Prevalence of asymptomatic Zika virus infection: a systematic review

1By 25 May 2017, 48 countries and territories in the Americas had confirmed autochthonous, vector-borne transmission of Zika virus disease and 26 had reported confirmed cases of congenital syndrome associated with the infection. Symptoms are often very mild or not present. When symptomatic, the infection may include rash, fever, arthralgia and conjunctivitis. Zika virus infection during pregnancy is a cause of congenital Zika syndrome and it may also be a trigger for Guillain‒Barré syndrome.,

2It has been widely reported that approximately 80% of people with Zika virus infection are asymptomatic. This statement is based on a household survey on Yap State in 2007 that has been cited in many publications on Zika virus. Among 557 residents who provided blood samples, 414 had immunoglobulin (Ig) M antibody against Zika virus and 156 of these (38%) reported an illness that met the definition for suspected Zika virus disease. However, 27 (19%) of the 143 residents who had no detectable IgM antibody against Zika virus also reported an illness that met the definition for suspected Zika virus disease. The authors concluded that, among participants who had IgM antibody against Zika virus, a total of 19% (38% minus 19%) had symptoms that were likely due to the Zika virus infection. When adjusted to the total Yap population aged 3 years or older, the authors estimated that 18% of those infected (95% confidence interval, CI: 10‒27%) had a clinical illness that was probably attributable to Zika virus. From these data we, and other authors, concluded that 82% of the population infected with Zika virus were asymptomatic.

3Lack of signs and symptoms of Zika virus infection does not necessarily imply protection from potential complications, such as microcephaly in babies and Guillain‒Barré syndrome in adults. This has implications for surveillance, treatment and research efforts. For example, an analysis was conducted of pregnancies completed between 15 January and 22 September 2016, and recorded in the United States Zika pregnancy registry. Among women with laboratory evidence of Zika virus infection, there was no difference in the prevalence of birth defects in babies born to asymptomatic (16/271, 6%; 95% CI: 4–9%) or symptomatic women (10/167, 6%; 95% CI: 3–11%). Thus, if the asymptomatic pregnant women had not been included in Zika virus surveillance the 16 babies born with birth defects may not have been attributed to Zika virus.

4Currently, with the exception of asymptomatic pregnant women, only people with suspected infection (i.e. symptomatic) generally undergo laboratory testing for Zika virus infection as part of national surveillance efforts. Thus, the true prevalence of infection and related complications is likely to be underestimated and biased towards those who seek care or develop a viral disease in response to infection. Knowing the prevalence of asymptomatic Zika virus infection is important for assessing the effectiveness and cost‒effectiveness of interventions, including vaccines, to prevent or treat infection. The prevalence is also needed for decision-making about the value of scaling-up surveillance efforts.

5The aim of the current review was to estimate the prevalence of asymptomatic Zika virus infection in the general population and in specific population groups from observational epidemiological studies.

6We used systematic review methods, including a meta-analysis., We registered the protocol on the International prospective register of systematic reviews (CRD42017059342) and followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis statement for reporting.

7We included general or specific population-based studies of participants of all ages and from any country: pregnant women, newborns and infants, children, adults, newborns with congenital abnormalities, and adults with Guillain‒Barré syndrome and other neurological diseases.

8We included studies if exposure to Zika virus was identified, using molecular or serological methods. We used the Pan American Health Organization (PAHO),World Health Organization (WHO) guidelines for laboratory testing wherever possible., For a confirmed case these guidelines require: (i) presence of ribonucleic acid or Zika virus antigen in any specimen (serum, urine, saliva, tissue or whole blood) tested by reverse-transcriptase polymerase chain reaction method; or (ii) positive anti-Zika virus IgM antibodies and plaque reduction neutralization test for Zika virus titres ≥ 20 and four or more times higher than for other flaviviruses; and exclusion of other flavivirus; or (iii) in autopsy specimens, detection of the viral genome (in fresh or paraffin tissue) by molecular techniques, or detection by immunohistochemistry. In practice, this definition was often not used in studies, especially in earlier research. We therefore included studies using alternative definitions for positive laboratory testing if the definition was clearly stated. One alternative definition was the PAHO‒WHO guideline for probable cases: presence of Zika IgM antibodies, with no evidence of infection with other flaviviruses.

