Systematic Review and Meta-Analysis of the Sero-Epidemiological Association between Epstein Barr Virus and Multiple Sclerosis

Background A role for Epstein Barr virus (EBV) in multiple sclerosis (MS) has been postulated. Previous systematic reviews found higher prevalences of anti-EBV antibodies in MS patients compared to controls, but many studies have since been published, and there is a need to apply more rigorous systematic review methods. Methodology/Principal Findings We examined the link between EBV and MS by conducting a systematic review and meta-analysis of case-control and cohort studies that examined the prevalence of anti-EBV antibodies in the serum of cases and controls. We searched Medline and Embase databases from 1960 to 2012, with no language restriction. The Mantel-Haenszel odds ratios (OR) for anti-EBV antibodies sero-positivity were calculated, and meta-analysis conducted. Quality assessment was performed using a modified version of the Newcastle Ottawa scale. Thirty-nine studies were included. Quality assessment found most studies reported acceptable selection and comparability of cases and controls. However the majority had poor reporting of ascertainment of exposure. Most studies found a higher sero-prevalence of anti-EBNA IgG and anti-VCA IgG in cases compared to controls. The results for anti-EA IgG were mixed with only half the studies finding a higher sero-prevalence in cases. The meta-analysis showed a significant OR for sero-positivity to anti-EBNA IgG and anti-VCA IgG in MS cases (4.5 [95% confidence interval (CI) 3.3 to 6.6, p<0.00001] and 4.5 [95% CI 2.8 to 7.2, p<0.00001] respectively). However, funnel plot examination suggested publication bias for the reporting of the anti-EBNA IgG. No significant difference in the OR for sero-positivity to anti-EA IgG was found (1.4 [95% CI 0.9 to 2.1, p = 0.09]). Conclusion/Significance These findings support previous systematic reviews, however publication bias cannot be excluded. The methodological conduct of studies could be improved, particularly with regard to reporting and conduct of laboratory analyses.


Introduction
Multiple Sclerosis (MS) is a complex, chronic, inflammatory, neurological disease that affects the central nervous system [1]. Worldwide there are 2.5 million people with MS [2] and in the United Kingdom (UK) alone the number is estimated to be 100,000 people [3]. The public health burden is huge and related to both direct medical care and loss of productivity from disability that the disease causes. It was estimated that the overall cost of MS in the UK was £1.5 billion in 1999 and a recent study estimated the mean annual cost to be over £30,000 pounds per patient [4].
The aetiology of the disease is still not well understood. Both genetic and environmental factors play roles in the development of the disease [5]. Environmental factors have been an area of intense research lately, particularly into infectious agents that could be linked to MS and many bacterial and viral agents have been studied [6] Of all infectious agents, Epstein Barr virus (EBV) has been most strongly associated with MS [7].
The odds ratio (OR) for MS patients to be EBV sero-negative was found to be 0.06 in a review by Ascherio and Munger [7], and in a more recent meta-analysis combining results from 22 studies the OR was found to be 0.18 [8]. It also appears that the titres of anti-EBV antibodies are significantly higher among sero-positive MS cases when compared with sero-positive controls. Prospective studies suggest this increase in anti EBV antibodies is apparent from 5 to 20 years before the onset of MS [9].
The most recent systematic review that updated the association between MS and sero-positivity for different anti-EBV antibodies was that of Santiago et al. from 2010 [10]. Literature search terms were limited and only publications in English and Spanish were included [10]. This increases the likelihood of publication bias affecting the results. Many studies have been published more recently, which could help to estimate the association more fully. Therefore, we undertook a systematic review and meta-analysis with a wider, more comprehensive search strategy, with no language restriction and including both older and more recently published studies.

Materials and Methods
The methods of the systematic review followed a pre-specified protocol developed by the authors.

