Monitoring sales of medications is a potential candidate for an early signal of a seasonal influenza epidemic. To test this theory, the data from a traditional, consultation-oriented influenza surveillance system were compared to medication sales and a predictive model was developed. Weekly influenza-like incidence rates from the National Influenza Sentinel Surveillance System were compared to sales of seven groups of medications (nasal decongestants, medicines for sore throat (MST), antitussives, mucolytics, analgo-antipyretics, non-steroidal anti-inflamatory drugs (NSAIDs), betalactam antibiotics, and macrolide antibiotics) to determine the correlation of medication sales with the sentinel surveillance system – and therefore their predictive power. Poisson regression and regression tree approaches were used in the statistical analyses. The fact that NSAIDs do not exhibit any seasonality and that prescription of antibiotics requires a visit to the doctor's office makes the two medication groups inappropriate for predictive purposes. The influenza-like illness (ILI) curve is the best matched by the mucolytics and antitussives sales curves. Distinct seasonality is also observed with MST and decongestants. The model including these four medication groups performed best in prediction of ILI incidence rate using the Poisson regression model. Sales of antitussives proved to be the best single predictive variable for regression tree model. Sales of medication groups included in the model were demonstrated to have a predictive potential for early detection of influenza season. The quantitative information on medication sales proves to be a useful supplementary system, complementing the traditional consultation-oriented surveillance system.
