The IHME Covid-19 Projections present historical data summaries and forecasts for infections and deaths. Daily death counts come from the Johns Hopkins database, which is a count of actual deaths directly attributed to covid (but not including so-called excess deaths that are likely covid-related). Infection rates, in contrast, are estimated rather than counted, “including those not tested.” How does IHME estimate infections?
Daily deaths is the best indicator of the progression of the pandemic, although there is generally a 17-21 day lag between infection and deaths… For estimated infections, we start with death estimates, then work backward, using infection fatality ratios to estimate infections based on deaths.
That’s how I do it too, using death as a lagging indicator. Divide today’s death count by the fatality ratio to arrive at the number of infections occurring 20 days ago. What fatality ratio does the IHME use? Doing the math on their graphs, it looks like they’re using a ratio of 0.85 percent. Where did they get that number?
We reviewed data from all locations where extensive COVID-19 testing has occurred, then used the infection fatality ratio from the locations where the ratios were lowest: the Diamond Princess cruise ship and New Zealand.
That’s how I arrived at a 0.6 percent fatality ratio, beginning in late April. Note that IHME intentionally uses the lowest fatality estimates; i.e., the actual ratio might well be higher than the 0.85 percent figure they’re using.
Two recent national seroprevalence surveys, in Italy and in the UK, come up with infection fatality rates that are not only higher than mine, and higher the CDC’s recently increased estimate (0.65%), but also higher than the IHME’s estimate. Adjusting for differences in national median age, the Italian and UK findings suggest a US fatality rate of around 1.1 percent.
Let’s say I split the difference between 0.6 percent and 1.1 percent: that’s 0.85 percent — the same as IHME’s current estimate. Raising the fatality rate raises the projected number of future deaths. The higher fatality rate also lowers the estimated infection rate — it takes fewer infections to cause the same number of counted deaths.
I’ll be using the 0.85 percent figure for the foreseeable future. What’s needed, of course, is a US seroprevalence survey, based on stratified random sampling of the population, to arrive at an accurate national estimate. It’s unconscionable that no such survey has been conducted. Or maybe it has… Here’s one of the IHME’s “major findings” from its August 7 update:
Lack of data sharing by the US government hampers our research: Our understanding of the drivers of the pandemic beyond mask use, mobility, testing, and seasonality is hampered by the lack of access to data. US CDC has many relevant datasets on the pandemic that they have refused to share with the research community. The switch of data reporting from US CDC to the US HHS has had little impact on our models since neither group is sharing much of their data with the research community. Some data that are critical to monitoring the response to the pandemic, such as mask use, are only collected through private-sector initiatives such as surveys conducted by Facebook, Premise, and SurveyMonkey. Federal government efforts to fill these critical data gaps have been limited to date.