Search strategy
Medline and Embase were electronically searched using a comprehensive list of MeSH and Emtree headings and text words that were informed by previous reviews and from search tools in Ovid Medline and Embase (Appendix S1). The search was conducted for articles published from 1960 to March week one, 2012, with no language restriction.
For a study to be included in the review, it had to be a casecontrol or a cohort study, recruited patients with MS diagnosis (confirmed or probable) and controls with no MS diagnosis (healthy or non-healthy). Participants could be from any age group and studies measured serum anti-EBV IgG antibodies (using any method) in both cases and controls. We had few exclusion criteria in order to be as generalisable as possible. Exclusion criteria were non-human studies, very small studies with less than 20 cases or 20 controls and studies that only measured anti-EBV IgM (indicating recent infection).
Titles and abstracts for inclusion were reviewed by one author. A second author reviewed part of the search results (140 out of 1056 reports) and a kappa statistic for the degree of agreement between the two reviewers was calculated and was high at 0.76. Full texts of potentially relevant articles were then screened by one author and uncertainties discussed with a second author. Reference lists from included studies were hand searched to identify any further related studies. Where extra data were required, the author of the article was contacted to obtain further information. Relevant data from each study were initially extracted independently by two authors (23 out of 39 studies). After sufficient agreement was reached, data were extracted by one author and checked by a second author.

Quality assessment
The quality assessment of included studies was based on the Newcastle-Ottawa assessment scale (NOS) [11]. We modified the exposure assessment criteria so that the subcategories would be applicable to serological studies. We added a two stars for blinding of blood sample analysts, one star for conducting the analysis in a clinical laboratory (away from investigators), one star for mentioning explicit laboratory cut-offs for sero-positivity, and one star for reporting the presence or absence of missing data.

Data analysis
The Mantel-Haenzsel odds ratio (OR) of sero-positivity to EBV was calculated for each anti-EBV antibody (anti-Epstein-Barr Nuclear Antigen 1 (anti-EBNA1) or EBNA complex, anti-Viral Capsid Antigen (anti-VCA) and anti-Early Antigen (anti-EA)), and for sero-negativity to EBV using Review Manager Version 5.1 [12]. The ORs were combined in a meta-analysis for each anti-EBV antibody and for overall EBV sero-negativity using a prespecified more conservative random effects model (anticipating study heterogeneity) with a 95% confidence interval (CI) for all analyses. I 2 was used to assess heterogeneity between studies [13]. In meta-analyses with sufficient number of studies, a funnel plot [13] was inspected visually to assess for publication bias.
We conducted subgroup and sensitivity analysis testing of the meta-analyses in Review Manager version 5.1 to compare the OR of sero-positivity for the three anti-EBV antibodies in the following categories: 1. Paediatric versus (vs.) adult studies. We considered studies with median participant's age of 20 or less as paediatric, 2. Studies above median latitude vs. those below median latitude.
All studies were assigned to latitude according to area of recruitment of participants. If the area of recruitment was not clear, studies were assigned to latitude according to the address of the lead author, 3. Studies controlling for age vs. those that did not, 4. Studies controlling for gender vs. those that did not, 5. Studies using McDonald [14]/Poser criteria [15] for MS diagnosis vs. those that did not or criteria used were not clear, 6. Studies that included only cases with confirmed MS diagnosis vs. those that included confirmed and probable cases, 7. Studies that used Immunofluorescence assays (IFA) vs. those which used Enzyme-linked immunosorbent assays (ELISA), 8. Studies with overall quality assessment scores above the median in the modified NOS scale compared to those scoring below the median.
To minimise the effect of multiple comparisons, we used 99% CI for all subgroup and sensitivity analyses.

