The burden of arboviruses transmitted by Ae. aegypti in Guayaquil has increased significantly since the 1980s, and targeted interventions are necessary to halt the spread of such diseases [3]. The findings of this study provide evidence that Ae. aegypti proliferation is influenced by specific household risk factors. The household models that performed best as determined by AICc contained the following variables: canopy use, large solid collection services, unemployment, water volume, water interruptions, biolarvicide, precipitation at week 0, precipitation at week 2 lag, and the interactions precipitation at week 0 * water interruption and large solid collection services * precipitation at a week 2 lag.
Canopy use was found to have a significant positive association with Ae. aegypti abundance, which is counterintuitive and may be attributed to other vector control practices being limited when bed canopies are in use. It is likely that those with higher mosquito populations inside the house are more prone to using bed canopies. The usage of these canopies may result in a false sense of security, or may be an indication of other unmeasured household risk factors that allow for high Ae. aegypti densities. The previously mentioned study from Machala, Ecuador, also cites an unclear relationship between dengue infections in a city near Guayaquil and bed canopy usage [2]. Furthermore, Ae. aegypti are daytime feeders, so these nets would only affect mosquito feeding if household members are napping during the day.
In our analysis, large solid collection services had a significant negative association with pupal abundance, which may be because there remain fewer untouched breeding sites available for Ae. aegypti. Regular bulky item pickup removes tires and other potential mosquito habitats where water could pool. This is corroborated by previous studies which have found that trash and flower pots are among the most common Ae. aegypti-positive containers [13]. These containers may be more regularly eliminated and maintained in higher socioeconomic areas through large solid collection services. In alignment with this finding, it was also found that unemployment had a significant positive relationship with pupal abundance. This corresponds with the comparison of the socioeconomic status map from Fig. 1 and the pupal index map from Fig. 2, where we find that there is a higher density of Ae. aegypti-positive households in lower socioeconomic areas. There is also a higher prediction error for these areas, as seen in Fig. 2 and Additional file 7: Fig. S3, suggesting that areas with higher unemployment need an emphasis on research to better understand the specific household risk factors attributed to Ae. aegypti pupal density (Additional file 6: Fig. S2).
Our study also found that average water volume had a significant positive relationship with pupal abundance. A 2012 study from the Tri Nguyen village in Vietnam found that containers where the water volume increased relative to the previous survey had a significantly higher count of Ae. aegypti pupae [14]. The study also found that the greatest increase in pupal abundance occurred after a rainfall event. This corresponds to our study’s findings in which both precipitation during week 0 and increasing water volume resulted in higher APC. Heavy rainfall is known to flush out existing containers, which could explain the negative (not statistically significant) association between rainfall during the week of sampling and APC. A negative association (although again not statistically significant) between rainfall with a 2-week lag and APC could conceivably result from the same flushing phenomenon, adjusting for the 8–12 days for Ae. aegypti eggs to develop into pupae [15, 16]. The interaction between large solid collection services and precipitation 2 weeks prior was significantly positively associated with Ae. aegypti pupal counts, which could be due to rainfall providing habitat for eggs to be laid which then develop into pupae 8–12 days later. The meaning of the interaction is somewhat unclear; however, wealthier neighborhoods had increased access to large solid collection services, so there was greater creation and destruction of mosquito habitat compared with poorer neighborhoods. Figure 2 shows that wealthier areas of Guayaquil have a lower number of high-density Ae. aegypti households. With fewer breeding habitats in wealthier neighborhoods, precipitation may have a larger and differential effect on pupal density, and therefore, any marginal effect may be picked up by the model.
This differentiation is further explained in the water storage practices and distribution of houses. In neighborhoods with higher employment rates and lower illiteracy, there are an increased number of natural areas where precipitation may collect, especially since houses are spread further apart. In less developed neighborhoods, there are different relationships with standing containers. During the rainy season, there are not as many water-holding containers because the water is constantly replenished by the rain. However, in the dry season, there are more standing containers because water is more scarce and needs to be stored for the households.
When the interaction term is included in the final model, precipitation at a week 2 lag has a negative correlation with the outcome. When the interaction term is not being controlled for, 14 days after a precipitation event correlates with higher pupal density. This suggests that there is a specific relationship between large solid collection services and precipitation at a week 2 lag on our outcome, pupal density. However, since large solid services may serve as a proxy for socioeconomic status, this may suggest a dynamic effect across socioeconomic statuses. These nuances are difficult to account for within the model context. For vector control efforts to be effective, it may require a more thorough understanding of the relationship between rainfall and socioeconomic factors that influence pupal density.
