Kinsa’s Atypical Illness Signal is a Leading Indicator of COVID-19 Outbreaks

Early detection of community spread is critical for containment of emerging epidemics, and Kinsa’s atypical illness signal is an advanced warning system for the COVID-19 pandemic. The signal provides two weeks advance notice of first deaths, demonstrates the impact of social distancing within days, and provides a 25-30 day notice of when outbreaks will peak, as demonstrated in New York City (Figure 1) and across the US (Figures 2-4). In the absence of widespread testing and contact tracing within the United States, Kinsa’s signal can play a pivotal role in detecting the emergence of and managing ongoing COVID-19 outbreaks.

Kinsa Real Time Atypical Illness Compared to New York City Reported Deaths

Figure 1: In New York City, atypical illness peaks 18 days before the peak death rates with a very strong correlation (r = 0.92; P < 0.0001), and there is a less than 1% chance this relationship occurred by chance. The atypical illness curve is offset 18 days forward on the bottom panel to illustrate the strong correlation between these signals.

Outbreak Onset

Early detection and containment is of central importance to curbing the spread of infectious disease, as countries with proactive testing, detection, and contact tracing efforts have successfully ‘flattened the curve’ (link). Our past research has demonstrated that Kinsa’s atypical illness signal is an early indicator of when and where COVID-19 outbreaks are about to emerge. We found that atypical illness warnings were first seen in US states 14 days before the first COVID-19 death was reported. This delay is in line with our growing understanding of the progression of COVID-19 illness, where it takes roughly 13 to 18 days for an individual to progress from first displaying fever symptoms to death. Our atypical illness, therefore, provides an early indicator of emerging outbreaks, as COVID-19 transmission appears in Kinsa’s syndromic fever monitoring data long before deaths are reported. In the example of New York City (Figure 1) as well as Louisiana and Michigan (Figures 3, 4), we clearly see atypical illness levels rise before both confirmed cases and deaths, demonstrating that atypical illness identifies the onset of the epidemic before official reports.

Intervention Efficacy

One complication of assessing the quality of social distancing interventions is it takes a long time to progress from COVID-19 infection to commonly measured outcomes, such as confirmed case counts or death. Factors contributing to this delay are the virus’s long incubation time (2-14 days), the 11-day duration from symptom onset to ICU admission (when cases would be confirmed), and an additional 7-day duration from hospital admission to death. With this in mind, we could expect there to be an 11+ day delay for the impacts of social distancing to be reflected in confirmed case counts, and a longer 18+ day delay to see the impact on deaths. This is a serious challenge for assessing the quality of social distancing measures and whether they are working now.

Kinsa’s atypical illness signal allows for the rapid assessment of social distancing directives, as the signal is available in real-time and responds quickly to policy interventions. The signal responds rapidly because our data includes all individuals expressing feverish symptoms, and this could be either individuals with COVID-19 or seasonal influenza infections. The relevance of this is that influenza incubates for only 2-3 days before fevers occur, and the response of reducing off disease transmission should be reflected in the fever data within a matter of days.

We see this expected response to social distancing across the United States, where our real-time illness signal began to quickly decline post-social distancing, as demonstrated in New York City, Santa Clara, and Miami, as well as across the rest of the country. In general, we find that the atypical illness signal responds to social distancing 3 days post-intervention, consistent with the incubation time for influenza. This is clearly seen in New York City, where atypical illness begins to decline roughly 3 days after the implementation of distancing directives (Figure 1). This rapid response should also be reflected in a later COVID-19 response, as both viruses are transmitted via social contact. This feature of our signal gives us insight into the future progression of epidemics that we would not have through COVID-19 case numbers alone. Understanding when a future COVID-19 epidemic will peak is of crucial importance, as hospital beds and ventilators are in short supply, and efficient allocation of these resources is required to minimize deaths.

Outbreak Peak

We do, in fact, see this delayed impact of distancing on COVID-19 outcomes that were initially visible in Kinsa’s atypical illness signal. We observe peak atypical illness levels on average 25 days before the peak in confirmed cases (+/- 10 days), and 30 days before the peak in daily deaths (+/- 9 days). These findings are aggregated from findings across 35 US states that had at least 100 COVID-19 deaths to-date. These delays are roughly in line with the duration from COVID-19 transmission to outcomes, where the virus might incubate for 2 to 14 days before expressing symptoms, take an additional 11 days to be confirmed upon hospitalization, and an additional 7 days until death. These findings suggest immediate dynamics observed for atypical illness may express themselves later in the COVID-19 outbreak.

This delay between peak atypical illness and COVID-19 can be seen clearly in New York State City. Here, the atypical signal peaks shortly after the implementation of distancing measures and quickly declines in response to distancing, followed by peak death rates 18 days later (Figure 1). These signals are highly correlated at an 18-day lag (r > 0.90), and the relationship is significant (P < 0.001), suggesting that dynamics in Kinsa’s atypical illness signal are indicative of the future progression of COVID-19 epidemics. This delay is expected as it takes an estimated 18 days from symptom onset to death. We see similar patterns in Louisiana and Michigan (Figures 3, 4), where atypical illness correlates strongly with the death outbreak 18 days in advance, as expected by COVID-19 disease progression.  For the entire state of New York, atypical illness is highly correlated with confirmed cases (r = 0.83; P < 0.001), hospitalization (r = 0.84; P < 0.001), and death rates (r = 0.90; P < 0.001), roughly one month in advance. This early insight into the future progression and dynamics of COVID-19 is of critical value to monitoring and managing outbreaks, and would not be available through monitoring either influenza or COVID-19 alone.


These findings suggest that dynamics in our atypical illness signal are indicative of what is to come in nascent COVID-19 outbreaks. Atypical illness is a leading indicator of when and where an outbreak is likely to occur, giving us a two-week lead on where to focus resources and implement social distancing interventions. Atypical illness responds to social distancing interventions within two to three days, allowing experts to rapidly assess the quality of interventions before the impacts on COVID-19 outcomes are realized. Finally, peak atypical illness is a one-month leading indicator of COVID-19 deaths and other consequences, demonstrating that atypical illness dynamics can provide insight into the future progression of COVID-19 epidemics.

Kinsa Atypical Illness Compared to New York State Hospitalizations

Figure 2: Daily rates of confirmed cases, hospitalizations, and COVID-19 deaths in New York State overlaid with Kinsa atypical illness. We observe similar dynamics between our atypical illness signal and these outcomes, where atypical illness peaks roughly one month before local death and case rates. Additionally, atypical peaks and begins to decline shortly after social distancing interventions, demonstrating that our signal allows for the rapid assessment of interventions. All of these outcomes are highly correlated with atypical illness at a one-month lead (r > 0.80).