Non-household environments make a major contribution to dengue transmission: implications for vector control

Abstract

The incidence of Aedes-borne pathogens has been increasing despite vector control efforts. Control strategies typically target households (HH), where Aedes mosquitoes breed in HH containers and bite indoors. However, our study in Kenyan cities of Kisumu and Ukunda (2019-2022) revealed high Aedes abundance in public spaces, prompting the question: How important are non-household (NH) environments for dengue transmission and control? Using field data and human activity patterns, we developed an agent-based model simulating transmission across HH and five types of NH environments, which was then used to evaluate preventive (before an epidemic) and reactive (after an epidemic commences) vector control scenarios. Our findings estimate over half of infections occurring in NH settings, particularly workplaces, markets and recreational sites. Container removal was more effective in NH than in HH areas, contrasting with the global focus on HH-based management. Greater reductions in dengue cases occurred with early, high-coverage interventions, especially in NH locations. Additionally, local ecological factors, such as uneven water container distribution, influence control outcomes. This study underscores the importance of vector control in both HH and NH environments in endemic settings. It highlights a specific approach to inform evidence-based decision-making to target limited vector control resources for optimal control.

Headnote

Keywords: dengue, agent-based model, non-household environments, vector control

The incidence of Aedes-borne pathogens has been increasing despite vector control efforts. Control strategies typically target households (HH), where Aedes mosquitoes breed in HH containers and bite indoors. However, our study in Kenyan cities of Kisumu and Ukunda (2019-2022) revealed high Aedes abundance in public spaces, prompting the question: How important are non-household (NH) environments for dengue transmission and control? Using field data and human activity patterns, we developed an agent-based model simulating transmission across HH and five types of NH environments, which was then used to evaluate preventive (before an epidemic) and reactive (after an epidemic commences) vector control scenarios. Our findings estimate over half of infections occurring in NH settings, particularly workplaces, markets and recreational sites. Container removal was more effective in NH than in HH areas, contrasting with the global focus on HH-based management. Greater reductions in dengue cases occurred with early, high-coverage interventions, especially in NH locations. Additionally, local ecological factors, such as uneven water container distribution, influence control outcomes. This study underscores the importance of vector control in both HH and NH environments in endemic settings. It highlights a specific approach to inform evidence-based decision-making to target limited vector control resources for optimal control.

1. Introduction

Vector-borne diseases are a group of infections caused by pathogens such as parasites, bacteria and viruses that are transmitted by biting arthropods; together, they put 80% of the world's population at risk [1]. Dengue virus, transmitted by Aedes mosquitoes, is among the most important vector-borne diseases because of the close relationship with human environments and its large and growing burden [2,3]. Dengue is estimated to cause around 400 million infections globally per year [4,5], and recent trends show an increase in cases annually [6].

This increase in dengue transmission is occurring despite the implementation of control activities in endemic settings. While there is some uncertainty about the effectiveness of actual control strategies due to lack of reliable evidence [7], some authors argue that vector control measures are inadequately implemented [8-10], and others add that an integrated, community-focused control requiring multisectoral, multi-disciplinary engagement and community participation sustained in time is necessary [11]. As a result, one of the main issues to be addressed is the improvement of vector control programmes [12].

The design of vector control strategies should be as effective as possible while minimizing the required costs, time and human resources. In search of this efficiency, vector control strategies have predominantly focused on households (HH) [13]. The rationale behind this assumption is that people spend more time in HH environments than in any other structure while sharing the same space with cohabitants and biting, breeding vectors [14]. As a result, most vector control guidelines exclude non-household (NH) locations as targets of interventions [14-16]. Yet, recent evidence suggests a larger role of NH environments in infection risk [17-21], which was supported by a significant reduction of cases reported during the COVID-19 pandemic lockdowns, when people spent more time in HH and less time in public or outdoor spaces [22,23].

To better understand the role of NH environments in dengue transmission, we developed an agent-based model of dengue transmission and calibrated it to data on mosquito breeding places, abundance and human activity space in various environments from two Kenyan cities [24]. With the model, we aimed to estimate the proportional role of both HH and NH environments in dengue transmission and evaluate the outcome when container removal control strategies are focused on either or both types of environments (HH and NH) and are either preventive (carried out before an epidemic) or reactive (carried out in response to an epidemic).

2. Methods

2.1. Model overview

2.1.1. Simulated populations and environments

To reach our goals, we developed an agent-based model incorporating data and conditions from Kenyan cities of Kisumu (located at the western part of the country next to Lake Victoria; figure 1) and Ukunda (a coastal city in the eastern part of the country; figure 1). Dengue virus circulation and endemicity have been described in these cities for a long time [25], where higher levels of dengue have been reported for Ukunda than Kisumu [26], spanning a wide range of endemicity levels within these populations.

To develop the model, we created two synthetic populations representing each of the Kenyan cities. To do this, we considered the number of inhabitants per HH for each city according to information provided by reports of the 2019 Kenya Population and Housing Census [27], which is also in accordance with previously reported HH occupancy information for both cities [24], where the mean number of inhabitants per HH is 4.6 for Kisumu and 7.3 for Ukunda. For tractability and scalability, we set the size of synthetic populations to be around 20 000 individuals (final size was 20172 for Ukunda and 20 160 for Kisumu), so the number of HH was set according to both the size of the human population and the mean number of house inhabitants.

Following the aim of this work, we additionally created a synthetic NH environment. For this, we considered information related to the presence of water containers in these locations and available data about the presence of people at different urban spaces (see electronic supplementary material and §2.2) to define five different types of NH environments: workplaces, schools, religious spaces (representing churches, mosques, etc.), markets (including shopping places of any kind) and recreational spaces (grouping any place where people attend for entertainment or gatherings, like bars, nightclubs, parks, etc.). The number of workplaces and schools were defined according to information extracted from previous reports informing the average number of either workers [28] (rounded to 19) or students [29] (rounded to 360) for each of the respective environments. Unfortunately, for the remaining NH structures, there was no available information related to their proportion or density within cities. By considering survey results from local populations, we defined their density as one market and recreational place for every 30 houses and one religious space for every 50 houses.

