Repeated Rapid Active Sampling Surveys Demonstrated a Rapidly Changing Zika Seroprevalence among Children in a Rural Dengue-endemic Region in Southwest Guatemala during the Zika Epidemic (2015–2016)

ABSTRACT.

Although Central America is largely dengue virus (DENV)-endemic, the 2015–2016 Zika virus (ZIKV) pandemic brought new urgency to develop surveillance approaches capable of characterizing the rapidly changing disease burden in resource-limited settings. We conducted a pediatric DENV surveillance study in rural Guatemala, including serial cross-sectional surveys from April through September 2015 (Survey 1), in October–November 2015 (Survey 2), and January–February 2016 (Survey 3). Serum underwent DENV IgM MAC ELISA and polymerase chain reaction testing. Using banked specimens from Surveys 2 and 3, we expanded testing to include DENV 1–4 and ZIKV microneutralization (MN₅₀), DENV NS1 IgG ELISA, and ZIKV anti-NS1 antibody Blockage of Binding (BoB) ELISA testing. Demographic risk factors for ZIKV BoB positivity were explored using multivariable generalized linear regression models. Of Survey 2 and 3 samples available (N = 382), DENV seroprevalence slightly increased (+1%–10% depending on the assay) during the surveillance period and increased with age. In contrast, ZIKV seroprevalence consistently increased over the 3-month period, including from 6% to 34% (P < 0.0001) and 10%–37% (P < 0.0001) using the MN₅₀ ≥100 and BoB ELISA assays, respectively. Independent risk factors for ZIKV seropositivity included older age (prevalence ratio (PR)/year = 1.12, 95% confidence interval (CI) = 1.07–1.17) and primary caregiver literacy (PR = 2.80, CI = 1.30–6.06). Rapid active surveillance (RAS) surveys demonstrated a nearly 30% increase in ZIKV prevalence and a slight (≤ 10%) increase in DENV seroprevalence from October to November 2015 to January to February 2016 in rural southwest Guatemala, regardless of serologic assay used. RAS surveys may be a useful “off-the-shelf” tool to characterize arboviruses and other emerging pathogens rapidly in resource-limited settings.

INTRODUCTION

With an estimated 51 to 96 million cases of dengue fever and 105 to 390 million dengue infections annually, dengue virus (DENV) is a major global public health concern.^1–3 Zika virus (ZIKV), which shares significant sequence and structural homology with DENV and causes congenital Zika syndrome,^4–6 first arrived in the Americas in 2014 and rapidly spread throughout the region in 2015–2017.^7,8 Worldwide, approximately 3.9 billion people are at risk for DENV and ZIKV due to the presence of their shared vector, Aedes mosquitoes.¹ In Guatemala, DENV is widespread, with all four DENV serotypes causing outbreaks, seasonal peaks every rainy season (summer and fall), and recent larger outbreaks occurring in 2005, 2007, 2010, and 2013.⁹

Accurate, timely, and local burden of disease data for DENV and ZIKV are critical for allocating limited public health resources, including vaccines, to the areas of greatest impact. Recommendations by the WHO Strategic Advisory Group of Experts on Immunization to provide CYD-TDV only to DENV-seropositive individuals (DENV-naive vaccinated individuals are at increased risk for severe disease when infected) and in regions with high DENV seroprevalence underscore the importance of developing tools to estimate local DENV seroprevalence rapidly using limited resources.¹⁰ Likewise, accurate, timely ZIKV seroprevalence and risk factor data will be important for directing future public health policies to reduce risk of ZIKV to pregnant women, and for the testing, introduction, and targeting of ZIKV vaccination programs.^6,11

Currently, most low and middle-income countries (LMICs) rely almost exclusively on passive surveillance systems for disease burden estimates of DENV, ZIKV, and other emerging pathogens, especially at subnational levels. These existing, primarily hospital-based systems, which may or may not include diagnostic testing, generally underestimate community burden, including the numerous asymptomatic or mild (subclinical) infections,^1,6 and they are often unable to identify high-risk subpopulations. To address the needs of public health services and stakeholders for DENV and ZIKV vaccines, active community-based surveillance tools are needed that can provide accurate estimates of disease burden, identify risk factors, and be performed quickly and with limited resources.^9,12,13

