Predicting Lyme Disease: A One Health Approach.

Abstract

Lyme disease is the most common vector-borne disease in North America. Predicting Lyme disease incidence is a key component of public health preparedness. Previously, we demonstrated that the volume of data searches on Google Trends for terms related to Lyme disease, such as "Lyme" and "tick bite", can be used as a tool to predict monthly human Lyme disease incidence at the state level. The objective of this project was to build upon our previous work by adding environmental and canine data to our predictive models for the prediction of state-level human and canine Lyme disease incidence. Human data were acquired from state health departments. Canine data were acquired from IDEXX Laboratories. We hypothesized that incorporating a One Health approach with human, animal, and environmental data would improve the predictive ability of the models. The One Health model performed significantly better (Mean Absolute Error [MAE] = 12.1) in predicting human disease incidence in 6 out of 16 states compared to the environmental data model (MAE = 16.5), human search terms model (MAE = 21.4), canine data (search terms + case count) model (MAE = 31.1), and the canine case data model (MAE = 32.0). For canine Lyme disease incidence, the One Health model performed worse (MAE = 330.5) compared to the canine search data model (MAE = 282.3), the human data (search terms + cases) model (MAE = 248.4), and the environmental data (MAE = 221.5) model. However, even the best-performing models had large prediction errors, which limit practical utility. Future studies should incorporate alternative data streams, such as electronic health records and insurance claims, to test predictive ability.