We used data on proportions of water-holding containers in HH and NH environments that we previously published [24]. With those proportions, each structure was assigned a specific number of water containers, each with an assigned size based on our data on the size distribution of containers [24].

Each individual of the synthetic population was assigned an age following the proportions reported in the Census of 2019 [27]. Each individual was also assigned an occupation, being either student (can be as young as 3 and up to 17 years old), worker (from 15 and up to 64 years old), both or none (e.g. toddlers or retired), following the general student and worker age reported in the Kenyan Quarterly Labour Force Report (2021) [28] and the Basic Education Statistical Booklet (2019) [29]. We assumed an initial baseline prevalence of dengue of 0.08%, estimated from previous studies reporting age-structured seroprevalence [30], with an incidence rate per year estimated as IR = - 1 - prevalence).

2.1.2. Population dynamics

The model was configured to simulate on a daily basis what happens in each structure where humans and vectors are present, and hence infections can take place. The presence of humans at a given location depends on the type of structure, where HH, workplaces and schools are daily attending locations, and religious, market and recreational places are daily randomly assigned attendings (figure 2). In this sense, for each structure falling into the first category, the same individuals attend daily. For the latter three types of structures, the number and selection of individuals occur randomly on a daily basis (see figure 2 and electronic supplementary material). Finally, we also included movement among HH with a probability of 0.1 for a given HH to receive a non-resident individual each day. To consider mosquito movement, we turned to information from studies that release large numbers of mosquitoes, Which may promote mosquito movement, which have shown that roughly 90% of mosquitoes are recaptured either within 30 m of the release point or even in the same house [31-33]. For this reason, we assumed that mosquito movement between buildings for an already established subpopulation is negligible (no movement). Instead, in the model, the human movement is the primary driver of virus spread.

When a given structure has mosquitoes and humans, there is a chance of transmission if any of them are infected. The probability of a human being bitten in a given structure depends on the number of both mosquitoes and humans and the time that humans spend in the structure. For simplicity, each structure was assigned a specific number of hours for people to spend that is defined according to the type of structure. Data to define the number of hours per structure were estimated based on fieldwork conducted in the same study cities (see 82.2 and electronic supplementary material).

Mosquito population dynamics are not determined at a city-wise level but at a structure-wise level, recognizing that different patterns of transmission are obtained when space fragmentation is considered [34]. To do this, we considered that the main limiting driver of mosquito population growth occurs at the larval stages when mosquitoes are more metabolically active: eating, growing and molting and are subject to competition for resources [35,36] and hence depending on the breeding place availability, their volume and larval density. Accordingly, we developed a density-dependent function to estimate the number of individuals in the next discrete time (day) describing the larval survival as a function of the density of mosquito immature stages and temperature, as follows:

(ProQuest: ... denotes formula omited.)

where

d = -0.166+0.08T -0.0014T2.

In this equation, a = 0.55 and b = 0.09 are calibrated values (see electronic supplementary material), d is a term that relates temperature (T) with the remaining terms in the equation and D is the larval density expressed as the ratio of the number of larvae and the number of litres of water available for breeding in the structure. This density-dependent function was applied alongside a comprehensive set of temperature-dependent biological processes, accounting for development, fertility and mortality, among others, previously described through mathematical functions [37-39] and integrated with statistical distributions. While distributions like quasi-Poisson and negative binomial effectively fit mosquito count data, our model's population dynamics are primarily driven by biological processes. This approach ensures biologically realistic population dynamics while incorporating natural variability that mechanistic functions alone cannot capture (see electronic supplementary material for details related to implementation of functions in this work).

2.1.3. Infection dynamics

For a susceptible mosquito that bites an infected human, it can be moved from the susceptible to the exposed stage according to temperature-dependent vector competence and later be moved to the infectious stage with the temperature-dependent extrinsic incubation period following functions previously described (electronic supplementary material, table S2) [37]. Infectious mosquitoes remain in this stage until death (figure 2), for which the rate also depends on temperature using a previously published function (electronic supplementary material, table S2) [37].

Humans that are bitten by infected mosquitoes are moved to a stage where they remain for 5 days. At the end of this period, they are moved to the infectious stage, lasting 7 days before moving to a recovered stage (figure 2). Since the model does not explicitly simulate the circulation of dengue virus serotypes, we considered a period of complete protection before returning to a susceptible state to account for multiple serotype infections. Based on Sabin's classic works of experimental infections [40,41], complete heterotypic protection can be lost by 3 months after exposure, so we set a complete heterotypic immunity lasting for 100 days.

The type of structure where each infection took place and the date were recorded. Each computational run considers a temporal window of 731 days comprising between 1 January 2020 and 31 December 2021. Data are temporarily grouped, yielding the number of cases happening every week. Results are expressed as the median and interquartile range (IQR) of infections among 400 runs. The model was coded in Julia language v. 1.8, and all simulations were run on the Sherlock cluster at Stanford University (Stanford Research Computing Center).

2.2. Data acquisition

Information related to the number of containers in different environments was obtained from fieldwork performed between 2020 and 2022 in both study sites and previously reported in detail [24]. Briefly, 400 m? urban areas were sampled with four different strategies targeting different stages of mosquitoes: ovitraps to obtain information on eggs and egg-laying females, container surveys to obtain information on larval stages and container availability, Prokopack aspirators to obtain information related to adults, and BG-Sentinel to obtain information from oviposition site-searching females. Information on the sample location was recorded, including the type of environment (HH or NH), number of water containers and their size category, the Aedes positivity status and the number of house inhabitants.

