infection with, or reactivation of, the eight human herpesviruses: specifically, herpes simplex virus types 1 and 2 (HSV-1 and HSV-2), VZV, Epstein-Barr virus (EBV), CMV, herpesvirus 6, 7, and 8. The exposure definition could be self-reported or a confirmed diagnosis, either through clinical or laboratory criteria. Vaccination against herpesviruses (e.g. Zostavax vaccine) and treatment for herpesviruses (e.g. antivirals) were also considered as effect modifiers, to investigate whether preventing or treating human herpesviruses attenuated stroke risk.

Comparators
Eligible studies were required to include a comparison group of people (or person time for SCCSs or case-crossover) without the herpesvirus exposure of interest.

Outcomes
Studies were included if stroke (first ever or subsequent) was an outcome, clinically diagnosed or self-reported. Those studies meeting the inclusion criteria were additionally assessed for secondary outcomes: TIA [12] and subtypes of stroke (ischaemic versus haemorrhagic).

Information sources
We searched for eligible articles in six databases, originally from dates of inception to January 2017, and then again in July 2018 limited to the years 2017 and 2018. The databases included Cochrane Central Register of Controlled Trials, Embase, Global Health, Medline, Scopus and Web of Science. We additionally searched the clinical trials registers (ClinicalTrials.gov) and grey literature sources, including the New York Academy of Medicine Grey Literature Report (www.greylit.org) and the Electronic Theses Online Service through the British Library (http://ethos.bl.uk).

Search strategy
We searched medical subject heading terms and free text (in the title and abstract) for the concepts 'human herpesviruses' and 'stroke' (combined with the Boolean logic operator AND). Search terms were developed for the database Medline, reviewed by all collaborators and subsequently transcribed into search terms for the remaining databases (supplementary information S1 Appendix for search terms). Reference lists of eligible articles and relevant reviews were scanned for additional papers.

Study selection
Eligibility assessment was performed independently in a blinded standardized manner by two reviewers (CWG and HF); all retrieved titles and abstracts were screened.

Data collection process
Data were extracted using a pre-defined standardised template. Extraction criteria were based on the PICOS [13] (Population, Intervention, Comparator, Outcomes and Study design) framework. As this is an aetiological study, "exposure" replaced "intervention" and "study characteristics" broadened to "study design" (S3 Appendix for all items extracted). We also recorded: the most fully adjusted effect estimates (odds ratios, hazard ratios, incidence rate ratios, risk ratios) for the association between the exposure and stroke; confounders adjusted for; and results of additional analyses relevant to our non-primary objectives. If there were no events in one arm of the study, a continuity correction was applied (adding 0.5 to each cell [13]).

Risk of bias in individual studies
Two authors independently assessed risk of bias in three studies and HF completed the remaining studies. In keeping with the Cochrane Collaborations approach, [14][15][16] a pre-specified set of domains were considered, including bias due to: 1) confounding; 2) selection of participants; 3) differential and non-differential misclassification of exposure and outcome; 4) missing data; and 5) reverse causation. For each domain, a-priori criteria were set-out to assign 'high', 'low', 'moderate', or 'unclear' risk. A summary risk of bias table was produced; when a domain had more than one item the highest risk of bias judgment was used (unless the only item at high-risk was non-differential misclassification, which would bias results toward the null).

Synthesis of results
We synthesised the results into a narrative, grouping studies by herpesvirus exposure and study design; subgroup analyses were also described. We classified exposures as past infection or recent infection/reactivation. IgM and IgA antibodies, and DNA, are present in the blood for a limited period following herpesviruses exposure, therefore their presence suggests recent infection or reactivation (though IgM has poor sensitivity for detecting acute infections and poor specificity in immunosuppressed). [17] Conversely, IgG antibodies, although also raised during an acute infection or reactivation, remain during latent infection, therefore were classified as a past infection. [17] We also presented results from studies of high versus low IgG titre, as high IgG titre may reflect recent reactivation.
When at least two studies assessed the same herpesvirus as a stroke risk factor, meta-analysis was considered. For pooling, we required studies to have identical study designs, the same measurement for the herpesvirus (e.g. IgG seropositivity) and identify the outcome within a similar time-frame. We pooled effect sizes (referred to as "summary estimates") irrespective of the type of effect estimate, due to stroke being rare. Random effects meta-analysis was used throughout, to ensure a consistent approach to all analyses was employed; the I 2 statistic indicated moderate heterogeneity (I 2 >25%) for many subgroups. We investigated sources of heterogeneity (where there were at least three studies in the meta-analysis) by removing studies at high-risk of bias.