Rapid cross-sectional seroprevalence surveys, paired with appropriate diagnostic testing, may serve as one such cost-effective tool. Recent reviews found that existing DENV and ZIKV seroprevalence studies demonstrate significant heterogeneity in their methodology and often introduce selection bias due to a lack of probabilistic sampling.^14,15

We examined the feasibility and performance of community-based, serial cross-sectional rapid active sampling (RAS) surveys as a standardized seroprevalence survey tool to provide timely, low-cost DENV and ZIKV disease burden estimates and identify risk factors in an at-risk pediatric population living in rural Guatemala. We describe seroconversion of DENV and ZIKV infection, as well as risk factors for ZIKV infection, among rural Guatemalan children early in the ZIKV epidemic.

MATERIALS AND METHODS

Study setting and population.

This study was conducted in 25 communities (estimated population 30,000) within a 200 km² catchment area along the coastal lowlands of southwest Guatemala. The populations living in these communities suffer from high levels of food insecurity and poverty; low access to healthcare; and high levels of diarrheal, respiratory, and arboviral disease.^16,17

Surveillance methodology.

All households were screened and enrolled using a “30 × 7” two-stage cluster sampling strategy adapted from the WHO Lot-Quality Assurance method.^18–23 Household clusters were sampled from a predetermined, satellite map–based grid of the catchment area without replacement. Thirty separate clusters of seven households (210 total households) were enrolled into each of three cross-sectional Surveys. The first Survey (Survey 1) was the enrollment visit for a prospective cohort for acute febrile illness (AFI), which occurred April through September 2015. The two RAS surveys (Survey 2 = October–November 2015 and Survey 3 = January–February 2016) were each conducted as single household visits in a separate randomized sample.^24,25 The RAS surveys were designed to collect representative but timely seroprevalence data using minimal resources. Thus, instead of identifying blocks/clusters of households a priori, the RAS surveys enrolled the first (index) house in a cluster, then followed a predetermined pattern to enroll subsequent households (Supplemental Figures 1 and 2). Within eligible households, children aged 6 weeks to 17 years were enrolled in the study after parental consent (and child assent if ≥ 7 years). Study nurses collected demographic, epidemiologic risk factor, and clinical data and a venous blood sample on all study participants. All interviews were done in person and in Spanish, and data were entered via a smartphone application (Integra IT, Bogota Colombia).

Laboratory testing.

Whole blood samples were collected at the household during the visit. Serum was separated within 24 hours and frozen at –20°C for up to 4 weeks, then at –80°C thereafter. Serum from children with self-reported AFI were tested for DENV by reverse transcriptase polymerase chain reaction (RT-PCR), using the CDC DENV-1–4 assay.²⁶ Serum samples from all study children underwent DENV IgM Capture ELISA (InBios Inc., Seattle, WA), as previously described.²⁷

In 2018, all remaining samples from Surveys 2 and 3 underwent additional testing for DENV 1–4 and ZIKV microneutralization to an inhibition of 50% (MN₅₀), DENV nonstructural protein 1 (NS1) IgG ELISA, and ZIKV anti-NS1 antibody Blockage of Binding (BoB) ELISA assay, as previously described.²⁸

Case definitions.

AFI was defined as self-reported fever for ≥ 2 days. Three systems were used to define DENV and ZIKV infection status based on MN₅₀ titer: 1) a strict cutoff, in which a MN₅₀ titer > 100 was considered “positive”; 2) a relative cutoff, in which the both DENV and ZIKV titer were taken into account, when categorizing an individual as “negative,” “possible,” “positive,” or “indeterminate” (Figure 1); and 3) the ZIKV-to-DENV ratio, in which individuals were considered “positive” for ZIKV if their ZIKV MN₅₀ was greater than their maximum DENV-1–4 MN₅₀ titer.²⁹

Figure 1.

Zika and Dengue serostatus classification system based on relative Zika virus (ZIKV) and dengue virus (DENV) microneutralization (MN₅₀) titer. The serologic categorization takes into account both DENV and ZIKV MN50 titer to classify an individual’s DENV/ZIKV exposure status as “negative,” “possible,” “positive,” or “indeterminate.”

