The World Health Organization outlines five stages of community response intervention for epidemics: anticipation, early detection, containment, control and mitigation, and eradication or elimination. In order to reduce the impact of an infectious disease, early detection is a vitally important intervention tool that allows for a better-informed response with containment and control. In the case of COVID-19, our atypical illness map started monitoring for unusual fever clusters on March 1 in cities around the United States. Our approach has detected unusual fever clusters across the US that are significantly correlated with confirmed COVID-19 cases for individual counties and whole states. In essence, healthweather.us, and Kinsa’s atypical illness detection, have been validated to be an accurate early detection system for COVID-19 outbreaks.
To have a successful early detection tool, you first need a high quality data set that provides a good representation of each community, as well as, the ability to track and identify outbreaks of illness before people go to the hospital. During the COVID-19 outbreak in the United states, preventative measures such as widespread and rapid testing are not always available. In the absence of these safeguards, Kinsa was able to develop an early warning system that monitored atypical illness levels across the country and identified areas of potential COVID-19 outbreaks.
To verify that Kinsa’s atypical illness signal is a leading indicator of COVID-19 outbreaks, we compared the first day of atypical illness within a state to the earliest reported statewide COVID-19 death. We used death reports because confirmed COVID-19 cases underreport the actual number of cases due to a lack of widespread testing. Our source for these reports was the New York Times coronavirus death count.
On average, we observe illness anomalies within states 14 days before states report their first COVID-19 death. Of the 47 states that have experienced COVID-19 deaths, Kinsa’s atypical illness signal recorded illness anomalies at least five days prior to the first fatality in 41 states. In six states, we did not find any anomalies prior to the first death. In four of these states (Minnesota, Wisconsin, North Dakota and South Dakota), Kinsa did not observe anomalies.
In a majority of instances, healthweather.us was able to detect an abnormality five days prior to a death from COVID-19, with an average lead time of 14 days. This means that Kinsa’s atypical illness signal can serve as an early detection system in future outbreak monitoring. Though a work in progress, Kinsa’s atypical illness signal has effectively identified COVID-19 outbreak epicenters in advance of reported deaths. The team here at Kinsa is working every day during the COVID-19 pandemic to improve our insights as we work toward our mission to stop the spread of infectious disease.