Post #1351: Principal versus secondary diagnosis of COVID-19: Why this matters a lot.

Posted on December 15, 2021


Hospitalizations with a principal diagnosis of COVID-19 are:

  • Persons being treated for COVID-19.
  • Hospitalizations that were caused by COVID-19.
  • Hospitalizations that would not exist if COVID weren’t here.
  • Hospitalizations the are added on top of hospital’s existing case loads.
  • Hospitalizations that can fill up all available beds if too many occur.
  • An additional expense burden generated by COVID-19.
  • An indicator of the severity of illness of the current COVID strain.

Hospitalizations with a secondary diagnosis of COVID-19 are:

  • None of the above.

I may have glossed over a fine point or two above.  But that’s the big picture.  By and large, hospitalizations where COVID is a secondary diagnosis are hospitalizations that would have occurred anyway.

In this post, I’m going to explain:

  • why this distinction hasn’t much mattered, in the pandemic, up to now.
  • why the cited statistics out of South Africa regarding the split of hospitalizations by principal and secondary diagnoses are so eye-popping.
  • what this seems to imply for asymptomatic Omicron infections.

Conceptual model of hospitalizations with secondary diagnosis of asymptomatic COVID-19.

This next part is not rocket science. Though I will do my best to make it seem like it.

One statistic out of South Africa that I just cannot shake is that 76 percent of the persons hospitalized with COVID-19 in one large South African hospital system had that as a secondary diagnosis.  Based on the quotes in the reporting, most of that consisted of persons who were hospitalized for something completely different, and had an asymptomatic COVID-19 infection discovered only after routine testing in the hospital.

That may not jump out at you.  But it leapt off the page for me.  Because that is vastly different from any plausible estimate of COVID-19 inpatient cases in the U.S.

The South African data imply that either:

  1. there is vast pool of asymptomatic Omicron infections in South Africa,
  2. or almost nobody is actually being hospitalized for treatment of (with principal diagnosis of) COVID-19 in South Africa.

Let me just do a quick back-of-the-envelope with U.S. data to show you where I’m coming from.  They key concept is that, by any plausible estimate, persons with an active but asymptomatic COVID-19 infection account for a tiny fraction of the population.  If those people randomly show up in non-elective hospital admissions, you end up with just a tiny number of U.S. hospital admissions with asymptomatic COVID-19. 

Let me assume that, right now, the following is true:

  • Newly-diagnosed COVID-19 cases are averaging about 35 / 100K / day.
  • We have about 1.9 actual cases of COVID per diagnosed case (based on CDC seroprevalence studies, cited in earlier posts).  For purposes of this discussion, I’m going to assume that the excess 0.9 cases are asymptomatic.  That’s reasonably consistent with an estimate, early in the pandemic, by Dr. Fauci, that about 40% of total COVID-19 cases were asymptomatic.
  • About 7 percent of diagnosed COVID-19 cases end up hospitalized for COVID.
  • The U.S. has about 36 million total hospitalizations per year.
  • About 20% of U.S. hospitalizations are scheduled or elective.

Now let me work up a simple model for U.S. hospital admissions with COVID-19.

All this assumes is that U.S. non-emergency admissions have the same prevalence of asymptomatic COVID as the U.S. population as a whole.  In other words, if 1-in-1000 persons in the community has asymptomatic COVID, then 1-in-1000 non-elective admissions will have it.

So, reading down the lines in red:

  • The U.S. has about 24 / 100K / day non-elective hospital admissions, in a typical year.
  • The U.S. currently has about 2.4 / 100K / day COVID-19 admissions.  (Which, by itself, shows you the cost burden of COVID-19).
  • On any given day, about 0.3% of the U.S. population is probably walking around with an asymptomatic COVID-19 infection.  That’s 3-in-a-thousand.
  • Applying that fraction to all non-elective admissions, that means there are 0.08 / 100K / day such admissions with asymptomatic COVID-19.
  • And that amounts to about 3% of all U.S. admissions with any diagnosis of COVID-19.

