Zika virus infection as a cause of congenital brain abnormalities and Guillain-Barré syndrome: A living systematic review

Background: The Zika virus (ZIKV) caused a large outbreak in the Americas leading to the declaration of a Public Health Emergency of International Concern in February 2016. A causal relation between infection and adverse congenital outcomes such as microcephaly was declared by the World Health Organization (WHO) informed by a systematic review structured according to a framework of ten dimensions of causality, based on the work of Bradford Hill. Subsequently, the evidence has continued to accumulate, which we incorporate in regular updates of the original work, rendering it a living systematic review. Methods: We present an update of our living systematic review on the causal relation between ZIKV infection and adverse congenital outcomes and between ZIKV and GBS for four dimensions of causality: strength of association, dose-response, specificity, and consistency. We assess the evidence published between January 18, 2017 and July 1, 2019. Results: We found that the strength of association between ZIKV infection and adverse outcomes from case-control studies differs according to whether exposure to ZIKV is assessed in the mother (OR 3.8, 95% CI: 1.7-8.7, I 2=19.8%) or the foetus/infant (OR 37.4, 95% CI: 11.0-127.1, I 2=0%). In cohort studies, the risk of congenital abnormalities was 3.5 times higher after ZIKV infection (95% CI: 0.9-13.5, I 2=0%). The strength of association between ZIKV infection and GBS was higher in studies that enrolled controls from hospital (OR: 55.8, 95% CI: 17.2-181.7, I 2=0%) than in studies that enrolled controls at random from the same community or household (OR: 2.0, 95% CI: 0.8-5.4, I 2=74.6%). In case-control studies, selection of controls from hospitals could have biased results. Conclusions: The conclusions that ZIKV infection causes adverse congenital outcomes and GBS are reinforced with the evidence published between January 18, 2017 and July 1, 2019.


Introduction
The Zika virus (ZIKV), a mosquito-borne flavivirus, caused a large outbreak of infection in humans in the Americas between 2015-2017 (WHO Zika -Epidemiological Update). Since then, the circulation of ZIKV has decreased substantially in the Americas 1 but ZIKV transmission will likely continue at a lower level 2 . Smaller outbreaks have been reported from countries in Africa and Asia, including Angola, India 3 , and Singapore 4 . Regions with endemic circulation, such as Thailand 5 , have the potential for new ZIKV outbreaks with adverse outcomes 6 .
The World Health Organization (WHO) declared ZIKV as a cause of adverse congenital outcomes and Guillain-Barré syndrome (GBS) as early as September 2016 7 , informed by a systematic review of evidence structured according to a framework of ten dimensions of causality, based on Bradford Hill ( . We updated the systematic review to January 18, 2017 as a living systematic review by introducing automated search methods to produce a high quality, up to date, online summary of research 9 about ZIKV and its clinical consequences, for all the causality dimensions 10 .
Since 2017, understanding about the pathogenesis of how ZIKV causes congenital abnormalities has evolved 11,12 . The quality of diagnostic methods, especially for acute ZIKV infection, has also improved 13-15 . More importantly, understanding of the limitations of diagnostic testing, and the need for interpretation in the context of other flavivirus infections, has developed. Important epidemiological questions about the associations between ZIKV infection and adverse congenital outcomes and GBS remain unanswered, however. Much of the early epidemiological evidence, which relied on surveillance data, was limited in use because of issues with the quality of the reporting and case definitions. The reported strength of association between ZIKV and adverse outcomes has varied in studies of different designs and in different settings. Evidence for a doseresponse relationship with higher levels of exposure to ZIKV resulting in more severe outcomes, of clinical findings that are specific to ZIKV infection, or of adverse outcomes caused by different lineages of ZIKV was not found in the earlier systematic reviews.
The objective of this study is to update epidemiological evidence about associations between ZIKV infection and adverse congenital outcomes and between ZIKV and GBS for four dimensions of causality: strength of association, dose-response, specificity, and consistency.