Statistical analysis.

We first compared demographic variables between each Survey using χ² tests for categorical variables, and Student’s t test for two-way comparisons of normally distributed continuous variables. We next calculated overall seroprevalence of DENV and ZIKV using the DENV/ZIKV MN₅₀ approaches outlined in Figure 1, as well as DENV IgM and NS1 IgG ELISA (cutoff ≥ 50), ZIKV anti-NS1 antibody BoB ELISA (cutoff > 10), and ZIKV-to-DENV MN₅₀ ratio > 1. Significant differences in seroprevalence between surveys were calculated using general linear models, assuming equal variance across the three groups, for the Dengue IgM MAC ELISA test results, and χ² tests for the remaining test results, which were only available for Surveys 2 and 3 (see Table 1). Seroprevalence was compared between Surveys 2 and 3 by age group and gender for each testing modality using χ² to calculate statistical significance. Surveys 2 and 3 had more extensive diagnostic testing due to availability of samples. In the ZIKV risk factor analysis, univariate associations between each potential risk factor (independent variables) and ZIKV positivity (dependent variable) were calculated. A multivariable generalized linear regression model using a purposeful variable selection strategy was then used to identify significant independent risk factors for ZIKV seropositivity. SAS version 9.4 (Cary, NC) was used for all statistical analysis.

Table 1.

Dengue virus (DENV) and Zika virus (ZIKV) seroprevalence by assay and interpretation in rural Guatemalan children 2015–2016

	Survey 1 (April–Sept 2015)	Survey 2 (Oct–Nov 2015)	Survey 3 (Jan–Feb 2016)
Testing method	n/total (%)	n/total (%)	n/total (%)	P value*
Dengue
PCR (AFI only)	0/75 (0)	0/84 (0)	0/54 (0)	–
IgM MAC ELISA	5/343 (1)	11/229 (5)	25/188 (13)	< 0.0001
MN50 ≥ 100	–	105/196 (54)	119/186 (64)	0.04
MN50 “Positive”	–	86/196 (44)	98/186 (53)	0.14
MN50 “Positive” or “Possible”	–	130/196 (66)	138/186 (74)	0.14
NS1 IgG ELISA >50	–	128/146 (88)	133/153 (87)	0.85
Zika
MN50 ≥ 100	–	11/196 (6)	63/186 (34)	< 0.0001
MN50 “Positive”	–	1/196 (1)	54/186 (29)	< 0.0001
MN50 “Positive” or “Possible”	–	28/196 (14)	74/186 (40)	< 0.0001
anti-NS1 antibody Blockage of Binding ELISA > 10	–	20/196 (10)	69/186 (37)	< 0.0001
ZIKV:DENV MN50 ratio > 1	–	0/196 (0)	36/186 (19)	< 0.0001

AFI = acute febrile illness; BoB = Blockage of Binding; PCR = polymerase chain reaction.

* P values for categorical variables calculated using χ² test, and P values for continuous variable calculated using general linear models, assuming equal variance across the three groups.

Ethical oversight.

The study was approved by the Colorado Multiple Institutional Review Board, the Universidad del Valle de Guatemala Institutional Review Board, and the Guatemala Ministry of Health and Social Welfare National Ethics Committee. The local Southwest Trifinio Community Advisory Board for Research reviewed and agreed to the study.

RESULTS

RAS survey screening and enrollment.

Survey 1 enrolled 469 children from 207 households; Survey 2, 402 children from 210 households; and Survey 3, 368 children from 210 households, for a total of 1,239 children from 627 households (Figure 2). The most common reasons for declining study participation (N = 164) included lack of perceived benefit to the child (42.1%) and discomfort with specimen collection (28.7%). Detailed subject and household characteristics are reported separately.²⁵

Figure 2.

CONSORT Diagram of Study Recruitment, Enrollment, and Completion. Survey 1 was the enrollment visit of a prospective cohort and enrolled from April to September 2015. Surveys 2 and 3 were cross-sectional rapid active sampling (RAS) surveys conducted on separate randomized households during a single visit from October–November 2015 (Survey 2) and January–February 2016 (Survey 3). Initial testing consisted of dengue (DENV) polymerase chain reaction (PCR) on all cases with acute febrile illness (AFI) and DENV IgM ELISA on all individuals. Additional DENV and ZIKV testing was performed on remaining serum samples for Surveys 2 and 3, as described.