Results
After removing duplicate reports, the search through Medline and Embase yielded 1056 reports ( Figure 1, adapted from Moher, et al., 2009 [16]). We then excluded case reports, non-human studies, studies with no serological studies performed or with unclear serological results, studies and conference abstracts with no obtainable full reports, retracted studies, studies with an updated version, studies using inappropriate cases (non-MS) or with unclear case definition and studies with cases or controls less than 20. Total number of included studies was 39  ( Table 1). We received extra unpublished data form five studies included in the review [28,29,35,49,52].
The characteristics of participants in the included studies are summarised in Table 1. All studies included were either casecontrol studies or nested case-control in cohort studies. There were 5020 cases and 5844 controls. The median sample size of cases and controls was 108 and 82, respectively. Around half of the studies specified using McDonald or Poser criteria for MS diagnosis (21 out of 39 studies) with cases including confirmed, probable and clinically isolated syndrome (CIS) patients. Controls included healthy and non-healthy participants (patients with neurological and non-neurological diseases), with the majority of studies recruiting healthy control samples (28 out of 39). Most studies recruited hospital controls (24 out of 39). The rest either recruited samples from the community or did not state the source. There were seven paediatric studies [17,19,31,37,41,45,54] and 32 adult studies.

Anti-EBNA IgG
There were 30 studies that tested for anti-EBNA IgG (anti-EBNA1 or EBNA complex IgG) sero-prevalence (for details of type of anti-EBNA IgG measured in each study, see Table 1). We divided the study by Sundstrom et al. [50] into two, retrospective and prospective, making the total number of studies 31. The median sero-positivity in the 31 studies in cases was 98% compared to 88% in controls.
The funnel plot revealed an asymmetrical distribution of studies around the line of identity, indicating the possibility of publication bias ( Figure 3).

Anti-VCA IgG
There were 24 studies that tested for anti-VCA IgG seroprevalence. After dividing the study Sundstrom et al. [50] into two the total number became 25. The median anti-VCA IgG seropositivity in the 25 studies was higher in cases than controls (99% compared to 92%).
The meta-analysis generated an OR of 4.51 (95% CI 2.84 to 7.16, p,0.00001) ( Figure 4). Again there was considerable heterogeneity between studies with I 2 = 59% (p = 0.0001). A funnel plot revealed a slightly less asymmetrical distribution of studies than for anti-EBNA IgG ( Figure 5).

Anti-EA IgG
There were 14 studies that tested for anti-EA IgG seroprevalence. The median sero-positivities in the 14 studies in cases and controls were similar (18% and 16% respectively).
The meta-analysis generated an overall OR of 1.4 (95% CI 0.9 to 2.1, p = 0.09) ( Figure 6 -Note: the study Wagner, et al., (2000) [56] in Figure 6 is an older version of the study Wandinger, et al., (2000) [53]). The heterogeneity higher than that of the previous two analyses with I 2 = 70% (p,0.0001). The funnel plot revealed a more symmetrical distribution of studies, suggesting publication bias is less likely (Figure 7).

Anti-EBV IgG sero-negativity
We considered sero-negativity data as adequate if the authors mentioned explicitly that cases and controls were negative for all anti-EBV anti-bodies (EBNA, VCA and EA) or mentioned that they were EBV negative without specifying which anti-EBV antibody they were negative for. Based on this, only seven studies  provided data for the analysis [17,30,33,38,39,41,53]. The median prevalence of anti-EBV IgG sero-negativity in the seven studies was lower in cases than controls (0.7% compared to 10%). The meta-analysis generated an overall OR of 0.13 (95% CI 0.05 to 0.33, p,0.0001) (Figure 8). Heterogeneity between studies was again significant I 2 = 56% (p = 0.03). Due to the small number of studies in the analysis, a funnel plot was not used to assess the risk of publication bias.