Methods
We performed a living systematic review, which we have described previously 10 . This review updates the findings of the previous reviews 8,10 and will be kept up to date, in accordance with the methods described below. Reporting of the results follows

Eligibility criteria
We considered epidemiological studies that reported original data and assessed ZIKV as the exposure and congenital abnormalities or GBS as the outcomes. We based the exposure and outcome assessment on the definitions used in the publications. We applied the following specific inclusion criteria (Extended data, Supplementary File 2 17 ): Strength of association: at the individual level, we selected studies that included participants both with and without exposure to ZIKV (Figure 1), such as cohort studies and casecontrol studies. At the population level, we included studies that assessed the outcome during the ZIKV outbreak and provided a comparison with pre or post-outbreak incidence of the outcome.
Dose-response relationship: we included studies that assessed the relation between the level of the viral titre or the presence or severity of the symptoms and the occurrence or severity of the outcome.
Specificity of the outcome for ZIKV exposure: we included studies that assessed whether the pathological findings in cases with the outcome are specific for ZIKV infection.
Consistency: we looked at eligible studies to determine the consistency of the relationship between ZIKV exposure and the outcomes across populations, study designs, regions or strains.

Search and information sources
We searched PubMed, Embase, LILACS and databases and websites of defined health agencies (Extended data, Supplementary File 2 17 ). We included search terms for the exposure, the outcome and specific study designs. We also performed searches of the reference lists of included publications. A detailed search strategy is presented in Supplementary File 2. For this review, the search covered the period from January 19, 2017 to July 1, 2019.

Study selection and extraction
One reviewer screened titles and abstracts of retrieved publications. If retained, the same reviewer screened the full text for inclusion. A second reviewer verified decisions. One reviewer extracted data from included publications into piloted extraction forms in REDCap (version 8.1.8 LTS, Research Electronic Data Capture) 19 . A second reviewer verified data entry. Conflicts were resolved by consulting a third reviewer.

Synthesis of evidence
First, we summarised findings for each dimension of causality and for each outcome descriptively. Where available, we calculated unadjusted odds ratios (OR) or risk ratios (RR) and their 95% confidence interval (CI) from published data for unmatched study designs. For matched study designs, we used the effect measure and 95% CI presented by the authors. For publications that presented results for multiple measures of exposure and/or outcome, we compared these results. We applied the standard continuity correction of 0.5 for zero values in any cell in the two-by-two table 20 . We used the I² statistic to describe the percentage of variation across studies that is due to heterogeneity for reasons other than chance 21 . Quantitative synthesis was performed using R 3.5.1 22 . We conducted random effects meta-analyses using the R package metafor (version 2.0-0) 20 . Finally, we compared descriptive and quantitative findings from this review period with previous versions of the review 8,10 .
Searching and screening frequency Daily searches of PubMed, Embase and LILACS are automated and monthly searches are performed manually for other information sources in the first week of the month (Extended data, Supplementary File 2 17 ), with screening of all retrieved publications on the same day. The search strategy consisted of a combination of free terms and MESH terms that identified the exposure and outcomes (Extended data, Supplementary File 2 17 ). Searches from multiple sources were combined and automatically deduplicated by an algorithm that was tested against manual deduplication. Unique records enter a central database, and reviewers are notified of new content.

Frequency of results update
The tables and figures presented in this paper will be updated every six months as a new version of this publication. As soon as new studies are included, their basic study characteristics are extracted and provided online https://zika.ispm.unibe.ch/assets/ data/pub/causalityMap/.

Duration of maintenance of the living systematic review
We will keep the living systematic review up to date for as long as new relevant data are published and at least until October 31, 2021, the end date of the project funding.

Risk of bias/Certainty of evidence assessment
To assess the risk of bias of cohort studies and case-control studies, we compiled a list of questions in the domains of selection bias, information bias, and confounding, based on the quality appraisal checklist of the United Kingdom National Institute for Health and Care Excellence (NICE) and literature 23 .
Two independent reviewers conducted the quality assessment. Disagreements were resolved by a third reviewer.