Serological data were obtained only for purposes of calibrating the model (electronic supplementary material). Data consisted of sero-incidence estimates based on individuals of all ages yielding a negative serology followed by a positive test during a follow-up examination 6 months later. Though the study contains data from 2014 to 2022, only those individuals recruited during the same temporal window of this study were considered for calibration, i.e. from 2019 to 2022. Enrolment of individuals [42,43] and serological methods [26] have been previously described.

Temperature data were collected in both cities by using temperature data loggers (HOBO®), and the daily average was calculated during the entire simulated period comprising from 1 January 2020 until 31 December 2022.

2.3. Human movement survey

We carried out a semi-structured interview (SSI) to gather information about people's movement routine in Kenyan settings from 1 November to 2 December 2021. SSIs have been previously employed to gather data on routine human movement in other settings [44]. The survey was carried out in two community cohorts corresponding to both Kenyan cities included in this work, i.e. Kisumu and Ukunda. Both cohorts are part of an ongoing longitudinal study [26]. We interviewed 201 individuals in Ukunda and 243 in Kisumu, carefully selected to represent the gender and age group distribution of their respective cohorts. The SSI included questions aimed at capturing commonly visited locations during weekly activities. Participants were asked to list locations, besides their HH, such as workplaces, markets and shops, schools and religious places that they usually visit in a week. Participants also provided an estimated amount of time spent in each location per week. The SSI was conducted by trained local technicians who provided an overview of the study. Questionnaires targeting children unable to respond on their own were completed by their guardians. For subjects under 18 years of age capable of answering the questionnaire, we obtained permission from their guardians. All locations listed by participants were georeferenced by the survey team, either by collecting Global Positioning System ground points or by gathering coordinates through Google Maps. The questionnaire used in this study is available as electronic supplementary material.

2.4. Vector control strategies assessed

The strategies tested in this model are focused on the reduction of vector populations through container reduction. The intensity of the control was quantified as the reduction percentage of water-holding containers available as mosquito breeding places. Container elimination was evaluated under two control scenarios termed 'preventive' and 'reactive'. The preventive control scenario refers to the elimination of containers on day 0 preceding the start of an epidemic. To ensure a proper start of the epidemic at the designated time, there were no infections happening in the 2 weeks preceding it. On the other hand, the reactive control scenario refers to the elimination of containers after the rise in cases defining the start of an epidemic. Because reactive control initiatives can take some days to be implemented for several reasons (planning, resource allocation, recruitment, among other stages), we considered 1, 50 and 100 days after the start of the epidemic as different reaction times (the day when control was accomplished, additional results for control implemented after 250 days are included in the electronic supplementary material).

For both control scenarios, the effort of the control strategies was quantified as the percentage of water containers removed. Given that our goal was also to quantify the contribution of urban HH and NH environments in the total number of cases, we additionally evaluated the effectiveness of control when it is focused only on one or both of these categories of environments. In addition, we also included the outcome of the control when it is focused on large (those with a volume >10 1) or small (with a volume <10 I) containers to provide insights on different mosquito population-related parameters and their effects on transmission and subsequent control effectiveness. In a realistic scenario, implementing control activities would take several days, with varying durations across different environments-likely longer in NH. While this may seem unrealistic, in our simulations, control measures were applied in a single day. This approach was chosen to theoretically describe the effects of interventions rather than to simulate specific activities or policies.

3. Results

3.1. Infections in non-household environments are higher than expected

During a period of 731 days (i.e. 2 years), the model yielded a median of 784 (IQR: 350-1557) infections in Kisumu and a median of 3971 (IQR: 2680-5445) in Ukunda (figure 3). By explicitly quantifying the number of infections happening in different urban environments, the model estimated a slightly higher proportion of infections taking place in NH structures, accounting for around 57.4% (IQR: 47.7-58.2) of infections from Kisumu and 56.3% (IQR: 53.1-57.4) of infections from Ukunda. Workplaces were the most frequent NH site of infections, accounting for 77.1% (IQR: 54.7-99) in Kisumu and 30.9% (IQR: 28.8-32.4) in Ukunda. Markets and shopping locations accounted for 9.3% (IQR: 0-18.8) and 27.4% (IQR: 26.2-29.2) of NH infections for Kisumu and Ukunda, respectively. Recreational locations accounted for 10.9% (IQR: 0.0-19.4) of NH infections in Kisumu and 21.4% (IQR: 20.3-21.8) in Ukunda. Finally, religious places and schools had the lowest number of infections, accounting for 2.7% (IQR: 0.0-7.1) and 0% (no infections recorded) for Kisumu and 18.9% (IQR: 17.9-19.1) and 1.3% (IQR: 0.0-5.1) for Ukunda.

In addition, we wanted to understand how infection risk for children (15 years old and younger) and adults (those older than 15 years) was distributed in different environments. Though infections in schools only took place in Ukunda, in that setting, a higher proportion of infections in children is happening in schools (for children from 3 up to 15 years old), as expected, while a higher proportion of infections in adults happen in workplaces. Following schools, the model predicted the highest proportion of infections in children happening in HH, followed by recreational places for both cities (figure 3).

3.2. Differential effectiveness in preventive control among urban environments

NH environments provide a powerful lever for vector control through container removal, especially when vector control was preventative. Control by removing containers is more effective in the NH environment, while the impact was less pronounced in the HH environment; e.g. considering 50% control intensity in Kisumu, HH-only control reduced cases by 54.2% (IQR: 10.1-73.3) while NH-only control reduced cases by 68% (IQR: 48.4-80.4), and controlling both reduced cases by 76.6% (IQR: 61.3-85.7). In Ukunda, the difference is even more dramatic: 35.6% (IQR: 2.2-66.2) reduction in cases from control in HH containers alone, 84.6% (IQR: 62.9-95.4) reduction for NH containers alone and 93.2% (IQR: 80.4-97.5) reduction for control in both HH and NH (figure 4).

