Assessing the Utility of Over-The-Counter Medication to Track Influenza Season Timing in a South African Population

Laboratory-based influenza surveillance in low- and middle-income countries may have limited timeliness and geographic coverage. We evaluated whether over-the-counter (OTC) respiratory medication sales track seasonal influenza activity and provide early warning of influenza onset in South Africa.

Daily OTC medication sales were obtained from a national pharmaceutical chain and analysed nationally and in Gauteng, KwaZulu-Natal and Western Cape. Weekly laboratory-confirmed influenza cases were obtained from sentinel surveillance programmes for 2012-2019 and 2022-2023. Seasonal characteristics were compared using the Moving Epidemic Method. Lead-lag relationships were assessed using Pearson cross-correlation, with 3-week smoothing and first differencing as sensitivity analyses. Forecasts at 1- to 5-week horizons were evaluated using influenza-only, OTC-only, combined influenza-plus-OTC generalized linear model (GLM), Random Forest and naïve persistence models. Performance was assessed using root mean square error and area under the receiver operating characteristic curve (AUC).

OTC sales preceded increases in influenza detections nationally and provincially. Sales-defined seasons began and peaked earlier in 70% of national seasons and 67%-89% across provinces. Primary analyses indicated that OTC sales preceded influenza detections by 2-3 weeks (mean peak correlations, 0.77-0.81). After first differencing, OTC changes preceded influenza changes in 5-8 of 10 seasons across locations (correlations, 0.42-0.50). Influenza-only models produced the lowest forecast errors. OTC-informed models achieved AUC ≥ 0.90 at 1- to 2-week horizons. All significant squared-error comparisons favoured naïve persistence.

OTC respiratory medication sales may complement virological surveillance by supporting earlier identification of seasonal influenza activity but should not replace laboratory surveillance.