Surveys 2 and 3 coincided with the onset of the ZIKV epidemic in Guatemala and Central America, with the first case reported by the Guatemala Ministry of Health and Social Welfare reported in November 2015 (Figure 3).

Figure 3.

Timing of the three cross-sectional dengue virus (DENV)/Zika virus (ZIKV) seroprevalence surveys (1–3) and Onset of the Zika Epidemic in Guatemala, 2015–2017. The surveys were conducted in April–September 2015 (Survey 1), October–November 2015 (Survey 2), and January–February 2015 (Survey 3). Black bars represent suspected and confirmed ZIKV cases reported to the Pan American Health Organization for Central America/Guatemala (https://www.paho.org/hq/dmdocuments/2017/2017-phe-zika-situation-report-gut.pdf). The striped arrow represents the first polymerase chain reaction–confirmed ZIKV case detected at the Trifinio study site in May 2015.

DENV and ZIKV testing results.

There were 1,239 children enrolled in the three surveys. Table 1 and Figure 4 show DENV and ZIKV seroprevalence estimates during each of the three surveys, stratified by testing modality. Of those with serum available, 213 (17%) reported AFI in the preceding week and none were DENV RT-PCR-positive. AFI samples were not available for ZIKV RT-PCR.

Figure 4.

Dengue and Zika seroprevalence in Surveys 2 and 3 in rural Guatemalan Children 2015–2016. Striped bars = Survey 2: October–November 2015; Black bars = Survey 3: January–February 2016.

Of the 1,239 children, 760 (61%) had serum available for DENV IgM. Children with serum available were more likely to be older (9.9 years versus 7.7 years, P = 0.002). In Surveys 2 and 3, which included the expanded testing, 382 children (196 [86%] from Survey 2 and 186 children [99%] from Survey 3) had serum remaining for MN₅₀ and anti-NS1 antibody BoB testing, and 299 (146 [64%] from Survey 2 and 153 children [81%] from Survey 3) had serum available for DENV NS1 IgG. The groups were not significantly different in terms of demographics: the 196 children from Survey 2 were 45% male, with a mean (SD) age of 9.6 (4.3) years, whereas the 186 children from Survey 3 were 43% male with a mean (SD) age of 10.2 (4.0) years.

DENV IgM positivity increased from 1% to 5% to 13% during the three survey periods (P < 0.0001). In Surveys 2 and 3, with the expanded testing, DENV seroprevalence trended upward over the 3-month period and increased significantly, relying on the MN₅₀ ≥ 100 definition (54%–64%, P = 0.04); prevalence varied by test and interpretation. The DENV NS1 IgG ELISA provided a significantly higher overall seroprevalence estimate (87%) than all the MN₅₀ classifications, including MN₅₀ ≥ 100 (75%, P < 0.0001).

In contrast to DENV, ZIKV seroprevalence increased significantly over the 3-month period by all testing modalities (P < 0.0001) (Table 1; Figure 4). The highest estimates for Surveys 2 and 3 were obtained using the anti-NS1 antibody BoB ELISA (10% and 37%, P < 0.0001) and the MN₅₀ positive and possible classification (14% and 40%, P < 0.0001). The ZIKV-to-DENV MN₅₀ titer > 1 provided a lower estimate of ZIKV positivity (0% and 19%, P < 0.0001). Seropositivity for both DENV and ZIKV increased with age (Figure 5), but there was no significant difference based on sex.

Figure 5.

Dengue virus (DENV) and Zika virus (ZIKV) age-based seroprevalence by assay and interpretation in Surveys 2 and 3 in rural Guatemalan Children 2015–2016. Age groups of 2 to <5 years, 5 to <9 years, and 9 to 18 years demonstrate increasing DENV and ZIKV seroprevalence by age, and prevalence varying by assay used. Striped bars = Survey 2: October–November 2015; Black bars = Survey 3: January–February 2016.

