The IVM intervention took place in Caguas City, Caguas Municipality (18.23412 N, −66.0485 W). The population and number of buildings in the municipality were 142,893 (United States Census Bureau 2010) and 65,852 (Municipal Revenue Collection Center 1998), respectively. The treated area (23.1 Km²) included 61,511 inhabitants and 25,391 buildings. We developed a Geographical Information System (GIS) to monitor progress and produce maps with layers of roads (Puerto Rico Highways and Transportation Authority), buildings (Municipal Revenue Collection Center), satellite images (DigitalGlobe^™; June 25, 2014), location of mosquito surveillance traps and mosquito abundance, houses treated or untreated (abandoned, no response, refused treatment), and type of treatment. The city was divided into eight clusters with similar numbers of buildings and the order of IVM treatment was selected at random (Figs. 1, 2; Table 1). A flowchart summarizes the study (Fig. 2).

The proposal was presented, discussed, and approved by municipal and state health authorities, and a written agreement stating the scope and length of the intervention was signed by all parties. The Centers for Disease Control and Prevention determined that this intervention did not involve human subjects under 45 CFR 46.102(f). We explained to adult residents participating in the intervention the nature of the project and asked for their written permission to conduct IVM on their properties, return every two months to service the AGO traps, and in some of them return every week to count the number of mosquitoes in surveillance traps. Residents could opt out at any time.

We conducted six focus groups with 50 community leaders representing residents of 37 communities in the City of Caguas to explain the project and learn about residents’ acceptance of the use of mosquito traps in their communities, including installing three mosquito traps per home (CDC 2017). We further reached out to the residents in the city by conducting town hall meetings involving 154 community leaders from 150 communities to discuss how to protect against Zika infections and our projected application of IVM methods, including education, container removal, larvicides, and mass trapping with AGO traps. We established an agreement with the Government of Puerto Rico’s Call Center (311) to provide information to concerned residents and to orient them who to contact for further information.

We began monitoring adult female Ae. aegypti mosquitoes in October 2016; six weeks before the initiation of vector control. We deployed 360 fixed-position surveillance AGO traps (SAGO traps) that were monitored every week throughout the study (Fig. 1). AGO traps attract and capture gravid Ae. aegypti females on a sticky surface located inside a 3.8 liters, black plastic capture chamber that is partially inserted into a 19 liters black plastic bucket containing 10 liters of water and a 30 grams hay packet (Barrera et al. 2014b). Mosquitoes were counted in these surveillance traps by a team of eight technicians every week under CDC’s supervision. We did not monitor mosquitoes in the AGO traps that were used for intervention purposes (IAGO traps. 3 per structure). When a house had one SAGO trap then only two IAGO traps were placed in that house. SAGO traps were serviced every two months to replace water, sticky surface, and hay. SAGO traps (38–53 per cluster) were deployed to uniformly cover the eight clusters, avoiding areas that were 100 m from the edge of neighborhoods, highways, fields, industrial complexes, and streams. Distances between SAGO traps were 109 – 130m. Each SAGO trap’s geographical coordinate was used as a centroid to produce one Tiessen polygon per trap covering the entire intervention area, so that analyses could be performed at both cluster and polygon (SAGO trap) levels. A Thiessen polygon is drawn so that each SAGO trap is at the center of each polygon and the borders of polygons define areas that are the closest to each SAGO trap’s center considering all neighboring SAGO traps. Each of the 360 Tiessen polygons was used to aggregate data at that level (e.g., number of IVM-treated and untreated structures around a given SAGO trap).

Females of Ae. aegypti collected from every SAGO trap every week were transported to the laboratory where they were pooled by trap (1– 20 specimens per pool) and stored at −80ºC until processing by a trioplex RT-PCR to detect RNA of DENV, CHIKV, and ZIKV (Santiago et al. 2018). RNA of these viruses can be detected in Ae. aegypti mosquitoes exposed in the field for more than one week (reviewed in Barrera et al. 2018b).

Vector control activities were contracted to a pest control company (Rentokil/Oliver Exterminators). Vector control started on November 14, 2016 and the project ended on August 4, 2017, when contractors began removing control traps. Trap removal and subsequent follow up was interrupted by Hurricane Irma on September 6, 2017 and then by Hurricane Maria on September 20, 2017. IVM consisted of distributing educational handouts to adult residents, source reduction (removing only small disposable containers), applying larvicides (Four Star, Bacillus thuringiensis israelensis CRG) to containers with water not used for pet or human consumption, and setting three IAGO traps or more per building, depending on the size of the property (residential house, commercial store, etc.). Traps were acquired from SpringStar Inc. (Woodinville, WA) at $14.50 each.