Results
Search results from January 19, 2017 to July 1, 2019 (Update 2) From January 19, 2017 to July 1, 2019 we screened 1941 publications, of which we included 638 based on title and abstract. After reviewing the full text, 249 publications were included (  131 . The studies assess adverse pregnancy outcomes including infants born with microcephaly, according to exposure to ZIKV for cases. Of these, all studies matched controls, based on gestational age and/or region. During the review period up to January 18, 2017, we included one case-control study 271 , which we replaced with a publication reporting the final results of the study 126 . The meta-analyses incorporate estimates from studies identified in all review periods. In meta-analysis, we found that the odds of adverse congenital outcomes (microcephaly or congenital abnormalities) were 3.8 times higher in ZIKV-infected mothers (95% CI: 1.7-8.7, tau 2 =0.18, I 2 =19.8%, Figure 4). Moreira-Soto et al. (2018) found that in Bahia, Brazil, Chikungunya infection was also associated with being a case 127 .
In this review period, one cohort study reported on strength of association, in 610 pregnant women returning from ZIKVaffected areas in Central and South America to the USA 138 . Maternal ZIKV exposure was measured using RT-PCR or IgM followed by plaque reduction neutralisation test (PRNT). Among the 28 infants born to ZIKV-infected mothers, none was diagnosed with microcephaly and, one was born with a major malformation. In the ZIKV-unexposed group, eight out of 306 had major malformations. A complete overview of different outcomes assessed is presented in the extended data, Supplementary File 3 17 . During the review period up to January 18, 2017, we included two cohort studies, one in women with rash and fever (Brasil et al. In meta-analysis of all three studies, we found that the risk of microcephaly was 3.5 times higher in ZIKV-infected mothers of babies (95% CI: 0.90-13.51, tau 2 =0, I 2 =0%, Figure 5).
Population level: At a population level, data from Mexico collected at different altitudes during the ZIKV outbreak, showed      In this review period, evidence emerged that transmission through sexual contact with infected travellers also resulted in foetal infection 58,59 .

Study designs:
The association between ZIKV infection and congenital abnormalities was consistent across different study designs (Table 2).

Lineages:
We found no new evidence of consistency across different lineages from observational studies. The currently observed adverse congenital outcomes are linked to the ZIKV of the Asian lineage.

Risk of bias assessment
In all case-control studies, uncertainty about the exposure status due to imperfect tests could result in a bias towards the null. Some studies might suffer from recall bias where exposure was assessed by retrospectively asking about symptoms 125,131 . For the cohort studies 138,274 , the enrolment criteria were based on symptomatology. As a result, even in the absence of evidence of ZIKV, the unexposed groups might have had conditions that were unfavourable to their pregnancy. We expect this to bias the results towards the null or underestimate the true effect. Owing to imperfect diagnostic techniques, both false positives (IgM, cross reactivity) and false negatives (due to the limited detection window for RT-PCR) might occur, potentially resulting in bias; the direction of this bias would often be towards the null. None of the studies controlled for potential confounding. Extended data, Supplementary File 4 provides the full risk of bias assessment of the studies included in the meta-analysis 17 .  [255][256][257][258] . In Bangladesh, ZIKV transmission was endemic 259 . Exposure assessment was based on serology 255,256 or a combination of RT-PCR and serology [257][258][259] . Extended data, Supplementary File 3 shows the variability in ORs according to criteria for ZIKV exposure assessment, based on unmatched crude data extracted from each case-control study 17 . Figure 6 shows the association between GBS and ZIKV infection, using the diagnostic criteria that were most similar across studies. Heterogeneity was considerable (I 2 =78.3%), but was reduced slightly after stratification based on the method of selection of controls. The summary OR was higher in studies that enrolled controls from hospital (OR: 55.8, 95% CI: 17.2-181.7, tau 2 =0, I 2 =0%) 258,277 than in studies that enrolled controls at random from within the same community [255][256][257] or from the same household 259 (OR: 2.0, 95% CI: 0.8-5.4, tau 2 =0.46, I 2 =74.6%). Amongst studies with community controls, ORs were lower when enrolment and assessment took place several months after onset of symptoms 255,256 than in studies with contemporaneous enrolment 257,259 . To further illustrate the heterogeneity in exposure assessment between and within the studies, we provide additional aggregations of the data in Extended data, Supplementary File 3 17 . Specificity.  compared Puerto Rican GBS cases reported through public health surveillance that were preceded by ZIKV and cases that were not preceded by ZIKV infection 249 . Clinical features involving cranial nerves were observed more frequently in ZIKV-related cases and, at a sixmonth follow-up visit, residual cranial neuropathy was noted more often in this group. However, clinical symptoms did not allow a distinction to be made between ZIKV and non-ZIKV related GBS.

