Large Scale Multivariate CAR Modeling

• Activity Number: 46 - Advances in Spatio-Temporal Epidemiology
• Type: Invited
• Date/Time: Sunday, July 30, 2017 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Robert Lund* and Christopher McMahan and Yan Liu
• Companies: Clemson University and Clemson University and Clemson University
• Keywords: CAR Model ; Lyme Disease ; Heartworm ; Space-time Modeling ; Autoregression ; Multivariate

Abstract

This talk examines conditional autoregressive (CAR) space-time modeling in large-scale problems. The methods are needed to describe and forecast the Companion Animal Parasite Council's test bank data, which contains millions of annual heartworm and Lyme disease test results for United States dogs from 2011-current in the 48 contiguous United States. Our work entails fitting a Poisson regression with space-time correlations to the 3000+ counties in the contiguous United States, a big data setting. After devising methods for this situation, we show how the model and several auxiliary covariates can be used to forecast next year's disease prevalence, an aid to the veterinarian. As Lyme is also a debilitating human disease --- and human Lyme test results are typically private --- we jointly examine Lyme disease in humans and dogs, hoping that the dog forecast would also work well for humans.

This is joint work with Chris McMahan and Yan Liu at Clemson University.