We distributed the following handouts to households in the study area: notification that residents were to be visited by authorized and identified personnel from a specific pest control company to conduct IVM in their properties, illustrated explanation about the use of IAGO traps and actions residents can take to eliminate Ae. aegypti mosquitoes in their premises, and notifications that residents were to be visited to service the IAGO traps and repeat IVM. If mosquito larvae were detected in containers when visiting properties every two – three months to service the IAGO traps, the containers were either removed (small ones), treated with larvicides, or the resident was advised to cover or protect the container. In large public areas such as shopping centers, industries, and parks IAGO traps were placed at least 50m apart from each other. The initial goal of the intervention was to treat 80% or more buildings in each of the eight clusters because such a treatment coverage achieved significant and sustained control of Ae. aegypti in previous studies (Barrera et al. 2014a, b). We expected to gain insight at this scale about the minimum coverage required to bring down the mosquito population below two-three females per trap per week.

Hired personnel were instructed on the transmission of viruses by Ae. aegypti, its biology and ecology, how to monitor and control this mosquito, filling data forms, using maps to orient themselves in the field, and use of portable electronic devices to enter data (Table S1). Additional training sessions were provided given the high turnover of personnel throughout the project. The number of field personnel conducting vector control fluctuated between 34 and 100 per month. More personnel were required as the study progressed to service the traps that had already been deployed, besides adding traps to yet untreated structures. Vector control personnel were asked to visit each structure up to three times on different times and days, including weekends, to be able to contact the residents.

We gathered the following data: name of resident, address, geographical coordinates, telephone number, date of visits, unique structure identifier, treatment cluster, status of premise (no response, rejected treatment, treated, abandoned, vacant), types of containers treated or managed (covered, placed under a roof) to prevent mosquito production, number and identifier of each IAGO trap installed, and if larvicide was used. Data were collected using a password protected, encrypted software application that was developed in-house and was capable of running on cell phones, tablets, and personal computers. Data were stored and managed in a CDC server.

The number of adult Ae. aegypti mosquitoes eliminated by mass trapping during the study was calculated by multiplying the average number of mosquitoes captured in SAGO traps by the number of deployed IAGO traps per week. This method is based on an investigation carried out in Villodas, Guayama Municipality, Puerto Rico from December 2016 through April 2017, where we found no significant differences between the numbers of female Ae. aegypti mosquitoes captured per trap per week in 27 SAGO traps (1 per house) and in 60 randomly selected IAGO traps out of a total 570 IAGO traps deployed (three per house; CDC unpublished). Characteristics of the Villodas study site are published (Barrera et al., 2014b).

Because clusters were treated at different times and because Ae. aegypti populations are mainly driven by rainfall and relative humidity in Puerto Rico (Lega et al. 2017), we used meteorological parameters as covariates in all analyses. We placed and operated meteorological stations (HOBO Data Loggers, Onset Computer Corporation, Boume, MA) in the center of each of the eight clusters, where daily temperature, relative humidity, and rainfall were recorded throughout the intervention. To relate weather variables to mosquito abundance, we averaged daily temperature and relative humidity for three weeks before sampling and accumulated rainfall during the third and second weeks before sampling. Assuming a mosquito life span of three weeks, the number of adult Ae. aegypti mosquitoes captured in the traps may have been influenced by temperature and relative humidity during the previous three weeks. Accumulated rainfall during the week before sampling is not expected to modify the number of adult mosquitoes in the traps as immature development lasts about a week.

We tested the hypothesis that the average number of mosquitoes captured in SAGO traps significantly decreased as the percentage of treated houses increased. We used a Generalized Linear Mixed Model (GLMM) analysis on the number of female Ae. aegypti per trap per week with percentage of houses treated in the vicinity of each trap (Thiessen polygon) as the main factor and meteorological parameters as covariates. Clusters (1–8) were used as random factors (Hemming et al. 2015). We used a negative binomial model with log link for all analyses because the variance of the count variable was greater than the mean, and a first-order autoregressive function as the covariance structure of the repeated measures. We repeated this analysis using an interval-transformed variable of the percentage of houses treated, with the following values: 0, 1–20, 21–40, 41–60, 61–80, and > 80% that represent the progress in treatment coverage in time. We were particularly interested in investigating what would be the minimum treatment coverage necessary to cause a steady, significant reduction of the Ae. aegypti population. A posteriori mean comparisons were analyzed using adjusted sequential Bonferroni tests at a significance level of 0.05.

