Post #1367: COVID-19 hospitalization trend to 12/25/2021. No big uptick, and other surprising aspects of U.S. COVID hospitalization data

Posted on December 26, 2021

Main result

As of Christmas Day 2021, there had not yet been any jump in U.S. COVID-19 hospitalizations to match the recent jump in cases.  Below I show reported daily new admissions through 12/25/2021, and total cases in the hospital as of 12/26/2021.  As well as a separate graph of new admission and newly diagnosed cases.

But isn’t it too soon to tell, yet?  Nope.  The big uptick in daily cases for the Omicron wave started on 12/17/2021 or thereabouts.  Based on the historical lag between new cases and new hospitalizations, if there were going to be a big upsurge in hospitalizations, we should have seen it by now..

The lack of a big wave of hospitalizations is consistent with estimates of Omicron’s case hospitalization rate, from other countries.  Compared to Delta, for something like equivalent populations, that appears to be between 20% (South African analysis) and ~ 33% (UK and Scotland analyses).

It’s probably still no picnic to be (e.g.) an ICU nurse in the current environment.  As of today, based on the data below, COVID-19 patients occupy more than one-quarter of all U.S. ICU beds.

But for the U.S. as a whole, that’s not yet getting much worse, due to the Omicron wave.

Details follow.

Source:  Calculated from “COVID-19 Reported Patient Impact and Hospital Capacity by State Timeseries” data file , USDHHS, accessed 12/26/2021

Source:  Same as above for hospitalizations.  New cases calculated from NY Times COVID Github data repository, accessed 12/26/2021.



For the next week or two, it will be hard to track the true trend in new COVID-19 cases.  Holidays disrupt virtually every aspect of new-case reporting, both here and abroad.  Many data series are not updated, many are updated sporadically.  The behavior of the population itself changes temporarily over the holidays, with (what appears to be) a lower likelihood of seeking a COVID-19 test.

Near as I can tell, the only COVID-19 series that’s consistent over time and reported promptly is U.S. hospitalizations for COVID-19.  That’s a measure of the pandemic that greatly matters, because it tracks serious COVID-19 cases.  More to the point, hospitals don’t shut down over the holidays, and DHHS continues to update its national summary of hospital reports daily.

Normally, I’d just read the new hospitalizations data off the CDC COVID data tracker.  But CDC isn’t bothering to update that over the Christmas weekend.  So I downloaded the raw data from the U.S. DHHS (at this link), and summarized it myself.  (My numbers are quite close to, but not identical to, what CDC had been showing.)  I’m using SAS software, but there’s nothing here that couldn’t be done with Excel.

I ended up learning a lot of surprising facts about U.S. COVID hospitalization data.  Things that I just had not realized before.  For example, based on the DHHS hospitalization data above:

  • Only about a quarter of COVID-19 cases end up in the ICU.  That has been virtually constant since the start of the pandemic.
  • The average length of stay of COVID-19 cases is only about seven days.  That has also been nearly constant since early in the pandemic.
  • Equivalently, as shown above, total cases in the hospital lags new admissions by about a week, and the ratio of total to new cases has been virtually constant for the pandemic.
  • For every 100 admissions with lab-test-documented COVID, there’s another 50 or so with suspected COVID.  Judging from the data, most of those turn out not to be COVID.  CDC only reports admissions with confirmed COVID-19.  So that’s what I reported, above.
  • The CDC COVID data tracker website appears to have been reporting adult admissions, not total admissions.  (But it does not matter for trend data — pediatric share of admissions is small and appears to change only slowly.)  Again, that’s what I reported above.
  • As of today, COVID-19 cases account for just over one-quarter of occupied ICU beds in the U.S.

But what’s the time lag?

I need to be precise here.  What’s the time lag between two reported data series:  Seven-day moving average of reported new COVID-19 cases, versus seven-day moving average of reported new COVID-19 hospitalizations.