On the other hand, the importance of container size differed between cities, where small containers (<10 1 volume) were more important for Kisumu, while removal of large containers (>10 1 volume) yielded a greater reduction of cases in Ukunda. Thus, when 50% of control intensity is applied in both environments in Kisumu, a reduction of 66.6% (IQR: 45.6-79.6) of cases can be seen when only small containers are removed, while a 49.5% (IQR: 0.0-70.3) reduction resulted from removing only large containers in Kisumu. Nevertheless, a greater reduction is observed when the control is done, irrespective of the size of the container, with a reduction of 76.6% (IQR: 61.3-85.7) in the same city. On the other hand, in Ukunda, removing only small containers led to a 34.9% (IQR: 0.6-66.1) reduction of cases versus a 73.5% (IQR: 43.7-94.8) reduction when large containers are targeted. Similar to Kisumu, the greatest reduction in cases resulted when control was performed on all container types, with a 93.2% (IQR: 80.4-97.5) case reduction in Ukunda (figure 4). It is worth noting that the number of removed containers differs among cities when targeting different container sizes (electronic supplementary material, figure S5). Intriguingly, the number of removed containers is consistently higher for different control intensities at HH environments (electronic supplementary material, figure 55) than NH. In this sense, targeting NH should require less effort in terms of total containers removed than HH and hence more efficiency.

3.3. The effectiveness difference between household and non-household is consistent between preventive and reactive strategies

As expected, the greatest reduction in cases is observed when vector control strategies are implemented sooner at the beginning of the epidemic, showing timing to be as important as the environment. For example, considering a 50% control intensity, the effectiveness declines from 74.3% (IQR: 60.7-85.8) when control is implemented at day 1 to 44.3% (IQR: 20.5-62.7) when control is implemented at day 100 for Kisumu and from 89.1% (at day 1; IQR: 75.3-94.9) to 63.6% (at day 100; IQR: 44.9-78.8) in Ukunda, representing an increase in effectiveness of 30% and 25.5% for Kisumu and Ukunda, respectively. By increasing the intensity to 100%, an upper bound on the potential for control through container removal, the effectiveness shifts from 90.2% (IQR: 86.3-96.1) at day 1 to 52.9% (36.1-66.8) at day 100 in Kisumu and from 98.5% (IQR: 97.8-99.0) at day 1 to 80.4% (IQR: 72.5-82.9) at day 100 in Ukunda. On the other hand, when we examined the change in the effectiveness by increasing the control intensity at day 50, it goes from 68.8% (IQR: 57.5-77.4) to 44.99% (IQR: 8.3-65.0) when intensity is shifted from 100% to 25% in Kisumu. In Ukunda, the change is sharper, going from 91.8% (IQR: 88.6-94.4) effectiveness at 100% of control intensity to 37.7% (IQR: 7.25-62.9) effectiveness at 25% of control intensity (figure 5).

Like in the preventive scenario, for reactive control the highest reduction is achieved when container removal is applied in HH and NH environments, capturing their unique contributions to transmission. However, when control is applied in only one or the other, control targeted to NH environments was more effective than that targeted to HH (figure 5 and electronic supplementary material, figures S2 and S3).

4. Discussion

Recent fieldwork reported a higher number of vectors in NH environments than in HH in the Kenyan cities of Kisumu and Ukunda, suggesting potentially high risk for Aedes-borne virus transmission in these environments [24]. Consequently, this work expands our knowledge about the total burden of transmission varying based on human activity space within HH and NH environments as well as others like container type and density between cities. Here, we developed an agent-based model that incorporates field data on vector occurrence and abundance, container density and type, and human activity space across age structure in Kisumu and Ukunda to explore the consequences of NH environments for dengue transmission. Specifically, we tested the hypothesis that dengue vector control could be improved by extending it to NH spaces. The model supports this hypothesis, demonstrating that the contribution of NH spaces to transmission is as high as or higher than HH (figure 3), and though the higher efficiency is achieved by focusing only in NH environments (electronic supplementary material, figure S5), the higher reduction of cases was reached by controlling vectors in both NH and HH environments (figures 4 and 5).

Rather than focusing solely on the number of infections in NH, our model suggests that these environments act as spreaders of the virus among HH. In this way, while the number of new infections per HH is limited by HH size, the high levels of movement of individuals in NH environments provide a source of new infections and transmission spreading among HH. In line with this, previous modelling work had suggested the importance of the movement of people in dengue transmission [45- 48]. Itis very likely that this role is also mediated by intermediate spaces among HH, like those defined in this work as NH environments, such as workplaces, schools, social spaces, religious spaces and marketplaces, where the presence of vectors had already suggested a role in transmission [17,18,21].

Among the five categories of NH spaces, workplaces contributed most to transmission (figure 3). This is not the first work suggesting such a large contribution. Previous work developed on a Zika outbreak in Singapore by Prem and colleagues predicted an even higher proportion of infections, with an estimate of 64% (at least 51%) of infections happening at workplaces [49]. Besides HH and schools, workplaces are the locations where individuals spend most of their time, which increases the probability of being bitten by a mosquito (see electronic supplementary material for details of parametrization). Further modelling considering different types of workplaces like offices, open spaces (e.g. open-air markets or farms) or semi-enclosed spaces, as well as the proportions of each in each city, would enhance the understanding of risk at workplaces. For these spaces, we considered an average of 19 people per workplace (electronic supplementary material), which is higher than the average number of inhabitants per HH and hence increases the probability of having an infected individual at a given timepoint. By contrast, in schools, which have a considerably higher average number of students of 360 (higher than HH and workplaces; see electronic supplementary material), while the probability of having an infected person at any given time is high, the probability of a mosquito biting the infected individual among the entire student population is low, and the density of mosquitoes in schools is not high enough to compensate for this low per-person biting probability. This is especially true in Kisumu, where low overall incidence explains the lack of infections in schools, while the high incidence in Ukunda increases the presence of infected individuals in schools and hence the probability of human-to-vector infections and subsequent spread. Considering the population size with which we worked (see 82), previous reports of infection risk in schools [18,19] and the vulnerability of children that congregate in schools, our data suggest that this type of environment still represents some level of infection risk, and its inclusion in vector control activities is necessary. This is particularly true, as this is the main NH environment where children are at risk for dengue exposure.

