JMIR Publications

ABSTRACT

Background:

Traditional disease surveillance systems often rely on clinical or laboratory-confirmed data, which can delay detection of emerging outbreaks. The National Retail Data Monitor (NRDM), established in 2002, collects anonymized over-the-counter (OTC) health product sales data from more than 28,000 retail outlets across the United States and offers a complementary data source for syndromic surveillance.

Objective:

This study aimed to evaluate the use of NRDM data for detecting and tracking influenza-like illness (ILI), including data patterns that are suggestive of novel disease outbreaks.

Methods:

We analyzed NRDM data from Allegheny County, Pennsylvania, spanning 2016–2021. A probabilistic modeling framework was developed to estimate daily ILI activity from OTC purchasing patterns. An expectation–maximization algorithm was used to estimate conditional probabilities of product purchases by an individual given that they have an ILI disease. We also estimated similar probabilities given that they did not have such a disease. These estimates were applied to daily sales data to infer ILI-attributable purchases. We extended this framework into a monitoring system, termed NRDM Tracker, which uses daily log-likelihoods and false-discovery-rate–adjusted p-values to detect statistically significant anomalies in purchasing behavior.

Results:

NRDM-derived ILI estimates were moderately correlated with ED-based ILI counts (r = 0.66). Cough and cold medications showed a higher probability of purchase under ILI (0.792) compared with non-ILI conditions (0.695), while other product categories were more strongly associated with non-ILI use. Compared with ED visits (combined ILI and COVID-19 counts), NRDM-based ILI signals showed earlier increases during certain phases of outbreak emergence, particularly in early March 2020 in Allegheny County, when COVID-19 testing capacity was limited (resulting in low counts of confirmed cases) and public awareness and health-seeking behaviors were rapidly evolving. Moreover, NRDM Tracker detected significant anomalies in OTC purchasing behavior, including a marked anomaly in the log-likelihood in March 2020 corresponding to the onset of the COVID-19 pandemic. Multiple product categories exhibited statistically significant deviations in sales during major outbreak periods.

Conclusions:

OTC product sales data, analyzed using probabilistic modeling and anomaly detection methods, can provide timely insights into population-level health activity and support early detection of both seasonal and novel disease outbreaks. NRDM-based surveillance systems may serve as an effective complement to traditional clinical surveillance and enhance local and national biosurveillance capabilities.