In other words, if one line on the second chart above spikes upward (newly diagnosed cases), when should we expect to see the other line spike (hospital admissions)?  If it is, in fact, going to spike?

This question tends to get a lot of confused and ill-informed answers because people are not precise about what, exactly, they are talking about.  I’m talking about the two lines on the second chart above.  What’s the typical lag between those two lines?

For sure, there’s a fairly long lag between infection and onset of symptoms.  A typical estimate might be 4.5 days.  There can be a fairly long lag, even on average, between onset of symptoms and hospitalization.  There, numbers are all over the map, but I think a median of around three days is right.

But neither of those directly answers the question about the two lines above.

And, of course, when given this possible series of events:

  • Infection
  • Symptom onset
  • Diagnosis
  • Hospitalization

Everybody wants to report the “up to” or “as much as”.  COVID-19 patients are sometimes hospitalized as much as one month after onset of symptoms.  That’s true, but that’s rare extreme case is not relevant to the question I’m asking.

FWIW, I have seen at several scholarly reference suggesting that the typical time between actual diagnosis date and actual admission date is typically quite short.  Just about one day, on average.  I’m not sure how representative those studies have been.  For sure, the lag between symptom onset and hospitalization is longer than that.

And, again, those actual dates aren’t what I’m after.  I want to know, with all reporting lags included, how the timing of the two lines above works out.

By eye, it sure doesn’t look like there’s much of a lag.  Accordingly, I did a quick and dirty statistical analysis.  I tested the correlation between the two series, at various lags.  Between 2/2/2021 and 11/1/2021, the correlation wast greatest, at a one-day, between daily change in new cases and daily change in new admissions.

That is, empirically, for the two lines on the second chart above, (the change in) reported hospital admissions lags (the change in) reported newly-diagnosed cases by one day.   That’s the best fit to the data.

I’m persnickety about this because I keep having people call me out on it, citing some supposed authority.  Authorities clearly state that there’s a long lag between fill-in-the-blank and second fill-in-the-blank (infection … hospitalization, symptom onset … hospitalization, actual date of diagnosis … actual date of hospitalization).

What I’m saying is that, on average, there’s only about one day’s lag between reported newly-diagnosed cases, and reported new hospital admissions, for COVID-19.

This means that if there were going to be a spike in admissions proportional to the spike in new cases from Omicron, that should have shown up in the data by now.  But it hasn’t.  New admissions are up, but just a bit.

But the lack of decline in admissions remains a mystery

And yet, something is not right.  If it’s true that:

  • the case hospitalization rate for Omicron is 20% to 33% that of Delta.
  • Omicron now accounts for almost all cases.
  • New cases are up about 50%.

Then why aren’t new hospital admissions falling?  No, actually, why aren’t they falling by a lot?

Simply put, taking the numbers above, even accepting the higher 33% figure, then 33% of 150% of Delta admissions = 50% of Delta admissions.

In other words, if everything were as stated above, and there were no other unknown time lags, then daily hospital admissions should have fallen by 50% already. 

And, obviously, that hasn’t happened.  So at least one of the assumptions above must be wrong.

A little spreadsheet analysis shows that one way we can achieve a 50% increase in newly-diagnosed cases, and no change in new hospitalizations, is if half those new cases are Delta.  (That’s assuming the case hospitalization rate for Omicron is 33% that of Delta).

In other words, one simple way to make sense of this is if the CDC’s most recent “surprise” estimate that Omicron was 73% of new cases (as of the week ending 12/18/2021) was substantially in error.  If they managed to overstate that by a large margin, then that, by itself, would explain the current observed daily new hospitalization count.

In any case, this is exactly why I’ve been waiting to see a decline in COVID-19 hospitalizations in Great Britain.  Or some reduction in average apparent severity, if we’re just seeing a lot of cases with COVID, instead of cases for COVID.  So far, I haven’t see it there.

And now, I haven’t seen it here, either.  Not much else to say, except that I’m baffled.  And that hospitalizations are not (yet) rising in proportion to all those new Omicron cases.