Aedes aegypti are highly adaptable mosquitoes that were historically found in forested areas using tree holes for breeding but have since adapted to breeding in tires, vases, and other objects found in proximity to human habitations [17]. Their resilience and adaptability pose difficulties when searching for effective control methods, especially for outdoor areas [17]. However, in light of our analyses, certain types and materials of containers may be more or less productive than others.
In our study, vase-type containers were found to be a significant predictor and were correlated with lower pupae counts. Glass material composition was selected as an appropriate explanatory predictor and was correlated with lower pupae counts as well. Vases, other glass-type containers, metal material, and ceramic material containers may have more variable water temperature that impedes Ae. aegypti development. Contaminated water was found to be a significant predictor correlated with higher pupae counts. The survey had field technicians qualitatively assess whether water was contaminated, so it was not quantifiably measured. Research suggests that Ae. aegypti prefer “clean” water, but this is a relative designation, as some nutrients in the water may support mosquito populations [7]. Contaminated water may have organic components within the container that promote algal growth and support mosquito proliferation. Water that is contaminated is likely to be untouched and stagnant, allowing Ae. aegypti to lay eggs and develop, as opposed to cleaner water, which may be flushed more often [7]. This finding is corroborated by previous studies that have noted that poor sanitation and water storing habits provide viable habitats for Ae. aegypti [3].
These results suggest that trash collection services targeting large solids and monitoring of containers that could serve as juvenile mosquito habitat contribute to suppressing Ae. aegypti pupal proliferation and consequent adult mosquito densities. These predictive models provide household factors of interest that could be included in future surveys to test hypotheses or assessed in rigorous causal models.
For the top household model, the mean error was high (10 pupae off of the true value) relative to the mean pupal index (11.0836). However, the standard deviation of the data is 10.263 indicating that the high error is due to the relatively high variance of the data, and the maximum pupal count is 253. Overdispersion and high variance are common in insect count data; therefore, these results remain valid [18].
There were months without sampling in each of the years for each of the three parts of Guayaquil; however, they did not share the same months missing in each area, so it was not possible to address this through a time-series analysis to account for the repeated measurements on households. Predictive modeling has limitations. Best subset selection assesses 2p models, where p indicates the number of parameters, making the implementation of every interaction computationally infeasible when the number of parameters is large. Using previous literature, we assessed the most pertinent interactions and limited our model variable subset selection to 221 models (Additional file 5: Fig. S1). This study could be improved with the inclusion of zero APC households to differentiate between containers and households that have zero mosquito pupae compared with those that have positive counts. Additionally, a longitudinal study, as opposed to the cross-sectional study design here, could track temporal dynamics in pupae populations. With a longitudinal study, a time-series analysis would be able to assess changing exposures to vector control methods and the environment and any subsequent changes in mosquito populations.
Future studies could correlate pupae counts with household demographics such as age and sex of inhabitants. Noting behavioral differences across these characteristics could also inform efforts for reducing mosquito proliferation and arbovirus spread. Additionally, further studies should compare our estimates of household factors in Guayaquil to those in more rural settings. Household risk factors such as water service interruption and temephos use may have a larger impact in more rural areas, where water interruptions may be more frequent. A similar study placed on an urban-to-rural gradient may help capture these effects. Additionally, dengue serological data could be incorporated to assess correlation between household risk factors and past exposure to dengue, which would be closer to the health endpoint and valuable for the public health sector. Random-effects modeling may further assess our covariates and outcomes with contextual understanding of variable distributions between and within households. Lastly, an understanding of competing dynamics between Ae. aegypti and other species of mosquitoes for habitat, breeding, and feeding would provide further context for targeted interventions in areas where multiple species coexist.
Furthermore, in this study, we used average pupae per container as our outcome measure. In this research field, similar studies have used other measures such as the Breteau index and house and container indices [1]. For this study, house and container indices were not used because the data set does not contain records of negative containers that did not contain Ae. aegypti larvae necessary to compute such indices. The Breteau index was not appropriate for this study because we explored individual household-level characteristics; however, this could be applied to a neighborhood-level study in the future.