In line with this, the proportion of infections in children and adults in different environments likely reflects the age structure of individuals visiting these locations. In this model, we assume that the epidemic starts in a fully susceptible population; accounting for age-structured variation in pre-existing immunity would possibly alter the risk scenario for children [50]. This assumption is realistic for a new serotype invading the population at least 100 days after the most recent epidemic.

The scientific literature supports the presence of significant risk in spaces other than HH in other parts of the globe [18-21]. Some of these previously identified spaces that were not evaluated in this work include abandoned and open spaces and hotels because we did not have data available on time spent in these locations. Accordingly, it is possible that a slight increase of NH infections is still to be quantified by considering those environments.

Results under the preventive and reactive scenarios yielded similar results: higher effectiveness is achieved when container removal-based vector control strategies consider only NH compared to only HH, and a combined approach that includes HH and NH is most effective. This points to an important disconnect between our results and current vector control practices, which primarily focus on container removal in HH spaces and neglect NH vector control [14-16]. Scientific literature related to vector control HH-focused interventions is extensive [13,51] and rarely gives NH settings similar weight. Though we cannot definitively attribute the lack of successful dengue control to transmission in NH settings, our data-driven model suggests that this could be an important part of the problem, and we advocate for future studies in other locations studying this phenomenon as well as potential approaches to NH vector control.

Our model includes a novel mosquito density-dependent function that allows us to realistically model vector population dynamics from a larval perspective and at the scale of individual containers (as described by McCormack and colleagues [34]), which also allows us to estimate the relative importance of different container sizes in transmission and control (see electronic supplementary material). By using the function, our results suggest that the relative importance of different types of containers is city-specific since it depends on the frequency of these containers across cities. Accordingly, small containers in Kisumu are much more frequent than large containers (see electronic supplementary material), so slightly higher effectiveness is achieved when control by removing containers is targeting only them. The situation is different in Ukunda, where a higher effectiveness is observed when focusing on both types of containers, where control targeting large containers is more effective, as this type of container is more productive [52,53]. These results suggest that the effectiveness of container-focused vector control depends on a trade-off between the productivity of containers and their frequency, where large containers are more productive but less frequent than small containers. Containers larger than 10 1 were not included in this work as these were the less common container size, and the density-dependent function might yield unrealistic mosquito population growth for such large containers. Though the model is using fieldwork-derived data, the purpose of this model is understanding the proportional role of different urban environments. We think a separate analysis should be done in order to specifically evaluate the relative importance of different container types and the best approach to take advantage of their differential distribution to achieve the best outcome.

This model is meant to realistically represent the transmission conditions in both cities but has some limitations. Though agent-based models are excellent for capturing heterogeneity and variation within populations, especially when rich data are available for parametrization, our model does not estimate some other potential sources of variation, like mosquito movement or variability in time in the number of breeding places. In this model, mosquito movement between structures, such as those between HH and NH, is not considered. Potentially, mosquito migration could slightly alter the number of infections in environments, particularly if infected vectors are involved in these movements. Additionally, the addition of other types of buildings beyond the six types we included in this study might provide a more comprehensive perspective of urban locations where transmission can be happening, like those described previously by our team [24]. Our estimates of movement are mainly based on the time people spend in given locations, but other potential sources of variability were not included like intra-urban distances [47,54], decrease of mobility due to illness [55] and travel to other urban centres and rural areas. Movement data were collected using SSI (see §2), which relies on people's recollection. Consequently, the collected data may be influenced by recollection bias, a common limitation of SSIs. Recollection bias could lead participants to list only highly visited locations, potentially overlooking places that are less frequently visited. Additionally, the time spent in each location may be affected by participants' different senses of time, which are linked to their individual characteristics.

Unfortunately, data related to the density of mosquitoes in specific NH locations were not available or did not match with information collected from the human movement data survey. Some assessed spaces reported by Peña-García et al. [24] having mosquitoes like 'open spaces', 'gardens' or 'banana plantation" were not reported by people as places where they spend any time. Likewise, NH categories included in the model like 'workplaces' can group some of the categories reported by Pefia-Garcia et al. [24]. For these reasons, the initial conditions for the mosquito dynamics were the same for all NH buildings in the model. In this way, differences related to different NH locations are mainly due to the number of people attending the locations and the time spent in these. It is important to mention that mosquito positivity status per location and their number of containers were randomly assigned considering the total variability found for NH locations in the work of Peña-García et al. [24] (electronic supplementary material, Methods).

In conclusion, the results of this work suggest not only that NH locations are important in dengue transmission but also that vector control activities like those of container removal will be inefficient at reducing dengue burden if these spaces are not included. When exploring in detail the NH locations, differential risk depends on the number of people and the time they spend in these places, which can make some age groups particularly vulnerable to infection at given locations and certain locations critical for certain age groups. Equally, because cities differ in their abundance and distribution of container sizes, comprehensive vector control approaches that focus on multiple types of containers across HH and NH spaces are necessary to break chains of transmission. Through our model, we provide evidence-based insights for new directions aimed at designing new vector control strategies that can make limited resources be used in optimal control activities.

