Background Timely estimates of influenza activity are important for clinical and public health practice. However, traditional surveillance sources may be associated with reporting delays. Smartphone-connected thermometers can capture real-time illness symptoms, and these geo-located readings may help improve state-level forecast accuracy. Methods Temperature recordings were collected from smart thermometers and an associated mobile phone application. Using temperature recordings, we developed forecasting models of real-time state-reported influenza-like illness (ILI) 2 weeks before the availability of published reports. We compared time-series models that incorporated thermometer readings at various levels of spatial aggregation and evaluated out-of-sample model performance in an adaptive manner comparing each model to baseline models without thermometer information. Results More than 12 million temperature readings were recorded from over 500,000 devices from August 30, 2015 to April 15, 2018. Readings were voluntarily reported from anonymous device users, with potentially multiple users for a single device. We developed forecasting models of real-time outpatient ILI for 46 states with sufficient state-reported ILI data. Forecast accuracy improved considerably when information from thermometers was incorporated. On average, thermometer readings reduced the squared error of state-level forecasting by 43% during influenza season and more than 50% in many states. In general, best-performing models tended to result from incorporating thermometer information at multiple levels of spatial aggregation. Conclusion Local forecasts of current influenza activity, measured by outpatient ILI, can be improved by incorporating real-time information from mobile-devices. Information aggregated across neighboring states, regions, and the nation can lead to more reliable forecasts, benefiting local surveillance efforts.

Kinsa Inc. sells Food and Drug Administration–cleared smart thermometers, which synchronize with a mobile application, and may aid influenza forecasting efforts. We compare smart thermometer and mobile application data to regional influenza and influenza-like illness surveillance data from the California Department of Public Health. We evaluated the correlation between the regional California surveillance data and smart thermometer data, tested the hypothesis that smart thermometer readings and symptom reports provide regionally specific predictions, and determined whether smart thermometer and mobile application improved disease forecasts. Smart thermometer readings are highly correlated with regional surveillance data, are more predictive of surveillance data for their own region and season than for other times and places, and improve predictions of influenza, but not predictions of influenza-like illness. These results are consistent with the hypothesis that smart thermometer readings and symptom reports reflect underlying disease transmission in California. Data from such cloud-based devices could supplement syndromic influenza surveillance data.

BackgroundInformation regarding influenza activity can inform clinical and public health activities. However, current surveillance approaches induce a delay in influenza activity reports (typically 1–2 weeks). Recently, we used data from smartphone connected thermometers to accurately forecast real-time influenza activity at a national level. Because thermometer readings can be geo-located, we used state-level thermometer data to determine whether these data can improve state-level surveillance estimates.MethodsWe used temperature readings collected by the Kinsa smart-thermometer and mobile device app to develop state-level forecasting models to predict real-time influenza activity (1–2 weeks in advance of surveillance reports). We used state-reported influenza-like illness (ILI) to represent state influenza activity for 48 US states with sufficient surveillance data. Counts of temperature readings, fever episodes and reported symptoms were computed by week. We developed autoregressive time-series models and evaluated model performance in an adaptive out-of-sample manner. We compared baseline time-series models containing lagged state-reported ILI activity to models incorporating exogenous thermometer readings.ResultsA total of 10,262,212 temperature readings were recorded from October 30, 2015 to March 29, 2018. In nearly all of the 48 states considered, weekly forecasts of ILI activity improved considerably when thermometer readings were incorporated. On average, state-level forecasting accuracy improved by 23.9% compared with baseline time-series models. In many states, such as PA, New Mexico, MA, Virginia, New York and SC, out-of-sample forecast error was reduced by more than 50% when thermometer data were incorporated. In general, forecasts were most accurate in states with the greatest number of device readings. During the 2017–2018 influenza season, the average improvement in forecast accuracy was 24.4%, and thermometer readings improved forecasting accuracy in 41, out of 48, states.ConclusionData from smart thermometers accurately track real-time influenza activity at a state level. Local surveillance efforts may be improved by incorporating such information. Such data may also be useful for longer-term local forecasts.Disclosures I. Singh, Kinsa Inc.: Board Member, Employee and Shareholder, equity received and Salary. S. Pilewski, Kinsa Inc.: Employee and Shareholder, equity received and Salary. V. Petrovic, Kinsa Inc.: Employee and Shareholder, equity received and Salary.