We contrasted the density of Ae. aegypti before and after treatment for each of the clusters using Generalized Estimating Equations (GEE). We considered that a cluster had been minimally treated at the time when the percentage of treated buildings reached 60%. This resulted from the observation that the overall density of Ae. aegypti stabilized around two females per trap per week when treatment coverage exceeded 60%, as reported below. We considered a transition time of three weeks for the Ae. aegypti population to respond to treatment (Fig. 3), as observed before (Barrera et al. 2018a). Data collected during the transition time was not included in the analyses.

We used a GEE for a citywide comparison of final and pre-treatment Ae. aegypti densities. The pre-treatment period included eight weeks from October 10 – December 5, 2016 (vector control operations slowly started in November 14, 2016) and was compared with an eight-week period before the end of the project, from June 12 to August 4, 2017. Meteorological co-variates were used as before to control for differences in weather between final and pre-treatment periods. Results were reported as means and standard errors. Statistical analyses were carried out using IBM SPSS Statistics 25 software (IBM Corporation, Armonk, NY, USA).

We calculated the maximum likelihood minimum infection rates (MIR) of mosquitoes using PooledInfRate version 4.0 (Biggerstaff 2016). The Vector Index [VI = (N * P)* 1,000], an indicator of the expected number of infected mosquitoes per trap per week, was calculated as the proportion of infected mosquitoes (P = positive pools / total pools tested) times the average number of mosquitoes (N) captured per trap per week per 1000 (Jones et al. 2011).

Fifty community leaders participated in the focal groups. The leaders were 20–81 years old (average 56 years old) and 58% were females. Most leaders (98%) had consensus that AGO traps would reduce the number of mosquitoes in their homes and were willing to allow the placement of three traps in their yards. They also expressed that risk of infection with Zika would be reduced by using the traps. Community leaders felt that required trap maintenance would be easy. They suggested that community leaders and neighbors would be trusted sources for educating residents about AGO traps and assisting with trap maintenance, although residents were not asked to service the traps. In practice, community leaders were effective facilitators, particularly in some communities that required their presence to allow technical personnel to apply vector control.

Residents of 20,235 structures accepted installing three or more AGO traps, depending on the size of the property. We deployed 78,126 IAGO traps during the intervention with an average of 3.3 ± 0.01 traps per structure omitting structures with more than 10 traps (N= 511) to cover large areas such as cemeteries and parks. Treated structures reflected the predominance of single story houses but several large properties were also treated, such as shopping centers, industries, and cemeteries. Average time to servicing traps and reapplication of source reduction and larviciding when needed was 94.00 ± 0.17 days (median = 82 days). Commonly found containers that required disposal, management (store under a roof, cleaned up, etc.) or larviciding were: water meters, trash cans, cavities in the structure or its surroundings such as concrete or fencing, bromeliads, plastic buckets, flower pots, barrels, and lids on top of buckets or barrels.

Access to treatment varied among clusters, so that the actual sequence in which treatments was completed varied from the initial plan (Fig. 3). An important aspect affecting treatment schedule and slowing down operations was the need to revisit structures at least three times before giving up because of residents’ absenteeism. Structures for which we did not receive authorization (absent resident, refusal) were excluded from treatment. The first round of treatment was completed in all clusters by June 2017. Trap coverage (percentage of buildings treated) was 79 – 84% in six clusters (19,323 buildings) and 67% in two clusters (clusters 2 and 4; 6,068 buildings). Clusters 2 and 4 had the smallest percentage of residential properties of all clusters (25 – 37%; Table 1). The number of contractors dealing with IVM and trap servicing increased from 34 –77 during the first months (November – April) to 80 – 100 later because of the increasing number of traps requiring maintenance.

We conducted 15,480 SAGO trap weekly inspections (traps * weeks) from the beginning of the study until August 4, 2017 when trap removal started. We captured 62,760 female Ae. aegypti (4.06 ± 0.04 mosquitoes/SAGO trap/week; min, max = 0, 72). Rainfall peaked during late November 2016 and early December 2016, when we observed the largest density of mosquitoes in the SAGO traps (Fig. 4). Rainfall steadily decreased towards March 2017, as is normal during this period in Puerto Rico (Barrera 2010) and mosquito density concomitantly decreased. Mosquito density then stabilized at a density just above two mosquitoes per trap per week when trap coverage reached 60% or above (mid-March – May 2017; 2.10 ± 0.04 mosquitoes/trap/week) in spite of increased rainfall. In subsequent months (June – August 4, 2017) overall mosquito density stabilized below 2 mosquitoes/trap/week when trap coverage was at or above 80% in most clusters (1.4 ± 0.04; Fig. 4).

