Detecting anomalous patterns in pharmacy retail data

Bio-surveillance systems have recently gained a lot of attention and are growing more and more complex. Multiple sources of data (pharmacy sales, emergency department visits, weather indicators, census information, etc.) are now available, and these sources can be used to identify both natural disease outbreaks (e.g. influenza) and outbreaks resulting from bio-terrorist attacks (e.g. anthrax release). The bio-surveillance research community is actively developing intelligent algorithms to detect outbreaks in a timely manner, in order to save lives and costs. However, though many of these algorithms show impressive results under simulated environments, their performance tends to degrade when applied to real-world datasets. Seasonal and day-of-week trends, missing data, lack of known disease outbreaks, difficulties in designing test beds, and high costs associated with processing false positives are some of the many reasons that hinder development of a successful practical bio-surveillance system. We believe that incorporating expert knowledge from public health officials will provide valuable insight to this complex process of disease outbreak detection. An immediate goal is to provide a tool that not only shows the alarms to the expert users, but also allows them to provide feedback on the alarms. This feedback loop is essential for iterative refinement of outbreak detection tools. This paper highlights our experiences with developing such a biosurveillance system that currently monitors national level pharmacy sales of over-the-counter (OTC) drugs on a daily basis.