Quality of the evidence
The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) [18] approach was used to summarise the quality of cumulative evidence for each herpesvirus on stroke. Evidence was categorised as 'high', 'moderate', 'low' or 'very low' quality, with observational studies starting as 'low'; five reasons to rate down and three reasons to rate up the quality of evidence, were then considered. [19] Full criteria for grading is in S4 Appendix. We assessed publication bias when there were at least 10 studies by creating a funnel plot: effect estimates for the exposure on stroke risk were plotted against standard errors of the log odds, and symmetry was assessed visually and using Begg's test for small-study effects. [20] Ethics As this is a systematic review, ethical approval is not required.

Results
In our initial search 5012 titles and abstracts were screened and 41 observational studies were selected for review (Fig 1). Our updated search retrieved 607 studies, of which seven were selected for review, making a total of 48 studies for the final review.
Study methods and results are summarised in Tables 1 and 2 respectively, risk of bias for individual studies in Table 3 (S5 Appendix for detailed justification) and GRADE assessment in Table 4. Results and meta-analyses are displayed in  17 studies assessed the association between zoster and stroke (1 case-control study, [21] 13 cohort studies [22][23][24][25][26][27][28][29][30][31][32][33][34] and 3 SCCS [35][36][37]) (Table 1). Ten were based in the US or Europe and six in Asia and one in the Middle East; all studies used routinely collected medical records. Two studies involved an immunosuppressed population. 8/17 studies were considered at lowrisk of bias in all domains.  Zoster was associated with a 1.5-fold increased stroke risk four weeks following onset (summary estimate: 1.55, 95%CI 1.46-1.65), with the risk decreasing to baseline after around one year (Fig 2). Removing three studies at high-risk of bias eliminated statistical heterogeneity in cohort studies with "Over 1 year follow-up" (I 2 <0.01%, see S1 Table). There were no SCCS at high-risk of bias. There was moderate quality evidence of an increased risk of stroke following zoster, with evidence upgraded due to some strong associations and a clear dose-response gradient over time.
Two studies reported an increased risk of TIA following zoster. The first showed over 50% increased risk (IRR1.56, 95%CI:1.13-2.15) over a maximum of 10 years follow-up [34] and the second around 15% increased risk (HR1.15, 95%CI:1.09-1.21) during a median follow-up of 6.3 years. [22] Only a single SCCS study assessed the effect of zoster vaccination on stroke risk, using Medicare claims data; this study found no evidence that zoster vaccination attenuated stroke risk, however only 3% of study participants were vaccinated which limited the study's ability to detect an effect. [35] Results can be found in S1, S2, S3, S4 and S5 Figs. Ophthalmic zoster was associated with increased risk of stroke, of a larger magnitude than zoster at any site. The pooled estimate for stroke up to 4 weeks following ophthalmic zoster in SCCSs was 1.77 (95%CI:1.53-2.05), compared to 1.55 (95%CI:1.46-1.65) following any zoster (S1 Fig). Another study found the elevated risk of stroke among rheumatoid arthritis patients experiencing zoster was greatest in those patients with a neurological complication. [33] Antiviral agents appeared to attenuate stroke risk in two out of three studies, though the confidence intervals for effect estimates for zoster patients given and not given antivirals overlapped (S2 Fig). In one SCCS study, in the first four weeks following zoster there appeared to be no evidence of an increased risk of stroke among those given antivirals (IRR1.23, 95%CI:0.89-1.70), whilst for those not given antivirals there was an association (IRR2.14, 95%CI:1.62-2.83). A larger effect of zoster on stroke risk was seen in people aged below 40 years (S3 Fig); there was no difference of zoster on stroke risk by gender (S4 Fig); and little difference in stroke risk by stroke type (ischaemic versus haemorrhagic), except in one cohort study from Taiwan [24] where the magnitude of association was greater for haemorrhagic stroke (S5 Fig).       CMV infection, defined largely using laboratory criteria, was investigated in 22 studies [9,30,[38][39][40][41][42][43][44][45][46][47][48][49][50][51][52][53][54][55][56][57] using data from a variety of settings including electronic healthcare records, survey data and trial data (Table 1).
