In this post I propose using week-over-week case ratios as a better way of analyzing COVID-19 case growth changes. This method makes it easier to detect changes in growth rates and assess countries' mitigation efforts. The graph also shows the interesting trend that despite Italy's early explosion of cases, other countries' case growths appear to converge to the Italian trajectory.
While the go-to sites for COVID-19 statistics and visualizations (like Worldometers and the Financial Times) are a great service to the public, their visualizations make it difficult to see what progress is being made in each country as they start bending towards the inflection point (fewer cases and deaths from the day before). This post focuses on a different way of looking at the data that makes it easier to answer questions such as "what actions are having an impact?", "is there light at the end of the tunnel?", "when can I go back to work?", and "how long should I wait until it's safe to look at my 401k?"
As an American, I wanted to better understand just how impactful the missteps on testing and shelter-in-place orders have been and what can make a difference going forward. Accordingly, I specifically focused on western democracies in my analysis (i.e., EU, US, Canada), since those are the realistic benchmarks for the US response (and we’re more or less all going through this together). In order to normalize the data, I adjusted timelines to when the country hit at least 100 cases. Generally, I think that contact tracing is lost at this point and the country is fully in community spread.
The resulting graph looks like this:
The media likes to focus on daily changes because it’s the most immediate way to keep score (you’re either hitting new highs or reaching the inflection point) and probably the easiest way to communicate status to the public. However, I selected week over week for several reasons:
Saying that we expect X many times the number of cases at this time next week is actually pretty clear to communicate IMHO, especially for planning purposes.
It allows for the incubation period (around 5 days), which the daily comparisons ignore.
It smooths out the daily fluctuations in the graphs, which makes it easier to see what’s going on.
Because it compounds 7 days of growth, there is more meaningful separation of the different countries, again aiding viewing and interpreting how countries are doing.
So what are the takeaways so far :
Countries tend to be caught off guard because the early growth rates are typically very large. This is especially true of Italy where the growth rates were over 100x at the beginning. THIS IS NOT TRUE OF THE US, which appears to have squandered its early warning about the potential magnitude of the pandemic and let it spiral out of control.
Notwithstanding the early onslaught and the ongoing crisis of their health systems (note that I am plotting cases, not outcomes), Italy has been on a steady downward trajectory for several weeks.
Even more surprising, the rest of the countries seem to be converging to the same trajectory as Italy! Asymptomatic community spread perhaps?
Spain and especially the United States squandered their chances to control the virus. Whether this is due to lack of testing or dragging their feet on social distancing is a question to be answered.
So where to go with this? Some good questions are:
What is the impact of pervasive testing on managing growth rates? We know especially that the US has struggled to catch up with testing and still lags behind on a per capita basis.
What is the impact of social distancing? Knowing when each country enacted various shelter-in-place orders and what percentage of the population was included (i.e., Italy, US have locked down different states or regions at different times).
Is the growth rate governed by how “spread out” a country is? This would be estimated from the population and area of a country. If there is a readily available measure of percentage living in urban areas that would be excellent.
Is the growth rate governed by how democratic a country is or how much trust in government there is?
What does the analysis for death rates look like, including all of the factors above?
Comments