Ethics. This work specifically did not collect data from human samples or surveys. However, we did acquire information from other works obtaining such information. For those, ethical approval and oversight for data collection were obtained from the Institutional Review Board of Stanford University (IRB 31488), as well as the Kenya Medical Research Institutes (KEMRI SSC 2611) and Technical University of Mombasa Ethical Review Committee (TUM/ERC EXT/004/2019).

Data accessibility. Data and relevant code for this research work are stored in GitHub [56] and have been archived within the Zenodo repository [57].

Electronic supplementary material is available online [58].

Declaration of Al use. We have not used Al-assisted technologies in creating this article.

Authors' contributions. V.H.P.-G.: conceptualization, formal analysis, investigation, methodology, software, validation, visualization, writing-original draft; A.D.LaB.: conceptualization, funding acquisition, project administration, supervision, writing-review and editing; B.A.N.: data curation, investigation; F.M.M.: data curation, investigation; D.B.: data curation, investigation, writing-review and editing; J.R.A.: conceptualization, methodology, supervision, visualization, writing-review and editing; E.A.M.: conceptualization, methodology, supervision, visualization, writing -review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration. We declare we have no competing interests.

Funding. This research was funded by NIH through grant R01AI102918 (PI A.D.LaB.). In addition, V.H.P-G. is supported by grants R01AI102918 and R35GM133439; E.A.M. is supported by NIH grants R35GM133439, RO1AI102918, and RO1AI168097, and NSF grant DEB-2011147 (with Fogarty International Center); A.D.LaB. is supported by grants К01А1102918, R01AI149614, RO1AT155959, and D43TW011547.

Acknowledgements. We would like to acknowledge the teams that performed fieldwork and collected the data on which this work was based in both Kisumu and Ukunda cities: Joel Omaru Mbakaya, Samwuel Otieno Ndire, Gladys Adhiambo Agola, Paul 5. Mutuku, Said L. Malumbo, and Charles М. Ng'ang'a.

Sidebar

*VHP-G, 0000-0002-4856-3166; BAN, 0000-0002-8445-2885; FMM, 0000-0002-9000-7008; DB, 0000-0002-7832-2291; JRA, 0000-0002-5967-251X; EAM, 0000-0002-4402-5547

Cite this article: Peña-García VH, LaBeaud AD, Ndenga BA, Mutuku FM, Bisanzio D, Andrews JR, Mordecai EA. 2025 Non-household environments make a major contribution to dengue transmission: implications for vector control. В. Soc. Open Sci. 12: 241919. https://doi.org/10.1098/rsos.241919

Received: 4 November 2024

Accepted: 14 March 2025

Subject Category: Science, Society and Policy

Subject Areas: health and disease and epidemiology, computational biology, statistics

Author for correspondence: Victor Hugo Peña-Garcia e-mails: [email protected]; [email protected]

Electronic supplementary material is available online at https://doi.org/10.6084/ m9.figshare.c.7735452.

© 2025 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited.

References

References

1. World Health Organization. 2017 Global vector control response 2017-2030. Geneva, Switzerland: World Health Organization.

2. Scott TW, Amerasinghe PH, Morrison AC, Lorenz LH, Clark GG, Strickman D, Kittayapong P, Edman JD. 2000 Longitudinal studies of Aedes aegypti (Diptera: Culicidae) in Thailand and Puerto Rico: blood feeding frequency. J. Med. Entomol. 37, 89-101. (doi:10.1603/0022-2585-37.1.89)

3. Scott TW, Chow E, Strickman D, Kittayapong P, Wirtz RA, Lorenz LH, Edman JD. 1993 Blood-feeding patterns of Aedes aegypti (Diptera: Culicidae) collected in a rural Thai village. J. Med. Entomol. 30, 922-927. (doi:10.1093/jmedent/30.5.922)

4. Bhatt Setal. 2013 The global distribution and burden of dengue. Nature 496, 504-507. (doi:10.1038/nature12060)

5. GBD 2016 Disease and Injury Incidence and Prevalence Collaborators. 2017 Global, regional, and national incidence, prevalence, and years lived with disability for 328 diseases and injuries for 195 countries, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet 390, 1211-1259. (doi:10.1016/50140-6736(17)32154-2)

6. Yang X, Quam MBM, Zhang T, Sang $. 2021 Global burden for dengue and the evolving pattern in the past 30 years. J. Travel Med. 28, taab146. (doi:10.1093/jtm/taab146)

7. Bowman L, Donegan S, McCall P. 2016 Is dengue vector control deficient in effectiveness or evidence? Systematic review and meta-analysis. PLoS Neglected Trop. Dis. 10, e0004551. (doi:10.1371/journal.pntd.0004551)

8. Morrison AC, Getis A, Rigau-Perez JG, Santiago М, Reiter P. 1998 Exploratory space-time analysis of reported dengue cases during an outbreak in Florida, Puerto Rico, 1991-1992. Am. J. Trop. Med. Hyg. 58, 287-298. (doi:10.4269/ajtmh.1998.58.287)

9. Achee № Gould Е, Perkins T, Reiner В, Morrison A, Ritchie S. 2015 A critical assessment of vector control for dengue prevention. PLoS Neglected Trop. Dis. 9, 0003655. (doi: 10.1371/journal.pntd.0003655)

10. Lorenz C, Chiaravalloti-Neto Е. 2022 Control methods for Aedes aegypti: have we lost the battle? Travel Med. Infect. Dis. 49, 102428. (doi:10. 1016/j.tmaid.2022.102428)

11. Mulderij-Jansen V, Рипа P, Grillet ME, Lakiang T, Gerstenbluth |, Duits A, Tami A, Bailey A. 2022 Effectiveness of Aedes-borne infectious disease control in Latin America and the Caribbean region: a scoping review. PLoS ONE 17, 60277038. (doi:10.1371/journal.pone.0277038)

12. Morrison AC, Zielinski-Gutierrez E, Scott TW, Rosenberg В. 2008 Defining challenges and proposing solutions for control of the virus vector Aedes aegypti. PLoS Med. 5, e68. (doi:10.1371/journal.pmed.0050068)

13. Montenegro-Quifionez С, Louis V, Horstick O, Velayudhan В, Dambach P, Runge-Ranzinger S. 2023 Interventions against Aedes/dengue at the household level: a systematic review and meta-analysis. EBioMedicine 93, 104660. (doi:10.1016/j.ebiom.2023.104660)

14. World Health Organization. 2009 Dengue guidelines for diagnosis, treatment, prevention and control: new edition. Geneva, Switzerland: World Health Organization.

