Your HealthWeather Questions, Answered

Where does the data come from that’s used on HealthWeather?

What you see on Healthweather is an aggregation of many different data sources, allowing Kinsa a robust and well-rounded insight into illness. These data sources include COVID data from Johns Hopkins University, lab testing levels and diagnostic data, and most notably, aggregated, anonymized fever and symptom data from Kinsa’s network of more than 2.5 million smart thermometers in use across the country. Proprietary data from Kinsa smart thermometers allows us to track and predict the spread of illness earlier than any other existing system. Since the thermometer is one of the first things you reach for when you’re ill, we often see fevers in our data days before a doctor even comes into the picture. It’s also more complete. Imagine someone with limited access to healthcare: They might take their temp, follow our triage guidance and treat at home and never even see a doctor.

What happens to the data after it’s collected? 

It depends on the data. External data, like from JHU or the CDC goes through a process called ETL— extract, transform, and load —  where the data are normalized, standardized to fit within Kinsa’s system, and then incorporated with the rest of our data. Very little is changed about the data itself.  For data captured from Kinsa’s network of thermometers: Raw data are cleaned, harmonized, processed and aggregated to create Kinsa’s proprietary and unique core signals like percent ill, long-duration fever or secondary attack rate — all things I think HealthWeather readers are familiar with. Given our focus on privacy at Kinsa, all individual data are aggregated and anonymized to ensure no visibility as to who contributed data to a signal in a region. Then, we apply data science machine learning (ML) techniques. Think of these like advanced, adaptive statistical models. We use them to create unique illness insights including illness forecasts and local illness risk.

What are some caveats or limits to the data readers see on HealthWeather?

Since inception, Kinsa’s mission has been to curb the spread of illness through earlier detection and earlier response to symptoms. Capturing data from individuals moments after they fall ill — before they see a doctor or visit a lab — allows Kinsa’s illness signal (our aggregate data) to show illness levels rising earlier and faster than any other source (known as high sensitivity), but it cannot differentiate between what illness is spreading as accurately as, for example, diagnostic tests can (known as high specificity). Additionally, Kinsa does not capture illness indicators uniformly across the country. Density of our thermometer coverage tends to mirror population density, with highly populous urban areas having a stronger thermometer concentration than sparsely populated rural areas. While this is a caveat of our data, it is not a strong limitation. To understand illness trends in a less populated geography, we don’t need as many thermometers in use!  Kinsa also uses a technique known as geo-smoothing, or aggregating neighboring signals in areas of lower sensor penetration to get a good idea of how illness is moving through an area.