In the ZIKV risk factor analysis, ZIKV seropositivity by anti-NS1 antibody ELISA was independently associated with older age (prevalence ratio = 1.12 per year of age, 95% confidence interval: 1.07–1.17) and primary caregiver literacy (prevalence ratio = 2.80, 95% confidence interval: 1.30–6.06; Table 2). ZIKV MN₅₀ > 100 was associated with the same significant risk factors.

Table 2.

Univariate and multivariable associations of potential risk factors with Zika virus (ZIKV) seropositivity using the anti-NS1 antibody Blockage of Binding (BoB) assay in rural Guatemalan children, 2015–2016

Univariate Analysis
Variable	Mean (SD)	Prevalence ratio	95% confidence interval
Age	9.9 (4.2)	1.11	(1.07–1.17)*
No. of people in house (crowding)	6.2 (2.3)	0.96	(0.88–1.05)
Children age ≤ 5 years in house†	0.9 (0.9)	0.82	(0.66–1.02)
	n (%)	PR	95% CI
Gender = female	214 (55.9%)	0.97	(0.67–1.39)
Ethnicity = Latino	374 (98.4%)	NE	NE
Caregiver is literate	318 (83.7%)	2.66	(1.22–5.83)*
School attendance (age 6+)	233 (77.4%)	0.92	(0.59–1.43)
House material = wood	81 (21.4%)	0.65	(0.38–1.10)
Water source = well	291 (76.8%)	0.91	(0.14–5.77)
Standing water on property	117 (30.6%)	0.84	(0.55–1.27)
Multivariable analysis
Significant risk factor		PR	95% CI
Age (years)		1.12	1.07–1.17*
Caregiver is literate		2.80	1.30–6.06*

CI = confidence interval; NE = not estimable; PR = prevalence ratio.

* P value < 0.05.

^†Not including study subject.

DISCUSSION

Over a 3-month period from October–November 2015 to January–February 2016, the RAS surveys demonstrated a constant to slightly increasing DENV seroprevalence among children and a rapidly increasing ZIKV seroprevalence of nearly 30%. The surveys provided fairly constant seroprevalence estimates regardless of the assay used. The probabilistic, two-stage clustering approach to the surveys allowed rapid, efficient estimates of population- and age-based DENV and ZIKV seroprevalence in a resource-limited, rural region of Guatemala. RAS surveys could be a useful “off-the-shelf” tool to track changing burdens of flaviviruses and other emerging pathogens over time.

The force of ZIKV infection identified in this study population from rural, southwest Guatemala resembles that from other settings in Latin America,^30–32 including in Managua, Nicaragua, which estimated a similar (36%) seroprevalence increase over a 3-month period (July–September 2016).³² The timing of our seroprevalence increase is consistent with phylogenetic analyses, which support multiple ZIKV introductions into Central America beginning in 2014, with the dominant strain entering Guatemala, southern Mexico, and Nicaragua by early 2015.⁷ Indeed, PCR-confirmed ZIKV cases from southern Mexico and our Guatemala site, which borders Chiapas, support low-level circulation in the region, beginning in early 2015, with a first epidemic peak in late 2015.^33–35 This low-level ZIKV circulation before detection through traditional surveillance systems has been reported elsewhere.^36–38

Seroprevalence estimates for both ZIKV and DENV varied significantly depending on the assay and cutoffs, highlighting the importance of establishing consensus and careful interpretation of seroprevalence studies. In our setting, ZIKV seroprevalence ranged from 1%–14% (Survey 2) to 29% to 40% (Survey 3), with MN₅₀ ≥ 100 and anti-NS1 antibody ELISA, respectively, showing similar estimates. The ZIKV-to-DENV ratio, which has demonstrated some utility (and specificity) in identifying acute ZIKV infections following prior DENV infections, supported this increase as well.²⁹ In contrast, DENV seroprevalence varied broadly by assay, from 54% to 64% (MN₅₀ ≥ 100) to 87% to 88% (NS1 IgG ELISA), and we likely observed some cross-reactivity with ZIKV given that there was no evidence of ongoing DENV transmission in the community or on acute illness testing.³⁹ Despite the limitations of the individual assays used, an important observation is that the assays all showed internal consistency between the surveys, demonstrating a rapidly changing ZIKV seroprevalence (+26%–28%) and slightly increased DENV seroprevalence (+1%–10%).