15. Pan American Health Organization. 2019 Manual for indoor residual spraying in urban areas for Aedes aegypti control. Washington, DC: PAHO.

16. World Health Organization. 2016 Vector control operations framework for zika virus. Geneva, Switzerland: World Health Organization.

17. Huang CH, Lin CY, Yang CY, Chan TC, Chiang PH, Chen YH. 2021 Relationship between the incidence of dengue virus transmission in traditional market and climatic conditions in Kaohsiung city. Can. J. Infect. Dis. Med. Microbiol. 2021, 9916642. (doi:10.1155/2021/9916642)

18. Suwanbamrung C, Promsupa S, Doungsin T, Tongjan S. 2013 Risk factors related to dengue infections in primary school students: exploring students' basic knowledge of dengue and examining the larval indices in southern Thailand. J. Infect. Public Health 6, 347-357. (doi:10.1016/j. jiph.2013.04.006)

19. Pérez-Pérez J, Peña-García VH, Calle-Tobón A, Quimbayo-Forero M, Rojo R, Henao E, Shragai T, Rúa-Uribe G. 2021 Entomovirological surveillance in schools: are they a source for arboviral diseases transmission? Int. J. Environ. Res. Public Health 18, 6137. (doi:10.3390/ijerph18116137)

20. Kampango A, Furu P, Sarath D, Haji K, Konradsen F, Schiøler K. 2021 Risk factors for occurrence and abundance of Aedes aegypti and Aedes bromeliae at hotel compounds in Zanzibar. Parasites Vectors 14, 544. (doi:10.1186/s13071-021-05005-9)

21. Barrera R, Acevedo V, Amador M. 2021 Role of abandoned and vacant houses on Aedes aegypti productivity. Am. J. Trop. Med. Hyg. 104, 145 150. (doi:10.4269/ajtmh.20-0829)

22. Wilder-Smith A. 2021 Dengue during the COVID-19 pandemic. J. Travel Med. 28, taab183. (doi:10.1093/jtm/taab183)

23. Sasmono RT, Santoso MS. 2022 Movement dynamics: reduced dengue cases during the COVID-19 pandemic. Lancet Infect. Dis. 22, 570-571. (doi:10.1016/s1473-3099(22)00062-7)

24. Peña-García VH et al. 2023 The importance of including non-household environments in dengue vector control activities. Viruses 15, 1550. (doi: 10.3390/v15071550)

25. Shah MM et al. 2020 High dengue burden and circulation of 4 virus serotypes among children with undifferentiated fever, Kenya, 2014-2017. Emerg. Infect. Dis. 26, 2638-2650. (doi:10.3201/eid2611.200960)

26. Khan A et al. 2023 Spatiotemporal overlapping of dengue, chikungunya, and malaria infections in children in Kenya. BMC Infect. Dis. 23, 183. (doi:10.1186/s12879-023-08157-4)

27. Kenya National Bureau of Statistics. 2019 Kenya population and housing census. Population by county and sub-county. Nairobi.

28. Kenya National Bureau of Statistics. 2021 Quarterly labour force report. Quarter 1. Nairobi.

29. Kenya Ministry of Education. 2019 Basic education statistical booklet. Nairobi.

30. Inziani M, Adungo F, Awando J, Kihoro R, Inoue S, Morita K, Obimbo E, Onyango F, Mwau M. 2020 Seroprevalence of yellow fever, dengue, West Nile and chikungunya viruses in children in Teso South Sub-County, Western Kenya. Int. J. Infect. Dis. 91, 104-110. (doi:10.1016/j.ijid.2019.11. 004)

31. Ordóñez-Gonzalez JG, Mercado-Hernandez R, Flores-Suarez AE, Fernández-Salas I. 2001 The use of sticky ovitraps to estimate dispersal of Aedes aegypti in northeastern Mexico. J. Am. Mosq. Control Assoc. 17, 93-97.

32. Qualls WA et al. 2016 Movement of Aedes aegypti following a sugar meal and its implication in the development of control strategies in Durán, Ecuador. J. Vector Ecol. 41, 224-231. (doi:10.1111/jvec.12217)

33. Edman JD, Scott TW, Costero A, Morrison AC, Harrington LC, Clark GG. 1998 Aedes aegypti (Diptera: Culicidae) movement influenced by availability of oviposition sites. J. Med. Entomol. 35, 578-583. (doi:10.1093/jmedent/35.4.578)

34. McCormack CP, Ghani AC, Ferguson NM. 2019 Fine-scale modelling finds that breeding site fragmentation can reduce mosquito population persistence. Commun. Biol. 2, 273. (doi:10.1038/s42003-019-0525-0)

35. Smith DL, Perkins TA, Tusting LS, Scott TW, Lindsay SW. 2013 Mosquito population regulation and larval source management in heterogeneous environments. PLoS ONE 8, e71247. (doi:10.1371/journal.pone.0071247)

36. Walker M, Chandrasegaran K, Vinauger C, Robert MA, Childs LM. 2021 Modeling the effects of Aedes aegypti's larval environment on adult body mass at emergence. PLoS Comput. Biol. 17, e1009102. (doi:10.1371/journal.pcbi.1009102)