Among studies assessing CMV infection (past or recent), 19/22 studies had a least one domain at high-risk of bias, including: confounding (ten studies had no age-adjustment) and reverse causation (10 studies recorded CMV following stroke).
Recent CMV infection or reactivation was investigated in 11 case-control studies (Fig 3), using a variety of exposure definitions. In a meta-analysis of two studies, IgM seropositivity was associated with increased stroke risk (summary estimate:5.53,95%CI:2.83-10.81, I 2 <0.001%). When pooling three studies, CMV DNA was also associated with increased stroke risk (summary estimate:2.34,95%CI:0.95-5.74, I 2 = 81.4%). In two of three studies among immunosuppressed patients, clinical CMV reactivation was associated with around 3-fold increased risk of stroke.
There was very low-quality evidence suggesting there is no association between past infection with CMV and stroke and an increased risk of stroke following recent infection/reactivation with CMV. CMV was the only outcome for which sufficient studies were available to assess publication bias; there was no evidence of publication bias (see S6 Fig).
One case-control study assessed the association between HHV-6 and stroke; no association was found. [58] Four case-control studies examined the association between EBV and stroke ( Table 1); [50,51,53,55] three were hospital-based among older adults and one a multi-country study among children (under 18 years). All studies were small (N<500) and at high-risk of bias.
There was very low quality evidence of no association between past infection and an increased risk following recent infection/reactivation with EBV and stroke; the quality of evidence was downgraded due to high-risk of bias and imprecise estimates.
Associations between HSV-1 or HSV-2 and stroke risk were explored in seven studies [50-54, 56, 57] (Table 1) using population survey data and an RCT, and data from a hospital setting. A high-risk of bias was identified in all seven studies. No clear patterns were observed, although there was some indication that recent HSV1 infection/reactivation (IgM seropositivity or IgA high titre) was associated with increased stroke risk.
Two case-control studies [51,59] and one US-community based cohort study [57] assessed the effect of serologically-defined VZV infection on stroke risk (Table 1). Past infection (IgG seropositivity or high titre) was not associated with stroke risk in two studies (Fig 4); quality of evidence was graded very low due to a high-risk of bias. However, a multi-country case-control study among children (under 18 years) found recent infection/reactivation (IgM seropositivity) was associated with increased stroke risk; quality of evidence was graded as low, because although there was a high-risk of bias, the association was very strong.
Varicella and the risk of stroke among children was assessed in three studies from Canada and Europe; [60][61][62] a high-risk of bias was identified in 2/3 studies. Different study designs and time periods during which stroke was recorded were used, therefore estimates were not pooled. However, each study found varicella was associated with a greater risk of stroke within a year from diagnosis. Because of the dose-response gradient over time and very strong associations observed, the evidence was classified as moderate quality. The SCCS study also assessed the association among adults; an increased risk of stroke within 6-months of varicella was found (IR2.13, 95%CI:1.05-4.34). Although this association was strong, the confidence interval was wide, thus the evidence was graded low quality.
Five studies evaluated the short-term effect of VZV vaccination on stroke risk, by comparing vaccinated with unvaccinated people (or person time in the same individuals). One was a multi-country RCT [63] and the others used Canadian or US electronic healthcare records (Table 1); [64][65][66][67] these studies were at very low-risk of bias. No decreased risk of stroke in those vaccinated against varicella or zoster was noted (Fig 4); evidence across studies was graded very low and low quality for varicella and zoster vaccination, respectively.

