Containing outbreaks of infectious disease requires rapid identification of transmission hotspots, as the COVID-19 pandemic demonstrates. Focusing limited public health resources on transmission hotspots can contain spread, thus reducing morbidity and mortality, but rapid data on community-level disease dynamics is often unavailable. Here, we demonstrate an approach to identify anomalously elevated levels of influenza-like illness (ILI) in real-time, at the scale of US counties. Leveraging data from a geospatial network of thermometers encompassing more than one million users across the US, we identify anomalies by generating accurate, county-specific forecasts of seasonal ILI from a point prior to a potential outbreak and comparing real-time data to these expectations. Anomalies are strongly correlated with COVID-19 case counts and may provide an early-warning system to locate outbreak epicenters.

Forecasting influenza primes public health systems to respond, reducing transmission, morbidity and mortality. Most influenza forecasts to date have, by necessity, relied on spatially course-grained data (e.g. state- or country-level incidence), and have operated at time horizons of 12 weeks or less. If influenza outbreaks could be predicted farther in advance and with increased spatial precision, then limited public health resources could be adaptively managed to minimize spread and improve health outcomes. Here, we use real-time syndromic data from a distributed network of thermometers to construct city-specific forecasts of influenza-like illness (ILI) with a horizon of 30 weeks. Daily geolocated ILI data from the network allows for estimates of recurrent city-specific patterns in ILI transmission rates. These transmission templates are used to parameterize an ensemble of ILI forecasts that differ randomly in three parameters, representing city- and season- specific rates of susceptible depletion and reporting, as well as differences in influenza season onset timing. For nine cities across the US, the best-in-hindsight model matches the observed data, and the best forecast variants can be identified in the early season.