Fig 1.
Schematic overview of the individual-based model.
Table 1.
List of parameter values for the simulation model.
Fig 2.
Simulated monthly number of dengue infections and dengue seroprevalence following a permanent reduction in the mosquito population size by 70% in different dengue endemic settings.
Unless otherwise stated, model parameters were fixed at the following values: π = 4×10−5, ω = 0.005, ρ = 0, α = 0.25, ϕADE = 0.5, lC = 180, pRM = 0.7, σRM = 0 (Here we used σRM = 0 to denote spatially uniform vector control intensity).
Fig 3.
Simulated monthly number of dengue infections during the 30-year intervention period in different dengue endemic settings.
Each figure panel showed model outputs obtained from two simulation runs with the same starting condition, where one simulation was performed without vector control and the other with the vector population size permanently reduced by 70% at the start of the intervention period (identical to the corresponding model output shown in Fig 2).
Table 2.
Number of simulations where the 5-year intervention effectiveness remained positive throughout the entire 30-year intervention period.
For each overall level of vector control intensity, a total of 1000 simulations were performed.
Fig 4.
Impact of different predictor variables on the critical time point.
For (A), the original values of the critical time point obtained from the simulations were summarized. Each box was drawn from the first quartile (Q1) to the third quartile (Q3) of the data, with the horizontal line in the middle showing the median. The upper whisker boundary was the largest observation that was within Q3 + 1.5 · (Q3—Q1), and the lower whisker boundary was the smallest observation above Q1–1.5 · (Q3—Q1). Data points beyond the whisker boundaries were omitted from the graph for clarity. For (B)—(F), we used SHAP values to quantify the change in the predicted critical time point due to the value of each predictor variable, where each point represents a single simulation run and each curve was obtained by fitting a smoothing spline to the set of data points corresponding to a given overall level of vector control intensity. We excluded simulations where the overall level of vector control intensity was 0.8 or higher, due to the high probability that vector control remained effective throughout the 30-year intervention period.
Fig 5.
Impact of different predictor variables on the vector control effectiveness across the entire 30-year intervention period.
For (A), the original values of the intervention effectiveness obtained from the simulation runs were summarized. Each box was drawn from the first quartile (Q1) to the third quartile (Q3) of the data, with the horizontal line in the middle showing the median. The upper whisker boundary was the largest observation that was within Q3 + 1.5 · (Q3—Q1), and the lower whisker boundary was the smallest observation above Q1–1.5 · (Q3—Q1). Data points beyond the whisker boundaries were omitted from the graph for clarity. For (B)—(F), we used SHAP values to quantify the change in the predicted overall intervention effectiveness due to the value of each predictor variable, where each point represents a single simulation run and each curve was obtained by fitting a smoothing spline to the set of data points corresponding to a given overall level of vector control intensity.
Fig 6.
Distribution of the simulated intervention effectiveness under different overall levels of vector control intensity within each of the following time windows: (A) (0y, 5y], (B) (5y, 10y], (C) (10y, 15y], (D) (15y, 20y], (E) (20y, 25y], (F) (25y, 30y]. Each box was drawn from the first quartile (Q1) to the third quartile (Q3) of the data, with the horizontal line in the middle showing the median. The upper whisker boundary was the largest observation that was within Q3 + 1.5 · (Q3—Q1), and the lower whisker boundary was the smallest observation above Q1–1.5 · (Q3—Q1). Data points beyond the whisker boundaries were omitted from the graph for clarity.
Fig 7.
Impact of baseline PE9 on the predicted vector control effectiveness within each of the following time windows: (A) (0y, 5y], (B) (5y, 10y], (C) (10y, 15y], (D) (15y, 20y], (E) (20y, 25y], (F) (25y, 30y]. The baseline PE9 referred to the proportion of nine-year olds who had prior dengue exposure at the end of the warm-up period. Each point represents a single simulation run, and the SHAP value quantified the change in the predicted vector control effectiveness due to the value of baseline PE9. Each curve was obtained by fitting a smoothing spline to the set of data points corresponding to a given overall level of vector control intensity.
Fig 8.
Impact of individual-level heterogeneity in the biting risk score on the predicted vector control effectiveness within each of the following time windows: (A) (0y, 5y], (B) (5y, 10y], (C) (10y, 15y], (D) (15y, 20y], (E) (20y, 25y], (F) (25y, 30y]. The parameter referred to the standard deviation of the biting risk score controlling for body surface area. Each point represents a single simulation run, and the SHAP value quantified the change in the predicted vector control effectiveness due to the value of
. Each curve was obtained by fitting a smoothing spline to the set of data points corresponding to a given overall level of vector control intensity.