9We defined the primary outcome measure as percentage of all Zika virus-positive participants who were asymptomatic at the time of laboratory testing, or within 7 to 10 days of testing. The denominator was all participants who were Zika virus-positive. For the numerator, the PAHO‒WHO guidelines for signs and symptoms were used wherever possible, which require patients to have rash (usually pruritic and maculopapular) with two or more of the following signs or symptoms: fever, usually < 38.5 °C; conjunctivitis (non-purulent/hyperemic); arthralgia; myalgia; and/or periarticular oedema. In practice, not all studies used the PAHO‒WHO definition and we included studies using alternative definitions for symptoms if a clear definition was provided. Asymptomatic Zika virus-positive participants were those with no symptoms or with symptoms that did not meet the definition used for the particular study.

10We included cross-sectional seroprevalence studies, cohort studies of pregnant women, cohort studies of newborns and infants, case‒control studies of Guillain‒Barré syndrome and other neurological diseases, case‒control studies of microcephaly and case series with at least 20 participants. The cut-off value of 20 participants for case series was chosen as a reasonable minimum number for which prevalence data can be reported. A cross-sectional seroprevalence study in the general population is the most appropriate design to determine the prevalence of asymptomatic Zika virus infection. However, to make use of the limited information that was available, we chose to include other study designs and other populations. Published and completed unpublished studies were eligible for inclusion. Data from ongoing studies were also eligible for inclusion when results from a representative sample were available.

11Publications in English, French, Spanish or Portuguese were included. There was no restriction on year of publication.

12We excluded studies in which having symptoms of Zika virus was a criterion for inclusion of participants in the study. This is because it would give a biased value for percentage asymptomatic of 100% solely due to the inclusion criteria. We also excluded studies where the percentage of participants who were asymptomatic could not be determined.

13The search strategy and keywords used are shown in Box 1. The titles and abstracts of these references were checked by one author against the inclusion criteria. Additional published articles were also identified through separate manual searches of PubMed® and revision of Zika virus article alerts by another author. The full text of any potentially relevant papers were checked by a second author and disagreements resolved by discussion and consultation with a third author. Papers excluded after review by a second reviewer and discussions between reviewers were detailed in a table, together with the reason for their exclusion. We also made contact (by email or in-person at key Zika virus meetings) with known research groups conducting cross-sectional studies of Zika virus. These groups were identified through the PAHO‒WHO Zika virus research platform, which includes research protocols that detail ongoing research related to the virus.

14We searched PubMed®, Embase® and LILACS online databases from inception to date of search (4 November 2016, updated 7 March 2017 and 26 January 2018) using the term “zika” as text word for PubMed® and LILACS and “zika” as keyword (zika.mp) for Embase® (Ovid). References were imported into EndNote version X7 reference management software (Clarivate Analytics, Philadelphia, United States of America). The search was then limited using the terms: (cohort OR case control OR case-control OR series OR prospective OR retrospective OR longitudinal OR cross-sectional OR cross sectional OR observational OR transversal OR seroprevalence OR prevalence OR asymptomatic) in any field and then checked for duplicates.

15We extracted qualitative information into a Word version 14 table and quantitative data into an Excel version 14 spreadsheet (Microsoft Corporation, Redmond, USA). One author extracted the data and another author checked it: disagreements were resolved by discussion and consultation with a third author where necessary. We extracted the following data: country of study; region within the country; study design (cross-sectional, cohort, case‒control, case series); population (all ages, pregnant women, newborns and infants, newborns with congenital abnormalities, adults, adults with Guillain‒Barré syndrome); age range; period of study; definition of Zika virus positive according to laboratory tests; definition of symptomatic and asymptomatic Zika virus; preferential recruitment of participants with symptoms (yes/no); sample size calculation; and comments.

16Quantitative data extracted included: response rate; total number of participants; total number classified as Zika virus positive; number of Zika virus-positive participants classified as symptomatic and as asymptomatic; and percentage of the total sample who were symptomatic at time of recruitment. For the cohort studies we used Zika virus-positive status at any time during the pregnancy (for studies of pregnant women) or any time during the study (for studies of newborns and infants). We extracted quantitative data for relevant subgroups where the data and sample size allowed, including for population subgroups and different definitions of Zika virus exposure.

17The quality of the included studies was assessed independently by two authors using the critical appraisal checklist for prevalence studies, developed by The Joanna Briggs Institute. This tool includes the same dimensions as the Assessing Risk of Bias in Prevalence Studies tool, but was considered more useful for this review as it is applicable to a variety of study designs. The Joanna Briggs Institute tool also includes extra items related to sample size and subgroups. Disagreements were resolved by discussion and consultation with a third author where necessary.

