High-Resolution Vector-Borne Disease Risk Assessment

Goal

We wish to identify the risk of vector-borne diseases at the resolution of a neighborhood. This will enable effective and targeted public health interventions, such as integrated vector control measures.

Challenges

Human mobility plays a critical role in the spread of vector-borne diseases. Conventional data sources do not have the required spatial and temporal resolution. Furthermore, traditional models have difficulty accounting for demographic heterogeneity.

Approach

We use social media data to model human mobility. In particular, we use Twitter data to identify locations visited by people who have recently been to a disease-affected region.

  1. We first compute the flux of such persons into different counties using Natural Language Processing techniques on tweet content.
  2. We input this population flux into a meta-population model to identify counties at risk of disease importation.
  3. We then target high-risk counties for a fine-scale model. We leverage our novel deep learning workflow on Twitter metadata for determining home-locations of users to identify high-risk neighborhoods.
Flowchart of our approach
Identify home location of a user (red)

Results

Zika risk map at the county level (2016 outbreak)
Predicted high-risk neighborhoods Actual high-risk neighborhoods
Miami Beach Miami Beach
Wynwood Wynwood
Miami Airport Little River
Neighborhood level Zika risk prediction for Miami (2016 outbreak)

References

  1. High-resolution Twitter home location prediction
  2. Detailed version of ASONAM 2019 paper on Zika risk assessment