Using Disease Surveillance Data to Forecast Disease Risk

• Activity Number: 160
• Type: Topic Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Al Ozonoff*+ and Yorghos Tripodis and John Brownstein
• Companies: Boston University School of Public Health and Boston University and Children's Hospital Boston
• Keywords: disease surveillance ; time series ; ARIMA ; forecasting ; outbreaks ; influenza

Abstract

Events such as extreme weather events and natural disasters such as floods, or earthquakes, occur with unpredictable timing and magnitude with obvious risks to human health. Similarly, epidemics or pandemics of infectious disease are unpredictable and can lead to morbidity and mortality on massive scales. To address this concern we propose a methodology, inspired by similar work in econometrics and the physical sciences, to forecast future risk of a disease outbreak. We first fit linear time series models to surveillance data, then implement a resampling procedure to produce synthetic time series with similar dynamic characteristics. We use the resamples to generate an empirical forecast of future disease events. In this paper, we will discuss implementation issues related to seasonality, model selection, and resampling; then discuss an application to influenza mortality data.