Authored By: Alex MacAulay, Debbie Lai, Igor Kofman
When it comes to understanding COVID, one of the most useful things to know is the infection growth rate, or, as public health experts refer to it, R(t). In particular, it’s important to know how R(t) is changing in your local community in a timely manner, so that you and your local leaders can make appropriate decisions.
Today, we are making a change to the way we calculate R(t) to better serve lower population regions and regions with lower case counts, as well as to improve the timeliness of our R(t) metric. We want to let people know faster when COVID is growing or shrinking in their communities, so that they have the most up to date information on how policies and actions can achieve different outcomes.
For those of you interested in the technical details, you can see them in our Github here. For those of you who want a high-level explanation, there are two key things to know about the changes we’ve made:
(1) We now calculate R(t) based only on cases, rather than cases and deaths.
We had previously incorporated COVID death data into our estimation of R(t) due to the poor reliability of case data. But we have found that using COVID death data reduced our utility as a leading indicator because death data also tends to lag, i.e. there is typically a period of time between when an individual is infected and when an individual dies from infection. In addition, there are often too few deaths to generate an accurate R(t), particularly in low population counties.
We also found that incorporating death data made it more difficult to interpret our metric. When there were major changes in R(t), it was unclear if this result came from changes in cases or deaths. And increasingly, we are seeing that cases and deaths do not always align. This fact doesn’t mean that death data isn’t important to consider; we hope to incorporate that data in a separate metric, which we are working on.
(2) We’ve improved the responsiveness of calculations to new data.
Our R(t) estimate updates as new data on COVID cases comes in. How quickly R(t) shifts up or down in response to new data depends on the statistical significance of the new data. As a result, new data in high population / high COVID areas changes the R(t) estimate more quickly than new data in low population / low COVID areas. We’ve tweaked the way that this happens to increase how quickly new data affects the R(t) estimate in low population / low COVID areas in a way that reduces the overall “lag” from changes in data to changes in R(t).
As always, we anticipate that improvements will be an ongoing effort.