18We summarized the findings from the included studies in numerical and narrative tables. We conducted quality-effects meta-analysis using MetaXL version 5.3 (Ersatz, EpiGear International, Sunrise Beach, Australia) and the double arcsine transformation of prevalence.– We assessed heterogeneity using the Q and I2 statistics. We used Doi plots and the Luis Furuya‒Kanamori index to evaluate the presence of small-study effects, where asymmetry can indicate publication or other biases. A symmetrical mountain-like plot with values of the Luis Furuya-Kanamori index within ± 1 indicates no asymmetry; between ± 1 and ± 2 indicates minor asymmetry; and exceeding ± 2 suggests major asymmetry. Due to the high degree of heterogeneity in the results, we also checked whether the heterogeneity could be explained by population subgroups. The number of included studies was insufficient for testing multiple subgroups. We also tested the sensitivity of the results to excluding the largest study and to using the actual sample figure, rather than the population estimate reported by the authors that accounts for symptoms not attributable to Zika virus infection.

19We identified a total of 960 records from database searches and another 12 records through other sources (Fig. 1). No unpublished or in-process studies were identified. After screening, we assessed 64 full-text articles for eligibility (Fig. 1) and excluded 36 articles– for various reasons (Table 1). No studies were excluded due to language restrictions. A total of 23 studies from 28 articles met the inclusion criteria for the review (Table 2; available at: http://www.who.int/bulletin/volumes/96/6/17-201541).,,–

20We found only three cross-sectional seroprevalence studies of the general population, which are considered to be the most appropriate design to measure prevalence. These included the original study of Yap State residents, Federated States of Micronesia, conducted in 2007, a study of the general population and schoolchildren in French Polynesia conducted in 2014–2015 and a study in 2016 of the general population living near 19 index cases in San Juan, Puerto Rico. The majority of the studies were case series from population health surveillance programmes,,,,, systematic screenings of an at-risk population, or hospital-based screenings of an at-risk population.,,,,, A cohort design was used in four studies,,,, a case‒control design in two studies,, and a cross-sectional study of blood donors in one study (Table 2).

21There was considerable variation in the methods of laboratory testing and the definitions of Zika virus positivity used in the studies (Table 2). Also, few studies offered a definition for symptomatic or asymptomatic. Sample sizes in studies varied from 30 to over 9000 (Table 3).

22The risk of bias scores ranged from 1 to 9 out of a possible total of 10, with a mean score of 5.8 (Table 2). The most common limitations were: sample not clearly representative of the population (18 studies); response rate not reported, or large number of non-responders (19 studies); and not accounting for confounding factors or failure to identify subgroup differences (17 studies). The three cross-sectional seroprevalence studies of the general population had risk of bias scores between 6 and 8.

23The 23 studies included a pooled number of 11 305 participants positive for Zika virus, 6921 of whom were asymptomatic. Meta-analysis showed a combined prevalence of asymptomatic Zika virus of 61.8% (95% CI: 33.0–87.1%). However, there was substantial heterogeneity (Q = 3291, P < 0.001, I2 = 99%), suggesting that the pooled prevalence is probably not a robust estimate. Analysis based on subgroups of the population (general population, returned travellers, blood donors, adults with Guillain‒Barré syndrome, pregnant women or babies with microcephaly) was not able to explain the heterogeneity (Fig. 2). There was also significant heterogeneity within all subgroups.

24Both the funnel plot (Fig. 3) and Doi plot (Fig. 4) showed major asymmetry. The most likely explanations for the asymmetry are selection bias, including publication bias, or true heterogeneity in the included studies. The largest study (population-adjusted sample: 6892; actual sample: 557) had a weight of 40.7% in the meta-analysis. Excluding this study completely removed the asymmetry (Luis Furuya-Kanamori index: 0.05) but not the heterogeneity (Q = 1484.5, P < 0.001, I2 = 98%). The study’s exclusion also resulted in a substantial reduction in the pooled estimate to 45.2% (95% CI: 28.9–62.0%) and a narrowing of the confidence intervals. When the actual sample figures from this study were used instead of the population-adjusted figures the resulting pooled estimate was 46.5% (95% CI: 31.2–62.2%), with major heterogeneity (Q = 1537.1, P < 0.001, I2 = 98%) but no asymmetry (Luis Furuya-Kanamori index: −0.57).

