Study Setting

We conducted this study in 4 communities already participating in a long-term entomologic trial in the Salinas and Guayama municipalities on Puerto Rico’s southeastern coast. In 2015, the population of Salinas was 30,114 and the population of Guayama was 43,700; both had population densities of 434–672 persons per square mile (13). Both municipalities have young (median age ≈36 years) populations, ≈15% of whom are >65 years of age (13). The municipalities also have near equal sex distribution, and >50% of households are below the poverty line (13).

Community-level data showed that the 4 study sites were small, semiurban neighborhoods with comparable population sizes, household densities, and occupancy rates (Appendix Table 1). Average household size was 2.6–3.6 persons. Most homes were single-story with patios or gardens, and architecture and climate were similar across sites. All communities had piped water and waste removal services, including sewer or septic coverage depending on location (9,13,20). Those indicators support baseline demographic and infrastructural comparability of intervention and nonintervention sites.

The 2 intervention communities (La Margarita, Villodas) are geographically buffered by vegetation or roads (200–500 meters), reducing mosquito movement from adjacent areas (20). The nonintervention communities (Arboleda, La Playa) without AGOs are embedded in larger urban zones. Although not randomized, we selected sites that were demographically and environmentally comparable on the basis of census and field data (13,21). All 4 sites have had continuous mosquito surveillance since 2012. Prior analyses confirmed that differences in Ae. aegypti mosquito abundance between intervention and nonintervention areas emerged only after trap deployment (14,22). Consistent with prior evaluations, we observed similar female Ae. aegypti mosquito abundance during the predeployment baseline period, October–December 2011, in the original paired communities of La Margarita and Villodas (Appendix Figure 1). A CHIKV serosurvey in those sites found no systematic demographic or household-level differences (13). Other community-based serosurveys in Puerto Rico have shown consistent ZIKV, DENV, and CHIKV seroprevalence patterns, and variation was driven more by age and household factors than geography (23,24). Together, those findings confirmed sustained reductions in Ae. aegypti mosquitoes in intervention areas and limited underlying differences, supporting the inference that observed ZIKV effects were unlikely to reflect underlying community characteristics (13,20,25).

AGO Intervention

AGOs attract and capture gravid female Ae. aegypti mosquitoes (9,10,20). Each household in intervention communities received 3 traps, maintained every 2 months by trained staff. Deployment began during 2011–2013 and ultimately covered ≈85% of households. Previous studies documented that intervention communities experienced marked and sustained reductions in Ae. aegypti mosquito abundance compared with nonintervention sites (21). Weekly entomologic surveillance using sentinel AGOs provided data on Ae. aegypti mosquito abundance, which we linked to participant infection status (9,20,26) (Appendix).

Study Design and Sampling

We conducted a cross-sectional, community-based serosurvey during March–May 2017 to assess ZIKV infection among residents in the 4 communities. We conducted the serosurvey 6–9 months after peak ZIKV transmission in Puerto Rico during August 2016 to capture infections from the outbreak period. The survey overlapped with the seasonal arbovirus activity trough (March–April), when incident infections are uncommon (17). Therefore, IgM detection in this survey reflects infections acquired during the 2016 epidemic rather than new infections occurring at the time of sampling. Although the serosurvey was conducted in 2017, analysis and reporting were delayed because of competing public health response priorities and the time required for data harmonization, quality assurance, and linkage to longitudinal entomologic surveillance.

For sampling, we used a stratified random design, assigning a unique identifier to each residential structure and randomly selecting households to achieve ≈28.5% coverage of the total population. We chose 28.5% coverage to balance statistical power with operational feasibility for household-based venous blood collection. The 28%–30% coverage target was successfully applied in prior arboviral serosurveys in the same study communities and yielded representative samples for demographic and household characteristics (13,25). Field teams visited each selected household up to 3 times to recruit participants. If a household was vacant or the head-of-household remained unavailable after 3 visits, we randomly replaced that household to meet enrollment targets.

Eligible participants included all residents >5 years of age who slept in the selected household for >4 of the previous 7 nights. We excluded children <5 years of age because of the difficulty of venous blood collection and ethical considerations of venipuncture in that age group. Although younger children can provide valuable information on recent arbovirus circulation in endemic settings, ZIKV was newly introduced in Puerto Rico in 2015–2016; thus, all age groups were susceptible, and infection risk was broadly distributed. Our primary aim was to assess community-level ZIKV infection prevalence across the general population after the epidemic, which we could accomplish by including participants >5 years of age. Participant selection was independent of household AGO presence; in intervention communities, households were neither included nor excluded based on whether AGOs were installed at that specific residence.

