The 131st Annual Meeting (November 15-19, 2003) of APHA |
Elaine N. Florio, Steven F. Magruder, PhD, Sheryl L. Happel Lewis, MPH, and Richard A. Wojcik, MS. National Security Technology Department, Johns Hopkins University Applied Physics Laboratory, 11100 Johns Hopkins Road, Laurel, MD 20723, 240-228-1458, elaine.florio@jhuapl.edu
By monitoring point-of-sale counts of over-the-counter medication purchases, we have an opportunity for early indications and warning of disease. We conducted a detailed study after observing anecdotal evidence that the counts of units sold can have a strong association with factors other than illness. The ultimate goal was to use this knowledge to inform our modeling, have a better understanding of community illness and to better execute detection algorithms. The key findings of the analysis were that using information about season, weather, day of week and promotional offers were all significant in prediction error reduction, which can contribute to more optimal detector performance. Not only the existence, but the degree, of a promotional offer can drive sharp spikes in the number of units sold; this effect can be handled on-the-fly when electronic data on promotional offers are made available. Making brand-specific vs brand-independent distinctions may also be an important approach to monitoring. Day of week, particularly early in the week, is also a significant predictor of sales activity. Surprisingly, same-day outdoor temperature had a strong association with the number of units sold, suggesting a behavioral connection. Finally, models using the above factors were most accurate predictors when including seasonal effects. We also obtained some unexpected findings related to possible nontraditional uses of medications and sharp outdoor temperature anomalies.
Learning Objectives:
Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.