Study designs:
Across the different study designs, the relation between GBS and ZIKV is consistently shown. Table 2 and Extended data, Supplementary File 3 provide an overview of the included study designs 17 .

Lineages:
We still lack evidence on the consistency of the relation between GBS and ZIKV across different lineages from observational studies. The observed cases of GBS were linked to ZIKV of the Asian lineage.

Risk of bias assessment
Potential selection bias in case-control studies was introduced by the selection of controls from hospitals rather than from the communities in which the cases arose 258,277 . Uncertainty about the exposure status due to imperfect tests would tend to result in a bias towards the null. Two case-control studies did not conduct a matched analysis although controls were matched, and no study controlled for potential confounding by factors other than those used for matching. Exclusion criteria and participation rate, especially of the controls, were poorly reported. Extended data, Supplementary File 4 provides the full risk of bias assessment of the studies included in the meta-analysis 17 .

Discussion
In this living systematic review, we summarised the evidence from 249 observational studies in humans on four dimensions of the causal relationship between ZIKV infection and adverse congenital outcomes and GBS, published between January 18, 2017 and July 1, 2019.

Strengths and limitations
The strengths of this living systematic review are, first, that we automated much of the workflow 10 ; we searched both international and regional databases daily and we screen papers for eligibility as they became available, so publication bias is unlikely. Second, we have quantified the strength of association between ZIKV infection and congenital abnormalities and GBS and investigated heterogeneity of outcome and exposure assessment within and between studies. Third, for congenital outcomes, we included studies with both microcephaly and other possible adverse outcomes, acknowledging the spectrum of congenital adverse outcomes caused by ZIKV. This work also has several limitations. First, we have not assessed the dimensions of the causality framework that involve laboratory studies, so we have not updated the pathobiology of ZIKV complications, which was addressed in the baseline review 8 and the first update to January 2017 10 . Limiting the review to epidemiological domains has allowed more detailed analyses of these studies and we hope that laboratory scientists will continue to review advances in these domains. Second, the rate of publications on ZIKV remains high so, despite the reduced scope and automation, maintenance of the review is time-consuming and data extraction cannot be automated. Third, this review may suffer from continuity bias, which is important for the conduct and interpretation of living systematic reviews and results from changes in the author team. Careful adherence to the protocol will reduce this risk.
Interpretation of the findings ZIKV and congenital abnormalities: Since the earlier versions of the review 8,10 , evidence on the causal relationship between ZIKV infection and congenital abnormalities has expanded. Unfortunately, the total number of cases investigated in the published cohort or case-control studies remains small. In case-control studies in which infants with microcephaly or other congenital abnormalities are compared with unaffected infants, the strength of association differs according to whether exposure to ZIKV is assessed in in the mother (OR 3.8, 95% CI: 1.7-8.7, tau 2 =0.18, I 2 =19.8%) or the foetus/infant (OR 37.4, 95% CI: 11.0-127.1, tau 2 =0, I 2 =0%). This large difference in effect size can be attributed to the fact that not all maternal ZIKV infections result in foetal infection. In cohort studies, the risk of congenital abnormalities was 3.5 times higher (95% CI: 0.9-13.5, I 2 =0%, tau 2 =0) in mothers with evidence of ZIKV infection than without, which is similar to the OR for maternal exposure to ZIKV estimated from case-control studies. Further research is needed to understand the drivers of mother to child transmission. Higher maternal antibody titres were correlated with a higher incidence of adverse congenital outcomes in one case-control study 127 . However, amongst ZIKV-infected mothers followed prospectively, severity of ZIKV infection was not associated with more severe congenital abnormalities 136 . Convincing evidence on a dose-response relation is therefore still lacking.
ZIKV and GBS: Evidence on the causal relation between ZIKV infection and GBS has grown since our last review 10 . The body of evidence is still smaller than that for congenital abnormalities, possibly because GBS is a rare complication, estimated to occur in 0.24 per 1000 ZIKV infections 277 . In this review, the strength of association between GBS and ZIKV infection, estimated in case-control studies, tended to be lower than observed in the first case-control study reported by Cao-Lormeau (2016) in French Polynesia 277 . It is possible that the finding by Cao-Lormeau et al. was a 'random high', a chance finding 278 . Simon et al., however, found a similarly strong association in a case-control study in New Caledonia 258 . In both these studies, controls were patients in the same hospital. Although matched for place of residence, it is possible that they were less likely to have been exposed to ZIKV than the cases, resulting in an overestimation of the OR. In case-control studies in which controls were enrolled from the same communities as the cases, estimated ORs were lower, presumably because exposure to ZIKV amongst communityenrolled controls is less biased than amongst hospital controls 279 . Under-ascertainment of ZIKV infection in case-control studies in which enrolment occurred several months after the onset of symptoms 255,256 is also likely to have reduced the observed strength of association. There is also possible evidence of a dose-response relationship, with higher levels of neutralising antibodies to both ZIKV and dengue in people with GBS 260 . However, the level of antibody titre might not be an appropriate measure of viral titre, and merely a reflection of the intensity of the immune response. Taking into account the entire body of evidence, inference to the best explanation 280 supports the conclusion that ZIKV is a cause of GBS. The prospect of more precise and robust estimates of the strength of association between ZIKV and GBS is low because outbreaks need to be sufficiently large to enrol enough people with GBS. In the large populations that were exposed during the 2015-2017 outbreak, herd immunity will limit future ZIKV outbreaks.