Subgroup analysis
To examine the effect of different factors on the overall OR for anti-EBV sero-positivity, we conducted eight subgroup analyses (Tables 2, 3, 4, 5, 6, 7, 8, and 9), five of which were post hoc (Tables 4, 5, 7, 8, and 9). Of note is that none of the subgroup differences reached statistical significance although there are some trends seen in the results.
Paediatric studies had a slightly higher OR than adult studies for anti-EBNA and anti-VCA sero-positivity (above 5 compared to 4 or below) ( Table 2). Subgroup analysis by the latitude of studies did not reveal a clear trend, with a higher OR for anti-EBNA seropositivity in studies above the median latitude (45.37 north) while the OR for anti-VCA was higher in studies below the median latitude ( Table 3).
Studies that matched for age or sex had higher ORs for all anti-EBV IgG sero-positivities compared to those that did not (Tables 4  and 5). The ORs for all anti-EBV IgG sero-positivities was similar in studies that stated using McDonald/Poser criteria, compared to those that did not specify, or used different criteria (Table 6). However, when we compared studies with confirmed MS cases to those with confirmed and probable/CIS cases, the OR was higher in the former group for anti-EBNA and VCA sero-positivity ( Table 7). Studies that used IFA as method of detection of anti-EBV IgG, had double the OR for anti-VCA and EA seropositivity, but not for anti-EBNA, compared to studies that used ELISA (Table 8).

Quality assessment
In our modified NOS scale the maximum score that could be achieved by a study was 12 stars. The majority of studies scored below half this with an overall median of five. The highest scoring studies were by Ascherio, et al., Levin, et al., and Ponsonby, et al., with score of 11 stars [18,33,42] (Table 10).   For selection criteria, the majority of studies had adequate definitions of cases and controls. However only a quarter of the studies recruited an obviously representative or a consecutive sample of cases, and only six studies recruited adequate community controls without an obvious source of bias. For the comparability criteria, 28 out of 39 studies matched cases and controls for at least one factor, with half of the studies matching for age and at least one additional factor. For the exposure criteria, only a quarter of studies reported blinding sample analysts, with three studies reporting conducting the serology in clinical laboratory (distant from the investigators). Half of the studies reported explicitly serological cut off values with only eight studies reporting whether there were missing data (related to participants' recruitment and withdrawal/analytical failures).
To examine the effect of the quality of studies on the OR for anti-EBV sero-positivity, we compared studies with a quality assessment score above the median (a score of five) with studies scoring at or below the median in a post hoc analysis. The OR of   [56] in Figure 6 is an older version of the study Wandinger, et al., (2000) [53]. doi:10.1371/journal.pone.0061110.g006 sero-positivity was higher for all anti-EBV IgG outcomes in studies scoring above the median (Table 9), although there was no statistically significant difference.

Discussion
This review has again found an association between EBV seropositivity, based on anti-EBNA and anti-VCA IgG testing. We also found no evidence for a difference in sero-prevalence of anti-EA IgG, indicative of recent infection.
There was considerable between study heterogeneity, and we examined different factors that might have been influential. Although none of the subgroup analyses reached statistical significance, several trends emerged. There was a slightly higher OR in paediatric studies compared to adult studies. This might be connected to higher overall exposure of adults to EBV infection than paediatric controls. However, Pakpoor et al. [8] found the OR of sero-negativity to anti-EBV antibodies to be similar in paediatric and adult MS cases. This discrepancy in the findings may be due to chance or the small number of paediatric studies included (only three) by Pakpoor, et al. [8] We did not find a difference in combined OR between studies using more recent criteria (McDonald/Poser) compared to older criteria or failure to report criteria (which may simply reflect lack of reporting). The post hoc analysis according to the diagnosis status of MS participants revealed a higher combined OR in studies recruiting only confirmed vs. confirmed and probable/CIS patients for exposure to anti-EBNA and VCA IgG. This finding is plausible as unconfirmed MS patients are more likely to be misclassified and therefore have a lower prevalence of EBV exposure than true MS patients.
Studies matching for age and/or sex were found to have higher ORs for anti-EBNA and VCA IgG sero-positivity. Other important confounders are geographical location and socioeconomic status. We could not test for the effect of controlling for these factors because few studies explicitly mentioned controlling for them. Comparison of studies according to their latitude did not reveal a trend. In some studies the exact area of    Table 3. Combined OR in studies below median latitude vs. above the median.      Table 9. Combined OR in studies with quality assessment score of 6 (median) or above vs. below 6. recruitment was not mentioned and an estimate had to be used. A recent study by Disanto et al. [57] found that higher latitude was associated with higher prevalence of EBV exposure, independent from MS status.

