Methods/Statistics
Optimizing Environmental Surveillance Rebecca Smith* Rebecca Smith Anwesha Chakravarti Bo Li
Environmental surveillance requires collection of samples meant to represent a generalized risk in a specific area. For example, mosquito traps are used to test for the presence of West Nile virus (WNV) in mosquito populations, playing a crucial role in monitoring risk and informing response. But how do you decide where take a sample for environmental surveillance? What makes a good sampling site?
We present a multi-step statistical approach for using longitudinal environmental sampling data to determine the value, or score, of a location in predicting an outcome of interest. This score is then used to understand what landscape, infrastructure, demographic and socioeconomic factors are associated with predictive ability. As a practical example, we apply this method to mosquito trapping in the Chicago metropolitan area and its suburbs and its ability to predict human cases of WNV. We find a minimum threshold for human population in the vicinity of the trap is necessary for overall prediction, with a weighting towards prioritizing sensitivity due to the low number of cases and severity of disease. However, different landscape factors become important for maximizing either sensitivity or specificity of the prediction, indicating that the optimal sampling location may vary based on the relative importance of each of these values.
This approach enables resource-limited environmental surveillance programs to identify better locations for their sample collection, which may help in reducing the number of samples needed while increasing their individual efficiency.