We obtained written informed consent from all adult participants. Persons 15–20 years of age provided written assent with parental or guardian permission, and children 5–14 years of age provided verbal assent with written parental or guardian permission. All participants provided blood specimens, regardless of reported symptoms, and completed structured questionnaires on demographic and housing characteristics, mosquito prevention practices, and recent illness history (Appendix). This study was reviewed and approved by the CDC Institutional Review Board (protocol no. 6800).

Laboratory Testing

Field teams collected venous blood specimens and transported specimens on the same day to the CDC Dengue Branch (Division of Vector-Borne Diseases, National Center for Emerging and Zoonotic Infectious Diseases) in San Juan, Puerto Rico. Upon arrival, we centrifuged samples to separate serum, then aliquoted and stored serum at –20°C until testing.

We tested serum specimens for ZIKV IgM by using the CDC Zika IgM antibody capture ELISA (Zika MAC-ELISA), following CDC instructions, and tested for DENV IgM by using the DENV Detect IgM capture ELISA (InBios International, Inc., https://inbios.com), following manufacturer instructions. The Zika MAC-ELISA has high sensitivity and specificity for recent ZIKV infection in dengue-endemic settings, although some cross-reactivity with other flaviviruses, particularly DENV, can occur (27,28). To minimize misclassification, our primary outcome defined ZIKV infection as ZIKV IgM–positive and DENV IgM–negative results, an approach supported by evaluations of the Zika MAC-ELISA showing that IgM reactivity is strongest for the homotypic virus in dengue-endemic settings (27). We also conducted a sensitivity analysis by including participants testing positive for ZIKV IgM, DENV IgM, or both. We did not test for DENV or ZIKV IgG because available assays show substantial cross-reactivity among ZIKV-exposed persons in dengue-endemic settings, limiting the usefulness of those assays for distinguishing prior ZIKV from prior DENV infection (29).

Statistical Analysis

We estimated prevalence ratios (PRs) for ZIKV infection by using Poisson regression with robust SEs, adjusting for age, sex, and time spent at home. We explored effect modification across demographic and household subgroups. Sensitivity analyses included broader arbovirus IgM outcomes and models incorporating entomologic data. We conducted all analyses in R version 4.4.2 (The R Project for Statistical Computing, https://www.r-project.org) (Appendix).

Study Population

A total of 330 participants from 242 households completed the serosurvey: 71 households from Arboleda (nonintervention), 79 from La Margarita (intervention), 40 from La Playa (nonintervention), and 52 from Villodas (intervention). We selected households from among 1,228 total residential structures, of which 1,014 (82.6%) were occupied during enumeration (Appendix Figure 2). We prioritized household sampling by structures that participated in a prior 2015–2016 CHIKV serosurvey, then randomly selected replacement households to meet enrollment targets.

Of the 330 enrolled participants, we excluded 55 because of indeterminate serology results: 32 hemolyzed samples, 17 equivocal results, and 6 nonspecific results. We excluded another 4 participants who tested DENV IgM–positive. Thus, we included a total of 271 (82.1%) participants from 208 households: 65 households in Arboleda, 65 in La Margarita, 33 in La Playa, and 45 in Villodas. Among included participants, 136 (50.2%) lived in intervention communities and 135 (49.8%) lived in nonintervention communities. Participants’ median age was 59 (IQR 46–69) years; 165 (60.9%) were female and 106 (39.1%) were male (Table 1; Appendix Table 2). Most had lived in their communities for >10 years, and household characteristics were similar between groups.

The secondary sensitivity analysis included an expanded sample of 297 participants with valid ZIKV or DENV IgM results, including those who were DENV IgM–positive. That population had similar demographic and household characteristics to the primary analytic sample (Appendix Table 3).

Baseline Characteristics

Characteristics among participants in intervention and nonintervention communities did not differ substantially, including for age, sex, number of household residents, and duration of residence (Table 1; Appendix Table 2). However, participants in nonintervention communities reported higher levels of mosquito exposure than participants in intervention communities, including more frequent daily bites (24.6% vs. 8.1%; p = 0.004) and being bitten at home (79.3% vs. 66.9%; p = 0.031). Citronella use was more common in nonintervention areas (31.9% vs. 16.2%; p = 0.004), suggesting greater perceived or actual mosquito abundance or differing perceptions of citronella’s effectiveness relative to other control methods. Other prevention behaviors, such as repellent or coil use, were similar across groups. We observed similar patterns in the sensitivity analysis (Appendix Table 3).

ZIKV Seroprevalence

Figure 1

Figure 1. Adjusted prevalence ratios for ZIKV seropositivity by demographic and behavioral characteristic among participants in an evaluation of effectiveness of autocidal gravid ovitraps for preventing Zika virus infection, Puerto Rico, USA....

