Variables for habitat and vertebrate hosts of Ixodes scapularis are the best ecological predictors of the spatial spread of Lyme disease in the United States (2010-2019).

BACKGROUND

Lyme disease (LD) is a major public health concern in North America. The incidence of LD has increased in part due to the rapid expansion of Ixodes scapularis infected with Borrelia burgdorferi sensu lato (Bb), the causative agent of LD. Understanding how environmental factors contribute to the spread of LD in humans remains a major challenge.

METHODS

We aimed to measure the environmental associations and spatial dynamics of LD incidence across United States counties between 2010 and 2019 via a machine-learning (ML)-based model. We used LD incidence data in 1322 counties in 24 US states from the Centers for Disease Control and Prevention (CDC), categorized by four incidence classes (0-1, > 1 to 10, > 10 to 100, > 100 cases/100,000 population). Explanatory variables of climate, habitat, land cover, I. scapularis presence, and distribution of tick hosts and Bb reservoirs were used to train the ML models.

RESULTS

The performance of a random forest algorithm was high (area under the curve [AUC] = 0.89). As expected, surveillance-dependent variables for adjacency to LD-endemic counties (gain ratio: 0.22) and presence of I. scapularis (gain ratio: 0.133) were identified as the top individual predictors of LD spread. However, the strongest overall contributions to the model were driven by vertebrate-related variables (n = 8) (ReliefF: 0.237), with landscape features for forest growth, canopy, and forest edge (length) also identified as strong (gain ratio > 0.28) individual predictors. Climate predictors indicated the lowest LD incidence classes (< 10 cases/100,000 population) in warmer and drier counties and the highest LD incidence classes (> 10 cases/100,000 population) in the coldest and wettest counties.

CONCLUSIONS

Utilization of ML algorithms trained with variables impacting the circulation of Bb produced a comprehensive model of county-level LD incidence and captured the main factors acting on the spread of the pathogen. This represents an important step towards an integrated framework aimed at capturing LD incidence changes for preventive purposes.