A major challenge in responding to the current COVID-19 epidemic is the lack of early warning systems and limited number of testing kits. This makes it hard to know when and where outbreaks are happening, which in turn creates challenges for those in power to know where to distribute the necessary medical support and supplies. Current reports of positive COVID-19 cases from the Centers for Disease Control (CDC) rely on communication from healthcare facilities which results in identifying outbreaks days after it has already occurred. For example, COVID-19 tests are administered several days after patients first experience symptoms and some test results take a few days to process.
Kinsa has created an early warning system from their network of connected thermometers that helps public health officials identify where outbreaks of influenza-like illness (ILI) are occurring. This system has shown it can predict and alert communities of an ILI outbreak far earlier than the CDC, and recently, Kinsa’s detection of atypical illness has shown a strong correlation to outbreaks of COVID-19.
Kinsa’s atypical illness signal is the combination of two sets of data: Kinsa’s flu forecast, and our real-time illness data. In collaboration with Benjamin Dalziel, Associate Professor at Oregon State University, our data team has shown that they can forecast and predict the flu season in the United States up to 12 weeks out, and in some cases even farther into the future. In counties where there are enough active thermometers, these forecasts can be made down to the county level. The real-time illness data captures details such as fever and other symptoms which are submitted via the Kinsa thermometer and the app. This data is then aggregated to the county, state and national level, then normalized to match the CDC’s definition of ILI, which is the industry standard for tracking illness.
When comparing the cumulative cases of positive COVID-19 with Kinsa’s cumulative atypical illness data, a correlation is seen. This is shown in figure 1. Each data point on the graph represents one of 250 counties from 32 states and Washington D.C. These counties were analyzed because they had confirmed cases of COVID-19 and enough thermometer readings to calculate atypical illness. The graph shows that there is a significant relationship between Kinsa’s fever anomalies and confirmed COVID-19 cases. This is a promising indicator that atypical fever data from Kinsa’s connected thermometers point to areas of confirmed COVID-19 cases. There is a less than 1% chance that the relationship between these two datasets occurred by chance, making this correlation strongly statistically significant.
Kinsa publishes its atypical illness data at healthweather.us. The map from March 14, 2020, before the local government implemented aggressive social distancing orders, shows that Florida’s illness level was nearly two times the typical ILI. The state has since seen a surge in positive COVID-19 cases. The chart below tracks Kinsa’s cumulative atypical illness levels and the number of CDC confirmed cases of COVID-19 over a 12 week period. The chart shows that as the illness level increases, the reported COVID-19 cases from the CDC also increases at the same rate.
In the absence of widespread testing, we cannot definitively prove that an increase in Kinsa’s atypical illness levels means there is an outbreak of COVID-19. However, our research shows there is a statistically significant correlation between the two and that Kinsa’s atypical illness levels are a strong indicator of a COVID-19 outbreak.
If you’re interested in tracking your own illness and contributing to Kinsa’s public health data, you can pre-order one of Kinsa’s thermometers here. You can also track your temperature and symptoms without a Kinsa thermometer through Kinsa’s free app.