Tuesday, April 14th marked the projected US peak date in #coronavirus death based on the University of Washington model—which is what the White House regularly references. I want to take a moment to explain what this model does and doesn’t do, because people are confused.

1/20

Here’s what it does do.

First, the model projects #coronavirus deaths and the hospital beds, ICU beds, and ventilators needed per day. These projections were created from a mix of mathematical projections and historic data from Wuhan, China, and other epicenters.

2/20

These death and hospitalization estimates are the most important aspect of the model. It’s what makes the model extremely useful for leaders preparing capacity for their healthcare systems.

3/20

Even now, civic and medical leaders rely on the model as a benchmark for assembling hospital beds, acquiring ventilators, preparing morgues, and assessing staffing needs.

4/20

Second, the model projects the effect of strict social distancing on total deaths and hospitalizations based on data from other countries and taking time lags into account.

5/20

This made the model useful for leaders deciding whether to enact social distancing and when to expect a slowdown in the growth of cases. These projections on the effectiveness of “strict social distancing” was originally based exclusively on data from Wuhan, China.

6/20

As other countries, such as Italy and Spain, became epicenters, the model has been updated.

7/20

This is what led to a lower projected death toll and hospitalization than the original estimates based on Wuhan.

Last, the model uses the data mentioned above to predict the peak date for resource needs and for deaths.

8/20

This has helped leaders set targets to stay ahead of the curve of the virus. It also helps give a sense of when different parts of the country will need the most resources.

9/20

While the lack of national coordination has had tremendous costs, one benefit has been scattered decision making by states that has led to staggered peak dates. This allows resources to move around the country based on peak dates estimated by the model.

10/20

However, the peak dates can be misleading, which brings me to the second half of my comments on what the model doesn’t do.

First, the model doesn’t take into account what happens after the end of May.

11/20

This is why the model shows a “plateau” in total deaths for most parts of the US within the next few weeks. Many wishful commentators are misinterpreting this plateau, and using it for their argument, saying that it’s time to “reopen” the country.

12/20

These misleading ideas are dangerous and will kill thousands of people. Here’s why: because the model makes no post-May assumptions, the charts currently assume indefinite extreme social distancing.

13/20

If extreme social distancing measures are lifted, the model will have zero predictive value anymore. Second, the model doesn’t actually predict the number of total COVID-19 cases. It predicts deaths and hospitalizations, not cases.

14/20

That’s because we don’t actually know the number of cases due to our extremely limited testing capacity. As a result, we know very little about the actual fatality rate. We also know little about how many people are recovered and immune.

15/20

And finally, we don’t know how many people are asymptomatic—which makes it extremely hard to decide when and how to “open up.” As a result, the University of Washington model tells us next to nothing about when to reopen the country.

16/20

Currently, the researchers are working to update the model to account for relaxations of social distancing and its effects on potential deaths. But this is very hard to do accurately.

17/20

It’s my suspicion that the researchers haven’t accounted for post-May effects yet because they know that it’s impossible to predict until testing is more widely available. Finally, there are other challenges to the current data that feeds the model.

18/20

For example, there’s a growing consensus that public data grossly underestimates the total number of deaths. Many deaths, especially from nursing homes and from people with many other underlying health conditions, are not being counted in the official reports.

19/20

Without that data, the model can only do so much.

20/20