fbpx

Kinsa Publishes Updates to Maps and Underlying Signal on HealthWeather.us

You may notice the data on our HealthWeather maps looks a bit different today — we’ve rolled out a sweeping set of changes to the maps, interface and underlying data, which we think will offer a clearer view of what’s happening with the spread of influenza-like illness, including COVID-19. Here’s what’s new:

Improved County-Level Smoothing

The county-level illness signal that underlies all of our data is smoothed — differences between counties are averaged together to generate a more easily-interpretable map, but one that says less about individual counties in some cases than it does about regional trends. This is still true, but our updated signal does a smarter job of it: in counties with lower population density (and fewer active users), more illness data from the surrounding region is used to calculate the illness signal. In counties with higher population density (and more active users), the signal is much more specific to that county than it was in the past.

This change is responsible for the most visible differences you’ll notice in the map. We think it tells a more accurate story, particularly in urban areas, while still allowing us to tell broad regional stories about less-populated areas.

New Signal Strength Indicator

In concert with better county-level smoothing, we’re now publishing a signal strength indicator, just above the time series chart, here:

In general, areas with low signal strength incorporate more data from the surrounding region — for these counties, it’s best to think of the map and charts as depicting what’s going on in the region broadly — and areas with high signal strength tell very specific story about what’s happening locally in that county.

Updated Interface and Visualizations

We’ve also made some updates to our map interface, to make it easier to find and understand different modes, interpret time series data, and understand what’s happening with those atypical modes. We expect further incremental improvements along these lines to roll out over the next few weeks. Here’s a short list of what’s new

  • New Observed Rt mode — this new mode visualizes Rt values with respect to 1.0; values below 1 appear as negatively coded, since they indicate community spread of influenza-like illness is slowing. Values above 1.0 are positively coded to indicate transmission is increasing. This replaces the old “trends” mode, as it better reflects how community spread is changing over time.
  • Updated Atypical Rt mode — this mode only shows positive values now, similar to Atypical ILI. This highlights the important information here, namely areas where more community transmission is happening than expected.
  • Atypical ILI for all counties — in the past, we excluded counties with especially low population from our Atypical Illness analysis. Changes to the signal around geographic smoothing, and the new Signal Strength indicator, make this unnecessary, so we’re publishing Atypical illness for all counties, as of now.
  • New time series chart — we’ve updated the time series chart design for clarity and ease of interpretation.
  • New naming convention for modes — we’ve updated the names of our modes to make it clearer which data is being visualized. These are now:
    Observed Illness (ILI) — previously “Observed.”
    – Atypical Illness (ILI) — previously “Atypical.”
    – Observed Transmission (Rt) — new mode.
    – Atypical Transmission (rt) — no change.

What’s Not Affected

While these changes are far-reaching, several things are notably not impacted by them:

  • Older analysis posts — we started using this new signal in our analysis posts over the last few weeks as we were testing; older posts that used the previous signal will be updated with a note to that affect.
  • Technical write-up and academic papers — the science that underlies our signal and atypical illness analysis has not changed.
  • Public Healthweather API — the data we’ve provided to public health researchers for is not affected by these changes, and will continue to use the old signal. Note: we did update the “expected” model, which has the effect of providing more Atypical Illness data over the course of the summer, but smoothing and other changes are not reflected in the Public API.