A simple COVID-19 mortality model with John Hopkins University dataset for the US.

I am not an epidemiologist, nor a public health expert

-- Potentially useful (but heart-wrenching) data at the end of this post. --

I've pulled the JHU COVID-19 mortality dataset (for US only), fit a curve to it, and built a simplistic mortality model. This is not an epidemiological model, this is a simple curve fit.

Last week's fatality doubling-period was 2.53 days. Over the last week, that has increased slightly to 2.57 days (which is small, but good!)

For anyone who wants to plot the curve, the model is:
deaths =.0003 * e^(.2371 * days_since_jan_22)

The projections I observed for last week came in too low by about 25%. I hope this curve performs better than last week's, but I'm trying very hard to err on the side of being conservative.

Here are my projected aggregate US death tolls (at the end of each day) for the coming week:

Mon, Mar 30 -- 3,014
Tue, Mar 31 -- 3,821
Wed, Apr 01 -- 4,843
Thu, Apr 02 -- 6,139
Fri, Apr 03 -- 7,781
Sat, Apr 04 -- 9,863
Sun, Apr 05 -- 12,502

If we see three days straight of those predictions being higher than the actual number of recorded deaths, it's the best news ever: it means we've broken out of exponential growth and are heading into the peak of the bell-shaped mortality curve.

In the tragic event this forecast is correct, as many Americans will die of COVID-19 over the next four days as have succumbed to it over the last 30.

I'm beginning to grow skeptical that the mortality tallies we're getting are accurate. There are scattered reports of failing to do postmortem testing (because of a shortage of test kits, not due to any conspiracy), but it's the best data we have for the moment.

Many people expressed skepticism after the very slow growth in fatalities a week ago, my own projection of 1800+ aggregate mortality seemed unreasonably high.

It was low by 25%. Exponential growth is a bear.

Raw data is at the link