Frequently Asked Questions
What is Observed Illness?
Observed illness is Kinsa’s real-time measure of influenza-like illness incidence, similar to what CDC refers to as influenza-like illness (or ILI)1. This is aggregated from our network of more than 1 million Kinsa smart thermometers across the US. To learn more about the demographics of Kinsa users see our blog post on the subject.
What is Atypical Illness?
Atypical Illness refers to an unusual incidence of elevated flu-like illness levels. Kinsa quantifies atypical Illness by comparing real-time illness incidence2 (what is happening today) to our pre-COVID-19 illness forecast (the level of illness expected for this time of year). If the real-time illness is higher than expected, it is considered “atypical.” This allows us to estimate the proportion of the population in individual communities who are displaying unusually high levels of flu-like symptoms.
The “Atypical Illness” map mode captures atypical illness observed in the last day. Cumulative atypical illness is the sum of all atypical illness observed since March 1.
Key Takeaway: Atypical illness means levels of illness in a region have been higher than what is expected.
What is Illness Trend?
The illness trend map shows whether illness levels are increasing or decreasing. This map does not directly suggest anything about COVID-19 cases or the impact that it will have on local healthcare systems. Instead it suggests an overall trajectory of febrile illness, which includes normal cold/flu. Illness trend is measured over a one-week window.
Key Takeaway: Illness trend is the recent trajectory of community transmission of cold, flu and related illness, including COVID-19.
Has the map data changed over time?
There are two majors reasons that you may notice day-to-day variation in reported illness levels: (1) temperature readings may not sync with mobile devices immediately; a small percentage of such readings show up over the following week, which can change our illness signal slightly, and (2) we periodically release updates to our algorithms and data, as documented below.
- 5-22-2020: we updated our forecast of expected ILI levels by taking a new 12-week illness forecast covering May 15 through Aug 7 and integrating it with our previous forecast, which had been taken from March 1 and extended through May 24. This new forecast assumes aggregate social distancing behavior relaxes somewhat starting in mid-May, and will drive our Atypical illness signal moving forward. We expect to update it on an ongoing basis as conditions change.
- 5-13-2020: we made a calibration change to increase signal sensitivity at very low illness levels (e.g., below 0.4% ILI) — this affects both historical, current and forthcoming data. This change resulted in slightly higher current ILI across the board, and a small increase in cumulative atypical illness. We made this update because in the past, our signal was calibrated to capture large changes in ILI around the peak of flu season — not small changes when ILI is close to zero, which is what we’re seeking now in the context of social distancing. We believe this update will help us see changes in illness dynamics earlier than would otherwise have been possible as restrictions are eased going into the summer, and in similar low-ILI situations that might arise in the future.
- 5-06-2020: we updated our forecast to account for widespread social distancing starting on March 17 — this allows us to capture atypical illness levels continuously as influenza-like illness dropped following adoption of various social distancing measures. See our blog post for more details.
What does it mean for a geography to have 0% observed illness?
0% ill does not necessarily mean that no single thermometer has registered a fever or symptoms; it means that we are seeing relatively few illnesses compared to the number of active users we have in the geography. Due to widespread social distancing across the U.S., illness levels are at very low, off-season levels.
How many people need to be sick before it's noticeable in the data?
Kinsa’s illness data measures the percent of the population that is sick, not a raw number of infections. If only a few people are sick, the percent ILI for the population will not significantly change. When a larger number of people are sick compared to the overall population of a region, Kinsa will detect this rising level of illness.
Why is Kinsa showing low rates of illness when COVID-19 cases continue to go up?
Social distancing has caused incidence of flu-like illness to drop nationwide. We believe confirmed COVID-19 cases will continue to increase in the near term as testing becomes more widespread, even though overall population percent ILI has decreased nationwide.
How many users does Kinsa have?
Kinsa’s illness signal consists of data from over 1,300,000 thermometers with 60,000-160,000 readings driving our signal every day. To learn more about the demographics of Kinsa users see our blog post on the subject.
Why are many neighboring counties so similar?
Many cities span multiple counties, and residents of nearby counties regularly mix and share illness exposures. We also factor in information from neighboring counties in our estimate of illness levels, particularly in sparse rural counties with fewer thermometers. It’s probably best to think of our illness signal as representing a geographic area centered around a particular county, rather than a single county exclusively.
Key Takeaway: Similar illness levels across counties are expected due to patterns of human travel and interaction, which are not confined to counties.
Why are atypical illness levels not available for some counties?
While all three map modes are based on information aggregated from Kinsa thermometers, we take a more conservative approach to our atypical illness levels. In some parts of the country where there aren’t as many of our thermometers, we aren’t as confident in our typical flu season forecast. As our distribution of thermometers improves, we expect to be able to reach more of these areas, which are currently marked in gray in the atypical map mode.
Key Takeaway: In areas with fewer thermometers, there isn’t enough data to determine if the illness levels are atypical.
Are you seeing increased activity and does this affect your illness signal?
We are seeing increased use of Kinsa thermometers, but this does not affect our illness signal. Kinsa illness signals are based on the percent of active users that are sick at any given time. Increased thermometer usage even when the user isn’t sick will not change this percentage as these people were already included in our active users and, if not sick, will not drive up the percentage.
We also benchmark our signal against the Centers for Disease Control (CDC) at the end of every flu season when the CDC has finalized their illness reporting. We regularly see very high in-season correlation between the CDC’s illness incidence and Kinsa’s illness signal (R2 values greater than 0.95).
Key Takeaway: A significant and rapid change in temperature readings does not affect our illness signal.
How often are the maps updated?
Illness incidence levels are updated nightly.
Where can I learn more about your scientific methods?
Kinsa’s forecast extends the findings outlined in Dalziel et. al. (2018 Science)6 by applying similar modeling methods to Kinsa’s county-level illness incidence data.
For additional information, see our technical appendix here.
For prior work with others, and peer-reviewed publications, see www.kinsahealth.co/research.
How can I get in touch with you?
Do you work in public health, media, or at a business and want to explore a partnership with Kinsa? You can contact us here.
Information about our data privacy.
Kinsa takes individual privacy extremely seriously, and never shares individual health information. We strongly believe that there is a way to protect the individual user’s privacy, while also aggregating data for the greater good. The data that drives healthweather.us is composed of population level insights, so communities and healthcare systems can know where outbreaks are occurring. See Kinsa’s privacy principle for more information.
2Miller AC, Singh I, Koehler E, Polgreen PM. A Smartphone-Driven Thermometer Application for Real-time Population- and Individual-Level Influenza Surveillance. Clin Infect Dis. 2018;67:388–97.