STOR-i Seminar: Nedialko Dimitrov
Nedialko Dimitrov, Naval Postgraduate School
Friday 17 January 2014, 1200-1300
A54, Postgraduate Statistics Centre Lecture Theatre
Goal-oriented Design of Influenza Surveillance
The CDC employs a suite of data streams to achieve the multi-faceted goals of influenza surveillance: influenza-like-illness surveillance network (ILINet), IISP, WHO Labs, NRVSS. The data streams in many of these systems were largely assembled out of convenience or intuitive first principles.
As a result, the main challenge is how to use the available data to achieve existing and emerging surveillance goals. Next generation data streams such as Google Flu Trends are now also available. It is unclear which data streams are best to incorporate: some are expensive, others provide noisy data, others yet are unreliable.
In this presentation, we discuss a systematic process of design for influenza surveillance. Instead of constraining surveillance by convenience sampling, our process defines and selects the best available data through a four step process:
- Formalize surveillance objectives
- Specify candidate data sources
- Simulate data where none exists
- Select the most informative data sources
We present an example of this process by constructing a multi-objective influenza surveillance network in Texas. We also discuss ongoing efforts to introduce statistical guarantees into the national influenza surveillance system.