Specificity of Antibody
Comparing the OR in studies that used IFA to detect anti-EBV antibodies vs. those that used ELISA, revealed that the former had double the OR for anti-VCA and anti-EA IgG sero-positivity but similar results for anti-EBNA. These results are consistent with the analysis by Pakpoor et al. [8] who found that studies that used IFA generated a much lower OR for sero-negativity of anti-EBV antibodies in MS cases compared studies which used ELISA.   Other factors that may contribute to the differences in estimates of ORs are the selection and source of cases and controls. Few studies clearly described the method of selection of cases (consecutive, random, all, or a percentage of the eligible cases). Ideally, the community would be the best source for controls [58]. Only six studies were deemed to have appropriate community controls [18,33,35,42,49,50]. There were seven studies which recruited blood/bone marrow donors, which we considered inappropriate community controls, as such donors are selected for non-health risk behaviour. Looking at the results of six studies which recruited a control sample of patients with other neurological diseases (OND) [23,24,27,28,37,40] and comparing them to seven studies which recruited only seemingly healthy hospital controls [17,19,26,38,39,41,51], we observed that, among the former group, only two out of the six studies found a statistically significant higher anti-EBV IgG sero-prevalence in cases [27,37] compared to all seven studies in the latter group.
At least half of the studies scored poorly in our modified NOS scale for quality assessment. This may partly relate to poor reporting rather than poor conduct. This highlights the importance of following a standardised method for reporting to ensure that all relevant information is reported and to facilitate interpretation and validity assessment of studies. One such method is the STROBE statement (Strengthening the Reporting of Observational Studies in Epidemiology), which was developed to address standardised reporting of observational studies [59]. The NOS scale presently does not allow for uncertainty in quality assessment. Having the option to choose unclear in the criteria for assessment as used by the tool for assessing risk of bias in randomised controlled trials developed by the Cochrane Collaboration [13] would be helpful. Despite the limitation of the NOS quality assessment scale, we did find that studies with better methodological conduct are more likely to find stronger associations between EBV exposure and MS.
One major strength of this review is a more thorough search strategy with no language restriction. Despite this, we only found a single published study from areas other than Europe, North America and Australia (Asian study-Khaki et al. [30]). This emphasises the need for future studies to test the association of EBV with MS in other areas of the world with very different, generally lower prevalences of MS.
For consistent handling of data, we analysed the crude data in our meta-analyses, as very few studies provided adjusted odds ratios [18,19,54], nor adjusted for the same factors. We did not separate out prospective data presented in three studies [18,33,50] from the main meta-analyses, but removing these data has no effect on the results (data available from the authors). The funnel plot for anti-EBNA IgG sero-positivity revealed the possibility of publication bias, but this is unlikely to negate the main association.
Our review includes more than double the number of studies included in the most recently published systematic review by Santiago et al. [10]; the results are similar but more precise.
Although Santiago et al. [10] did assess the quality of included studies, they did not examine the potential for publication bias, and used more limited quality assessment. We found a similar result for sero-negativity to the systematic review by Pakpoor et al. [8] who did not undertake any quality assessment of the studies.
The possible role for EBV as a cause of MS provides the potential to prevent or reduce MS through vaccination against EBV [60]. Further research related to EBV and MS could focus on further characterising the risk of MS based on titres of anti-EBV antibodies. Although there have been studies which examined titre levels and MS risk [18,50,61,62], the data have not yet been combined in a systematic review. Better characterisation of anti-EBV antibodies titres and the associated risk of MS could potentially identify high risk groups and help in prevention or management.
In conclusion, there is strong epidemiological evidence that MS patients have higher sero-prevalence of anti-EBV antibodies compared to controls.

Supporting Information
Appendix S1 MeSH/Emtree headings and text words in search strategy.