Post #1391: Making sense of U.S. COVID-19 hospitalization data

Posted on January 9, 2022

Current level and growth in infections by state.

Daily new COVID-19 infections are increasing at a rapid pace almost everywhere in the U.S. There are only ten states where the growth rate in new cases is less than 50% per week  And (see below) most of those are areas that already have a very high level of cases.

Data source for this and other graphs of new case counts:  Calculated from The New York Times. (2021). Coronavirus (Covid-19) Data in the United States. Retrieved 1/9/2022, from”  The NY Times U.S. tracking page may be found at

With the exception of Rhode Island, we seem to be reaching the point where, on average, areas with the highest current level of infections are showing the lowest growth in daily new infections.  (Note the general downward slope in the scatterplot of level and growth in new infections below). That suggests were getting somewhere nearer the end of this wave.

That said, we seem to have a long way to go.  If you look in the middle of that mass of data points above, a typical U.S. state already has 200 new COVID-19 cases per 100K population per day, with new infections more than doubling in the past week.

Seems like the best we can expect is that a week from now, a lot of states will be where FL, DC, NJ, and NY are on the graph above.  Somewhere between 300 and 400 new COVID-19 cases / 100K population / day.  And a growth rate that’s not in the triple digits.

There’s not much we can do about that now.  Historically, it took about 12 days from the moment of infection, to the full reporting of that infection in the data.  What we’re seeing now are infections that occurred over the holidays.

And it sure looks like we aren’t doing much about it, in any case.  I’ve seen little or no re-imposition of (e.g.) state mask mandates or controls on public gatherings.  Nor has there been much of an increase in individuals reporting that they routinely wear a mask in public places.  We’re now at the point where we’re more cases and more hospitalizations than during last year’s winter wave.  And mask use is still nowhere close to where it was back then.

Source:  Calculations from NY Times (above) and U.S. DHHS unified hospital file.

Source:  Carnegie-Mellon University COVID Delphi project COVIDcast.

Sorting through state-level COVID hospitalization data.

It looks like a lot of state hospital systems are going to get a major stress test this week, from COVID-19 admissions.

That leads to the obvious question, what do COVID-19 hospitalization patterns look like right now, in the U.S.

To answer that, I stepped back and took a much more systematic approach to the U.S. DHHS COVID hospitalization data.  An approach that would allow me to start looking at the hospitalization data across all 50 states.

And that’s when things started getting weird.  Because, as it turns out, beneath the single U.S. averages lies a vast amount of variation.  Some of which makes sense, some of which does not.

And so, the answer to even the most basic question about hospitalization and COVID will depend strongly on where you live.  I don’t really think the actual practice of medicine varies that much.  But testing behavior and environmental factors do.

Here are a few questions I wanted to answer:

What’s the case hospitalization rate for new COVID-19 cases (i.e., what fraction get hospitalized).  Answer:  1% to 6.5%  This is a complete patchwork, and if there is any rationale for this variation, it’s certainly not apparent to me. 

My suspicion here is that a lot of this has to do with testing and test-seeking behavior.  And, possibly, with hospital testing practices (subject of a future post, but by that I mean, do they test every case coming in the door, and so find a lot of asymptomatic COVID cases?)

If individuals rarely seek testing, that will eliminate most of the lower-illness-severity segment of the population counted with COVID.  And of what’s left, you’ll see a high fraction hospitalized.  By contrast, locations were testing is encouraged or easily accessed, you’ll see a much broader population testing positive and a lower case-hospitalization rate.

Similarly, if hospitals test every person admitted as an inpatient, then they will find (and count) a lot of asymptomatic cases as hospital admissions with COVID.  And if they don’t test everyone, they won’t.  (An astute reader pointed me toward a press conference in the Kansas City area where hospital executives made it clear that they only test asymptomatic patients requiring general ansthesia (i.e., as a measure to prevent contamination of anesthesia equipment.)  In most other areas I have found, hospital systems advertise the fact that everyone admitted as an inpatient gets tested for COVID.


What’s the ICU use rate for COVID cases?  (Among those hospitalized, what fraction are in the ICU).  Answer:  10% to 33%.

The moment I saw this one, I recognized the pattern.  Higher elevation means less (partial pressure of) oxygen.  For a given level of lung impairment, you’re going to see lower blood oxygen saturation at higher elevations.  And so, you find that about one-third of the state-to-state variation in  ICU use per COVID-19 patient is associated with variation in elevation.

Anywhere from 10% to 33% of COVID-19 cases end up in the ICU.  And with the exception of Maine (a high outlier in this regard), that variation largely follows variation in elevation of the states.  Which, in turn, shows how much oxygen there is in the air.  The Mountain states show up with high average ICU use per case because … they’re mountain states.

This is particularly helpful to Washington, DC and to New York City, two areas very hard-hit by Omicron.  Roughly speaking, both of these locations are at sea level.

What fraction of COVID-19 admissions are children?  Answer:  1% to 14%.  Less than 1% of COVID-19 admissions in Maine are pediatric cases.  By contrast, 14% of admissions in Washington DC are pediatric cases.  If there is any obvious pattern to this, it escapes me.

What fraction of ICU beds are already in use:  Answer:  44% to 91%. 

What fraction of ICU beds are occupied by COVID patients?  Answer 11% to nearly 40%

I have no idea why Texas and New Mexico are such outliers, but based on the Federal data, they both have 92% of ICU beds already occupied.  But, for Texas at least, that’s not strongly linked to COVID-19.  They are only middle-of-the-road in terms of the fraction of ICU beds occupied by COVID-19 patients.

What is the trend in pediatric COVID hospital admissions as a share of all COVID-19 admissions?  It appears to have peaked.

This one gets so much press coverage, it’s worth putting up the national numbers, straight off the DHHS hospital file.  This is pediatric admissions with confirmed COVID as a fraction of all admissions with confirmed COVID.  The fraction of admissions that were pediatric rose from about 2.5% under Delta, to about 4.5% under Omicron.  That share now appears to have stabilized.

Certainly, the count of pediatric admissions will continue to increase.  But at the moment, that’s only in tandem with all admissions.