Overall, we detected recent ZIKV infection in 40 of the 271 participants, corresponding to 14.8% seroprevalence. Seroprevalence was much lower (9.6%, 13/136) in intervention communities than nonintervention communities (20.0%, 27/135). Overall crude PR was 0.48 (95% CI 0.26–0.89); after adjusting for confounders, the adjusted PR (aPR) was 0.49 (95% CI 0.27–0.90) (Figure 1; Appendix Table 4).

Lower ZIKV seroprevalence in intervention communities was consistent across most demographic and behavioral subgroups (Figure 1; Appendix Table 4). Among participants >60 years of age, seroprevalence was 5.1% in intervention and 23.3% in nonintervention communities (aPR 0.22 [95% CI 0.06–0.75]). We observed similar protective associations among participants in larger households (>4 residents; aPR 0.10 [95% CI 0.01–0.81]), without air conditioning (aPR 0.28 [95% CI 0.10–0.78]), and those who used mosquito coils (aPR 0.17 [95% CI 0.04–0.78]), although subgroup sizes were small.

Among participants who always kept windows or doors open, those in intervention communities had lower seroprevalence than those in nonintervention communities (6.5% vs. 24.0%; aPR 0.27 [95% CI 0.08–0.97]). Similarly, those spending >61 hours/week at home in intervention areas had lower seroprevalence compared with those in nonintervention areas (5.1% vs. 22.0%). In intervention communities, predicted infection probability rose with time at home, peaking at 52 hours per week before declining (Appendix Figure 3). In contrast, infection probability in nonintervention communities increased steadily, peaking at 84 hours.

Among the 297 participants in the expanded sensitivity analysis, 41 (13.8%) tested ZIKV IgM–positive and 5 (1.7%) tested DENV IgM–positive (Appendix Tables 3, 4). Overall arbovirus seroprevalence remained much lower in intervention communities than in nonintervention communities (9.2% vs. 21.5%; aPR 0.44 [95% CI 0.24–0.78]) (Appendix Table 5). The strongest protective associations persisted among older adults (aPR 0.21 [95% CI 0.07–0.62]), participants without air conditioning (aPR 0.25 [95% CI 0.09–0.69]), and those in larger households (aPR 0.10 [95% CI 0.01–0.80]), supporting the robustness of the primary findings.

Perceptions of AGO Effectiveness

Among participants in intervention communities, 105/136 (77.2%) reported a reduction in household mosquito density related to AGOs. Few reported an increase (3.7%, n = 5) or no change (8.1%, n = 11) in mosquito density, and 11.0% (n = 15) were unsure or did not respond to that question.

ZIKV Seropositivity and Acute Febrile Illness

Among ZIKV-seropositive participants, 37.5% (15/40) reported experiencing an acute febrile illness since November 2015, compared with 18.7% (43/231) of ZIKV-seronegative participants (p = 0.014) (Table 2). Seropositive participants more frequently reported common Zika symptoms than seronegative participants, including rash (27.5% vs. 8.7%), fever (30.0% vs. 15.2%), and joint pain (32.5% vs. 16.5%) (p≤0.040). However, care-seeking (20.0% seropositive vs. 12.1% seronegative; p = 0.270) and hospitalization (2.5% seropositive vs. 0.4% seronegative; p = 0.682) were infrequent and did not differ significantly by serostatus.

Entomologic Trends and ZIKV Seropositivity Association

Figure 2

Figure 2. Weekly Aedes aegyptimosquito abundance from an evaluation of effectiveness of autocidal gravid ovitraps (AGOs) for preventing Zika virus infection, Puerto Rico, USA. Mean mosquito counts per surveillance trap...

During the ZIKV epidemic, January 2016–May 2017, mean weekly Ae. aegypti mosquito abundance per surveillance trap was substantially lower in intervention than nonintervention communities (1.40 [95% CI 1.28–1.52] vs. 9.98 [95% CI 8.98–11.00] mosquitoes per trap) (Figure 2). In Poisson regression models adjusted for age category, sex, and hours spent at home, higher mosquito abundance was positively associated with ZIKV seropositivity. Each additional female mosquito captured per trap-week was associated with a 4% increase in ZIKV seropositivity risk at a 2-week lag (risk ratio [RR] 1.044 [95% CI 1.011–1.077]; p = 0.008) (Table 3). Associations at shorter lag times were similar in magnitude and reached statistical significance at a 1-week lag (RR 1.028 [95% CI 1.003–1.053]; p = 0.026), but not at 0 lag. Formal interaction tests provided no evidence that associations differed by intervention status (Appendix Table 6).