The GLMM model of the average number of mosquitoes captured in SAGO traps per week, as a function of vector control and weather variables, was significant (F_{4, 15067}= 628; P< 0.001). Significant effects were the cumulative percentage of houses that had been treated in the vicinity of the SAGO traps (F_{1, 15067}= 779; P< 0.001), temperature during three weeks before sampling (F_{1, 15067}= 30; P< 0.001), relative humidity during three weeks before sampling (F_{1, 15067}= 150; P< 0.001), and cumulative rainfall registered on the third and second weeks before sampling (F_{1, 15067}= 13; P< 0.001). The exponent of the coefficient for the percentage of houses treated (0.986) indicated a 1.4% reduction in the number of mosquitoes per trap per week per unit increase in the percentage of houses treated.

We repeated this analysis using the percentage of houses treated per week as an interval variable (0, 1–20, 21–40, 41–60, 61–80, and >80%) in order to estimate what mosquito reduction effects existed at each level of vector control. The model was significant (F_8,14588= 296; P< 0.001), with significant effects of the interval-coded percentage of treated houses per week treated in the vicinity of the SAGO traps (F_{5, 14588}= 146; P< 0.001), temperature during three weeks before sampling (F_{1, 14588}= 8.5; P< 0.01), relative humidity during three weeks before sampling (F_{1, 14588}= 163; P< 0.001 ), and cumulative rainfall registered on the third and second weeks before sampling (F_{1, 14588}= 13.2; P< 0.001). We saw no significant mosquito reduction of the average number of mosquitoes when control coverage was 0 – 20%, and significant reductions (α < 0.001) when control coverage increased to 21 – 40% (34.3% reduction), 41 – 60% (42.4%), 61–80% (62%), and >80% (81.5%; Fig. 5). The observed and predicted densities of Ae. aegypti per trap per week when 80% or more of the houses had been treated were 1.41 ± 0.05 and 1.68 ± 0.20, respectively.

Each cluster had specific characteristic changes in mosquito density, rainfall, and treatment schedule (Figs. 6, 7) but the general tendency of mosquito density of reaching small and steady values with increasing treatment coverage in each cluster was similar to the overall average tendency (Fig. 4). Clusters 1 and 8 had less rainfall than the other clusters and Cluster 8 had the highest average temperature and lower relative humidity, but in general, weather among clusters was not strikingly different.

The GEE comparisons of Ae. aegypti density before and after reaching 60% treatment (allowing a transition period of 21 days for the treatment to take effect) were significant for every cluster (Table 3). The meteorological parameters of relative humidity and rainfall were the covariates more frequently and significantly associated with changes in mosquito density other than vector control. The citywide comparison of final versus initial Ae. aegypti densities was significant for vector control treatment (χ2= 709.0; P< 0.001), rainfall (χ2= 59.0; P< 0.001), and temperature (χ2= 16.1; P< 0.001). Average initial and final densities of Ae. aegypti were 7.97 ± 0.12 and 1.41 ± 0.04, respectively. This result translates into an effective reduction of the mosquito population of 82.3% {[1- (Final density after treatment / Initial density)] * 100}.

We analyzed 12,081 pools of female Ae. aegypti for the presence of virus RNA of DENV, CHIKV, and ZIKV until 4 August 2017. We found 49 positive pools for ZIKV, seven for CHIKV, and two for DENV. The only positive pool found after March 2017 when the treatment coverage was above 60% was one positive pool of DENV in July (Fig. 8). ZIKV was detected in consecutive weeks from the beginning of the study in October 2016 until February 2017. We did not find any ZIKV positive pool in cluster 7, 15 positive pools in Cluster 3, and the rest of clusters had between four and seven positive pools each. During that period ZIKV MIR per week was 1.10 (0 – 2.81) mosquitoes (per 1,000) and VI was 8.87 (0 – 28.12) (per ,1000). The average density of Ae. aegypti per trap per week that yielded positive pools for ZIKV was 9.8 ± 0.1 (N= 49; range 1 – 37). CHIKV was detected in low, sporadic frequency from November 2016 to March 2017. Positive pools came from clusters 1, 3, 4, and 7. CHIKV MIR per week was 0.25 (0 – 1.32) mosquitoes (per 1000) and VI was 1.03 (0 – 2.81). Positive pools for CHIKV came from traps with an average density of Ae. aegypti of 10.3 ± 3.7 (N=7; range 1 – 29). The two DENV positive pools came from clusters 1 and 8. Positive DENV pools came from traps with 2 and 6 females of Ae. aegypti.

The total expected number of adult Ae. aegypti captured in AGO traps from deployment in November 2016 until August 2017 was 6,223,826. Given that the human population under treatment were 61,511 inhabitants, the expected number of mosquitoes removed per person was around 100 or 10 per person per month. The expected number of adult mosquitoes removed peaked in December 2016 and January 2017, and decreased afterwards. The cumulative number of mosquitoes removed per month steadily increased during the intervention without evidence of leveling off.