25Although we found 23 studies for this review, the high degree of heterogeneity in the studies made it difficult to form clear conclusions as to the true prevalence of asymptomatic Zika virus infection. Furthermore, subgroup analysis by population group was unable to explain the heterogeneity. While the prevalence of asymptomatic Zika virus infection appeared to be lower in returned travellers and adults with Guillain‒Barré syndrome, this could be due to the lack of representativeness of the samples, as those with symptoms are more likely to be tested.

26The large variation in prevalence of asymptomatic Zika virus infection in the general population, which ranged from 29% (95% CI: 24–24%) in schoolchildren from French Polynesia to 82% (95% CI: 81–83%) in the general population of Yap State could be due to several reasons. One possibility could be the lack of representativeness of the French Polynesia sample as the response rate was not reported. A second possibility is that the population prevalence in Yap State was overestimated due to the method of assessing symptom status, which was done retrospectively and then adjusted for the percentage unlikely to be attributable to Zika virus infection. The high degree of sensitivity of the results to the removal of this study lends supports to this possibility. A third possibility is that differences in definitions of symptoms and criteria for Zika virus infection (including the diagnostic test used) could have led to differences in prevalence estimates. This possibility is supported by the lower prevalence of asymptomatic Zika virus infection in pregnant women with confirmed recent infection than in those with possible recent infection (42% versus 63%; Table 3) in the United States. Finally, the difference could be real.

27The authors of a systematic review and meta-analysis of 55 influenza virus infection studies also found considerable heterogeneity in the proportion of asymptomatic infected persons. Despite the large number of studies, the heterogeneity could not be explained by the type of influenza, the laboratory tests used to detect the virus, the year of the study, or the location of the study. For Zika virus the amount and quality of the available evidence is insufficient to provide a single estimate of the prevalence of asymptomatic infection or to determine whether the heterogeneity found in this review is real.

28In relation to the heterogeneity in prevalence, comparing two included studies that presented data on completed pregnancies from the United States Zika pregnancy registry and used similar surveillance methods is important., One study in the USA found an asymptomatic Zika virus infection prevalence of 63%; this is consistent with an earlier report of 61% from the same population, suggesting little variation over time. The other study was of completed pregnancies in United States Territories (American Samoa, Puerto Rico and United States Virgin Islands) and the Federated States of Micronesia and Marshall Islands and found a prevalence of asymptomatic Zika virus infection of 38%. If the difference is real or a result of differences in ascertainment of asymptomatic Zika virus infection is difficult to know. The registry is based on surveillance systems, which depend on testing in clinical practice and which can be affected by the care-seeking behaviour of the population. This raises the issue of the ability of surveillance systems to provide unbiased results for Zika virus research questions.

29Although we included population subgroups in our meta-analysis there were insufficient data to study the effect of demographic variables on the prevalence of asymptomatic Zika virus. While three of the included studies reported on age, sex or geographical differences in symptomatic infection,– clear conclusions were not possible to make.

30A key strength of this review was the use of high-quality systematic review methods. Limitations of the review include the small number of studies found, especially cross-sectional seroprevalence studies, and the heterogeneity in the methods used across studies. The majority of studies included in the review were based on population health surveillance or screening programmes, rather than good-quality research studies. Furthermore, the included studies used various definitions of Zika virus positivity and rarely offered a definition for Zika virus symptom status. A variety of laboratory tests were used with varying degrees of validity, which can lead to potential misclassification error. A particular issue for Zika virus infection is the serological cross-reactivity of current IgM antibody assays with dengue virus, among other flaviviruses., The potential effect on the results is not known. In several studies there was also a bias towards inclusion of participants with symptoms due to the criteria for population surveillance or because symptomatic people are more likely to seek health care (e.g. travellers returning from Zika virus-endemic areas).

31One clear finding from this review is that, given the current state of the evidence, it is not possible to give an accurate figure for the prevalence of asymptomatic Zika virus. Nor is it known whether the prevalence varies between populations or over time. Better-quality research is needed to estimate prevalence in the general population and in specific population groups. The use of standardized protocols developed by WHO and partners, particularly the protocol for the cross-sectional seroprevalence study of Zika virus infection in the general population, will be important in this regard. The protocol aims to standardize the diagnostic tests and definitions used, as well as encouraging consistent reporting., Use of the protocol will ensure results can be compared across regions and countries and help to improve the quality of the studies by minimizing bias. In this way the results of studies will better inform future public health surveillance and interventions.