Implications for future research
The sample sizes of studies published to date are smaller than those recommended by WHO for obtaining precise estimates of associations between ZIKV and adverse outcomes This review highlights additional research gaps. We did not assess the complication rates within the infected group in studies without an unexposed comparison group; the adverse outcomes are not pathognomonic for ZIKV infection, making an appropriate comparison group necessary. Although there are no individual features of ZIKV infection that are completely specific, the growing number of publications on ZIKV will allow better ascertainment of the features of a congenital Zika syndrome 281 . In this review, we did not take into account the performance of the diagnostic tests in assessing the strength of association. Future research should include robust validation studies, and improved understanding of contextual factors in the performance of diagnostic tests, including the influence of previous circulation of other flaviviruses, the prevalence of ZIKV and the test used.
This living systematic review will continue to follow studies of adverse outcomes originating from ZIKV circulation in the Americas, but research in regions with endemic circulation of ZIKV is expected to increase. Such studies will clarify whether ZIKV circulation in Africa and Asia also results in adverse outcomes, as suggested by the case-control study of GBS from Bangladesh 259 . Increased awareness might improve the evidence-base in these regions, where misperceptions about the potential risks of ZIKV-associated disease with different virus lineages has been reported 282 . An important outstanding question remains whether the absence of reported cases of congenital abnormalities or GBS in these regions represent a true absence of complications or is this due to weaker surveillance systems or reporting 283 . The conclusions that ZIKV infection causes adverse congenital outcomes and GBS are reinforced with the evidence published between January 18, 2017 and July 1, 2019.

Data availability
Underlying data All data underlying the results are available as part of the article and no additional source data are required. This project contains the following extended data:  Table 1) • Table2.pdf (Most recent version of Table 2) Reporting guidelines PRISMA checklist and flow diagram for 'Zika virus infection as a cause of congenital brain abnormalities and Guillain-Barré syndrome: A living systematic review', https://doi.org/10.7910/ DVN/S7USUI and Figure 2.

Grant information
This work was supported by the Swiss National Science Foundation [320030_170069], end date: 31/10/2021.
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Disclaimer
The views expressed in this article are those of the authors and do not necessarily represent the official positions of the Centers for Disease Control and Prevention.