Discussion
Our review identified 48 studies assessing the association between infection with or reactivation of herpesviruses and risk of stroke. Consistent with previous reviews, there was moderate quality evidence that zoster was associated with a short-term increased risk of stroke, and that increased risk was greatest shortly after zoster (decreasing to baseline by around one year). Some evidence suggested the risk was greater among ophthalmic zoster patients, younger age groups and patients not prescribed antivirals. Moderate quality evidence suggests varicella was associated with increased stroke risk in children. Similar to findings for VZV, there may also be an increased stroke risk with recent CMV and HSV infection/reactivation, however the evidence was very low quality. Finally, there might be an increased stroke risk associated with recent CMV infection or reactivation based on studies carried out in immunosuppressed populations. Two main pathophysiological mechanisms are proposed by which herpesviruses may increase stroke risk. Systemic infection with, or reactivation of, herpesviruses induces acute inflammation, [2] which may lead to endothelial dysfunction accompanied by disruption of atheromatous plaques and hypercoagulability. [68] This biological hypothesis is consistent with our finding that latent herpesvirus infection (that is, presence of viral DNA in host cells without producing infectious viral particles) [69] does not appear to increase stroke risk, as latent infection does not cause acute inflammation in host cells. Herpesviruses may also directly invade cerebral arteries, producing vasculopathy, leading to increased stroke risk; [70] this could explain why younger individuals, normally free from traditional vascular risk factors, were at higher risk of stroke following a recent infection/reactivation of VZV. VZV is the only virus with clear evidence of virus DNA in cerebral arteries; the stronger association between ophthalmic zoster and stroke also supports this hypothesis. CMV is associated with vasculopathy in immunocompromised patients, however the mechanism, and the risk in immunocompetent subjects are unclear. [71] A larger effect of zoster on stroke risk was identified in people aged below 40 years. This has also been reported in a Korean-based cohort study. However, the absolute risk of stroke is low in younger ages, so a large relative effect may be small in absolute terms. This finding, together with the clinical efficacy of the currently available zoster vaccine becoming limited beyond 5-8 years, [72,73] means vaccinating younger age groups may not be cost-effective. This is the first study to systematically review the literature on all eight human herpesviruses as stroke risk factors and the results are broadly in-line with previous review assessing individual herpesviruses and cardiovascular disease (including stroke risk). [3][4][5][6][7][8]10] Strengths included: following a pre-specified protocol; undertaking a comprehensive search; using articles published in any language; and carrying out a complete risk of bias assessment for each study and an assessment of the accumulated evidence using GRADE. Most studies ascertained stroke from pre-existing health care records (n = 33/41), potentially leading to similar stroke definitions across studies. A further strength of this review is that it not only included studies of clinically apparent herpesviruses reactivation, but subclinical reactivation. A further strength of this review is that it not only included studies of clinically apparent herpesviruses reactivation, but subclinical reactivation. It is possible that those with clinical manifestations of reactivated infection (e.g. zoster), or those who are immunosuppressed (as in some CMV studies), may have higher viral titres which plausibly could affect the risk of short term triggering of stroke.
However, limitations included having little data from low-income countries, which make up around 75% of stroke deaths worldwide; [74] whether different populations have different susceptibilities to stroke following herpesvirus infections is unclear. Some meta-analyses combined very few studies, limiting the strength of our pooled results. Overall, the quality of evidence for CMV, EBV and HSV was low or very low.
The studies of VZV, particularly zoster, were well-powered to assess the association between VZV and stroke and rarely suffered from a high-risk of bias; however, subgroup analyses were underpowered, limiting confidence in the findings. Studies of the other herpesviruses (CMV, EBV and HSV) had more limitations; many had small sample sizes, inadequate adjustment for confounders In addition to this, the majority of non-VZV studies relied on laboratory, rather than clinical, identification of possible recent infection or reactivation. The strength of the evidence for zoster and stroke risk lies in the studies all using clear clinical diagnoses of reactivated VZV, which was recorded prior to stroke. In contrast to VZV infection or reactivation which presents with clear clinical symptoms, other herpesviruses may reactivate without any clinical symptoms. Studies that measured markers of infection after stroke may suffer from reverse causality (all but one cases-control study-see Table 3) herpesvirus exposures were defined following stroke and stroke may trigger stress, leading to detection of herpesviruses reactivation after 24 hours (and blood samples were rarely taken immediately after stroke). This may explain why CMV IgG high (versus low) titre was associated with increased stroke risk in most case-control studies, [18] but not cohort studies (in which CMV antibodies were recorded prior to stroke). However, most case-control studies used hospital-based controls, so any stress associated with hospitalisation itself may affect cases and controls equally.
In terms of future research, high-powered cohort or SCCS studies assessing the association between recent infection with, and reactivation of, herpesviruses (aside from VZV), ideally collecting serology samples regularly during follow-up are needed. Furthermore, as zoster vaccination uptake increases, better-powered studies could confirm our findings that vaccination is not associated with a short-term increased stroke risk, and establish whether the vaccine reduces the long-term risk of stroke.
In terms of clinical practice, this review indicated that antivirals might attenuate stroke risk among zoster patients. As patients with more severe zoster are more likely to get antivirals, and also potentially more likely to have a stroke, this might have led to underestimation of their effect through confounding by indication. Antiviral drugs shorten zoster healing time and reduce pain severity [75] therefore these drugs may plausibly reduce stroke risk, by reducing inflammation. Antivirals for zoster are under-prescribed in UK primary care [76] and this review strengthens the argument for better adherence to prescribing guidelines.
Our review highlights that we have a good understanding of a short-term increased stroke risk following VZV infection and reactivation. It also suggests infection and reactivation of other herpesviruses may increase stroke risk, yet better evidence is required. Herpesviruses are common, therefore improved understanding of whether they increase the risk of stroke could provide additional strategies for stroke prevention.