37. Mordecai E, Cohen J, Evans M, Gudapati P, Johnson L, Lippi C. 2017 Detecting the impact of temperature on transmission of Zika, dengue, and chikungunya using mechanistic models. PLoS Neglected Trop. Dis. 11, e0005568. (doi:10.1371/journal.pntd.0005568)

38. Caldwell JM et al. 2021 Climate predicts geographic and temporal variation in mosquito-borne disease dynamics on two continents. Nat. Commun. 12, 1233. (doi:10.1038/s41467-021-21496-7)

39. Huber J, Childs M, Caldwell J, Mordecai E. 2018 Seasonal temperature variation influences climate suitability for dengue, chikungunya, and Zika transmission. PLoS Neglected Trop. Dis. 12, e0006451. (doi:10.1371/journal.pntd.0006451)

40. Sabin AB. 1950 The dengue group of viruses and its family relationships. Bacteriol. Rev. 14, 225-232.

41. Sabin AB. 1952 Research on dengue during World War II. Am. J. Trop. Med. Hyg. 1, 30-50. (doi:10.4269/ajtmh.1952.1.30)

42. Kiener M, Shayegh N, Nyathi SV, Ndenga BA, Mutuku FM, LaBeaud AD. 2024 Low rate of asymptomatic dengue infection detected in coastal Kenya using pooled polymerase chain reaction testing. Am. J. Trop. Med. Hyg. 110, 738-740. (doi:10.4269/ajtmh.23-0650)

43. Nyathi S et al. 2024 Molecular epidemiology and evolutionary characteristics of dengue virus 2 in East Africa. Nat. Commun. 15, 7832. (doi:10. 1038/s41467-024-51018-0)

44. Paz-Soldan V, Reiner R, Morrison A, Stoddard S, Kitron U, Scott T. 2014 Strengths and weaknesses of Global Positioning System (GPS) data loggers and semi-structured interviews for capturing fine-scale human mobility: findings from Iquitos, Peru. PLoS Neglected Trop. Dis. 8, e2888. (doi:10.1371/journal.pntd.0002888)

45. Stoddard S, Morrison A, Vazquez-Prokopec G, Paz Soldan V, Kochel T, Kitron U. 2009 The role of human movement in the transmission of vector borne pathogens. PLoS Neglected Trop. Dis. 3, e481. (doi:10.1371/journal.pntd.0000481)

46. Stoddard ST et al. 2013 House-to-house human movement drives dengue virus transmission. Proc. Natl Acad. Sci. USA 110, 994-999. (doi:10. 1073/pnas.1213349110)

47. Reiner RC, Stoddard ST, Scott TW. 2014 Socially structured human movement shapes dengue transmission despite the diffusive effect of mosquito dispersal. Epidemics 6, 30-36. (doi:10.1016/j.epidem.2013.12.003)

48. Zhang Y, Riera J, Ostrow K, Siddiqui S, de Silva H, Sarkar S, Fernando L, Gardner L. 2020 Modeling the relative role of human mobility, land-use and climate factors on dengue outbreak emergence in Sri Lanka. BMC Infect. Dis. 20, 649. (doi:10.1186/s12879-020-05369-w)

49. Prem K, Lau MSY, Tam CC, Ho MZJ, Ng LC, Cook AR. 2019 Inferring who-infected-whom-where in the 2016 Zika outbreak in Singapore-a spatio-temporal model. J. R. Soc. Interface 16, 20180604. (doi:10.1098/rsif.2018.0604)

50. Katzelnick LC, Gresh L, Halloran ME, Mercado JC, Kuan G, Gordon A, Balmaseda A, Harris E. 2017 Antibody-dependent enhancement of severe dengue disease in humans. Science 358, 929-932. (doi:10.1126/science.aan6836)

51. Horstick O, Runge-Ranzinger S. 2019 Multisectoral approaches for the control of vector-borne diseases, with particular emphasis on dengue and housing. Trans. R. Soc. Trop. Med. Hyg. 113, 823-828. (doi:10.1093/trstmh/trz020)

52. Harrington LC, Ponlawat A, Edman JD, Scott TW, Vermeylen F. 2008 Influence of container size, location, and time of day on oviposition patterns of the dengue vector, Aedes aegypti, in Thailand. Vector Borne Zoonotic Dis. 8, 415-423. (doi:10.1089/vbz.2007.0203)

53. Ouédraogo W, Toé K, Sombié A, Viana M, Bougouma C, Sanon A. 2022 Impact of physicochemical parameters of Aedes aegypti breeding habitats on mosquito productivity and the size of emerged adult mosquitoes in Ouagadougou city, Burkina Faso. Parasites Vectors 15, 478. (doi:10.1186/ s13071-022-05558-3)

54. Perkins TA et al. 2014 Theory and data for simulating fine-scale human movement in an urban environment. J. R. Soc. Interface 11, 20140642. (doi:10.1098/rsif.2014.0642)

55. Perkins TA, Paz-Soldan VA, Stoddard ST, Morrison AC, Forshey BM, Long KC. 2016 Calling in sick: impacts of fever on intra-urban human mobility. Proc. R. Soc. B 283, 1834. (doi:10.1098/rspb.2016.0390)

56. vhpenagarcia. 2024 ABM_dengue. GitHub. See https://github.com/vhpenagarcia/ABM_dengue.

57. vhpenagarcia. 2024 ABM_dengue: code for research paper 'Non-household environments make a major contribution to dengue transmission: implications for vector control'. Zenodo. (doi:10.5281/zenodo.14036270)

58. Peña-García VH, LaBeaud D, Ndenga BA, Mutuku FM, Bisanzio D, Andrews J et al. 2025 Supplementary Material from: Non-Household Environments Make a Major Contribution to Dengue Transmission: Implications for Vector Control. FigShare. (doi:10.6084/m9.figshare.c. 7735452)

Footnote

1These authors contributed equally to the study.