The earliest possible detection from surveillance of human health is detection on the day that someone becomes symptomatic (ignoring tests capable of detecting incubating infections). Such detection requires self reporting, which could be obtained through frequent polling or methods to train individuals to report on the day they wake up ill.
Li and Aggarwal (2003) (see also Wagner et al., 2004) studied the responses of employees of the IBM T.J. Watson Research Center to a weekly e-mail asking about their health. Of approximately 2200 employees, 397 employees agreed to participate in the project. During the four-month period—January 25,2002 through May 31, 2002—each participant was sent e-mail once a week at 7 am on a randomly chosen work day. Participants were asked to self-assess their health for the day the e-mail was sent, choosing from one of the following categories: As usual, worse than usual (sick), or much worse than usual (very sick). If the participant reported being sick or very sick, there were a series of additional questions about the illness, including a question about symptoms (fever, headache, muscle ache, fatigue, cough, stuffy/runny nose, etc.). During the course of the survey, there was minimal dropout (<2%), and responses were prompt: 73% by noon and 92% on the same day as the e-mailing.
The researchers studied the informational value of polling data using correlation analysis with a reference time series that reflected disease activity in Westchester County. They constructed a reference time series from insurance claims data for patients seen in physician's office for respiratory complaints (ICD-9 codes 460-519). The correlation between the self-reporting obtained through polling and the physician office visits for respiratory illness was high, with the maximum correlation of 0.82 occurring with polling responses leading office visits by three days.
Although this result is promising, there are a number of factors that may limit the routine application of polling. Addressing privacy concerns will be critical in encouraging participation and obtaining reliable results. Liability issues must be fully understood. The responding process must be simple and non-intrusive enough to discourage dropouts. It may be that polling is most appropriate during situations of heightened alert or in permissive environments (see Section 5 in this chapter).
Was this article helpful?