Spatio-temporal analysis of the main dengue vector populations in Singapore

Abstract

Background: Despite the licensure of the world's first dengue vaccine and the current development of additional vaccine candidates, successful Aedes control remains critical to the reduction of dengue virus transmission. To date, there is still limited literature that attempts to explain the spatio-temporal population dynamics of Aedes mosquitoes within a single city, which hinders the development of more effective citywide vector control strategies. Narrowing this knowledge gap requires consistent and longitudinal measurement of Aedes abundance across the city as well as examination of relationships between variables on a much finer scale.

Methods: We utilized a high-resolution longitudinal dataset generated from Singapore's islandwide Gravitrap surveillance system over a 2-year period and built a Bayesian hierarchical model to explain the spatio-temporal dynamics of Aedes aegypti and Aedes albopictus in relation to a wide range of environmental and anthropogenic variables. We also created a baseline during our model assessment to serve as a benchmark to be compared with the model's out-of-sample prediction/forecast accuracy as measured by the mean absolute error.

Results: For both Aedes species, building age and nearby managed vegetation cover were found to have a significant positive association with the mean mosquito abundance, with the former being the strongest predictor. We also observed substantial evidence of a nonlinear effect of weekly maximum temperature on the Aedes abundance. Our models generally yielded modest but statistically significant reductions in the out-of-sample prediction/forecast error relative to the baseline.

Conclusions: Our findings suggest that public residential estates with older buildings and more nearby managed vegetation should be prioritized for vector control inspections and community advocacy to reduce the abundance of Aedes mosquitoes and the risk of dengue transmission.

Keywords: Aedes; Dengue; Spatio-temporal modeling; Vector control.

Fig. 1

Observed Aedes abundance (mean catch per trap per week) of each site during 2017–2018: (a) Ae. aegypti and (b) Ae. albopictus. We first created a 300 m buffer around each block, and all buffers for each site were merged into a single polygon, which was then colored according to the observed Aedes abundance value

Fig. 2

Estimated percentage change in the expected value of Aedes abundance (weekly mean catch per trap) due to a 1-SD increase in each covariate when all the other variables were held constant. Filled circles denote the posterior median estimates, and the solid lines denote the 95% credible intervals. The standard deviation of each covariate can be found in Table 1, and the estimated effects of lagged temperature covariates on the predicted Aedes abundance were visualized in Fig. 3

Fig. 3

Predicted values of Aedes abundance (weekly mean catch per trap) at different values of lagged weekly maximum temperature with all the other variables held fixed at their average values. Solid lines denote the posterior median estimates and the shaded areas denote the 95% prediction bands

Fig. 4

Box-plots of the observed site-level Aedes abundance (mean catch per trap per week) during 2017–2018 within each quintile based on the out-of-sample model predictions: (a), (b) leave-one-site-out cross validation and (c), (d) leave-one-planning-area-out cross validation. A small number of extreme values were omitted from the graph for clarity