Our ZIKV risk factor analysis demonstrated greater risk among older children, which may be due to increased risk of exposure to Aedes mosquitoes, as has been reported elsewhere,^40,41 or due to increased exposure to DENV and cross-reactivity.³⁹ Caregiver literacy was also independently associated with ZIKV seropositivity, which may be a proxy for living in a more urbanized setting and an associated increased risk of Aedes exposure.⁴²

Overall, the RAS survey methodology, which builds on the WHO “30 × 7” approach,^18–23 provided a cost-effective method to characterize a rapidly changing ZIKV seroprevalence in a highly DENV-endemic region. DENV, ZIKV, and other arboviruses are known to demonstrate significant and rapidly changing disease burdens, including on a subnational level, and policymakers require accurate estimates to make informed decisions about allocating limited public health resources, including vaccines. Cross-sectional seroprevalence surveys remain an important and accurate tool to estimate regional infection and disease burden. Our approach, which used satellite-based two-stage cluster randomization and predetermined sampling patterns by a study team of four trained nurses, provided an efficient and consistent method to obtaining community disease burden data over time, including age-based seroprevalence and risk factors for seropositivity.

The RAS approach may also be useful as an “off-the-shelf” tool for rapidly characterizing other emerging pathogens. The CDC CASPER (Community Assessment for Public Health Emergency Response) methodology,^43,44 which has been deployed in SARS-CoV-2 seroprevalence studies across the United States, is similar to the RAS survey methodology, with some key differences that may make the RAS methodology advantageous in certain low-resource settings. The CASPER design relies on traveling throughout the entire catchment area and selecting every ‘nth’ household for the survey, which may be less practical in large, geographically challenging, rural regions with limited resources. In such settings, the RAS approach, which randomly samples quadrants within the catchment area, assigns an index household, and then selects the next six households to the right of that index household, may provide a good approximation of prevalence while requiring less time and fewer resources. The interval between surveys could depend on whether they were being used for routine surveillance (annual) or to characterize an emerging pathogen rapidly (3 months); biobanked samples could also be rapidly disseminated to address urgent research needs (i.e., diagnostics), as was the case in our study.^35,38,45,46

Strengths of this study include the two-stage sampling design, which ensured representativeness of the study population to the region, and the use of multiple testing platforms to determine DENV and ZIKV prevalence for each survey, although all assays have the potential for cross-reactivity. Future studies may benefit from the use of standardized reference samples to calibrate variation in the assays used. Including ZIKV (or any additional) PCR testing during acute illness would provide greater insight into the burden of disease and alleviate concerns about flavivirus cross-reactivity, although this would need to be balanced with the additional cost. The lack of any positive DENV PCR testing during AFIs does suggest a low community burden at the time of the surveys. Other limitations of the risk-factor analysis include potential selection bias because literate parents may have been more aware of the risk of arboviral disease and thus more likely to enroll in the study, and social desirability bias may have influenced the predictor variables, which were collected via self and proxy report. However, such biased reporting of risk factors is likely to be nondifferential because the outcome (ZIKV status of the child) was not known or even being specifically investigated at the time of the predictor variable data collection. Thus, our results would be biased toward the null. Sampling bias was also possible, but this was minimized by the two-stage cluster design, and the surveys were demographically similar (data not shown). Finally, screening of mosquito breeding sites would provide greater understanding of exposure.

In conclusion, the RAS survey approach provided an efficient method to obtain community and age-based seroprevalence data for DENV and ZIKV, demonstrating a high ZIKV force of infection between October–November 2015 and January–February 2016. These findings were consistent regardless of the serologic assay used. Our results support existing phylogenetic studies of the early, undetected transmission of ZIKV throughout Central America and highlight the importance of developing and implementing serologic assays and consensus guidelines, including those that could be performed on site. RAS surveys offer a valuable tool to track changing seroprevalence rapidly and identify risk factors for novel and emerging infectious diseases and also to identify populations that would benefit most from public health interventions, such as vaccines.

Note: Supplemental materials appear at www.ajtmh.org.