The bottom line is that maybe 3 percent of U.S. hospitalizations with COVID-19 should be persons who were admitted with an asymptomatic infection. This compares to something approaching 76 percent in South Africa.

First, but least important, I think this explains why this issue of counting COVID-19 hospitalizations by principal versus secondary diagnosis has never been addressed in the U.S.  It has been a non-issue because there aren’t enough cases with COVID-19 as secondary diagnosis to produce any serious distortion in our statistics.

Second, and far more important, now ask this question:  What would it take to generate something like the 76% of cases in South Africa that are hospital admissions with COVID-19, but not for COVID-19?  Arithmetically, it would take a combination of:

  • One-tenth as many admissions for COVID-19, and 10x as many asymptomatic infections.  Total of all admissions with any mention of COVID-19 would be 60% lower, similar to what has actually been observed with Omicron relative to Delta in South Africa ( from this New York Times reporting ).
  • One-twentieth as many admissions for COVID-19, and 5x as many asymptomatic infections.  Total of all admissions with any mention of COVID-19 would be 80% lower.

And that’s why the South African data are so eye-popping.  We couldn’t come anywhere near that 76 percent ratio, given the situation in the U.S. under Delta.

Once again, my point is that the situation in South Africa under Omicron isn’t merely different from that of the U.S. under Delta, it’s vastly different.

The crude hospitalization rate in South Africa, under Omicron, is only 61% lower than it was under Delta.  (That’s calculated directly from this New York Times reporting on actual hospitalizations per 100K).  But that disguises the true situation, because such a large fraction of those hospitalizations under Omicron are unrelated to treatment of COVID-19.  They are hospitalizations with secondary diagnosis of COVID-19, not principal diagnosis of COVID-19.

Nobody expected that.  Nobody expected that the vast majority of hospital admissions with any mention of COVID-19 would be admissions for treatment of other conditions.  To the contrary, I think everybody’s history looks just like the U.S.’s history.  By and large, the overwhelming majority of hospitalizations with COVID-19 were hospitalizations for COVID-19.

That has now changed, for South African under Omicron.  But most hospital statistics are blind to that.  South Africa counts all hospitalizations with any positive test for COVID.  Some (but not all) U.S. statistics do the same thing, as I explained in the excruciatingly boring Post #1349.  Most counts of COVID hospitalizations simply combine all admissions with any diagnosis of COVID.  And if the mix of cases shifts from mostly principal diagnosis (admission for treatment of COVID) to mostly secondary diagnosis (admission for some other condition), that critical difference will just be swept under the rug.  As it has been in South Africa.


I remain a contrarian on the issue of Omicron severity of illness.  You will now see reporting that grudgingly acknowledges that Omicron cases may be somewhat less severe on average.  All the evidence I see so far suggests that Omicron cases are vastly less severe, on average.  No matter where I look, I find something like one-fifth to one-tenth the hospitalization rate, and virtually no deaths (so far).

The South African hospitalization case mix data floored me.  But to the extent that anyone else even noticed it, they seem to have shrugged as moved on.  I think that’s because they don’t grasp the full implications.  The only way you can get those numbers is with a combination of far fewer admissions for COVID-19, and far higher proportion of cases that are asymptomatic.  In other words, the only way to get that 76 percent figure is with a vastly lower average severity of illness.

And so, every day, I’ll check the WHO data.  And every day that I don’t see the deaths below, that’s another day that reinforces my opinion.  Omicron results in vastly less severity of illness, on average, than Delta.

Source:  WHO, accessed 12/15/2021

Addendum: What am I glossing over?

What’s the fine point that I’m glossing over?

I have neatly divided cases into those where the individual is being admitted for treatment of severe respiratory illness, and cases where COVID is present but not medically meaningful.  In the latter, cases where the person had an asymptomatic COVID-19 infection that was discovered after admission.

The world is never quite that net.

You may have some cases where COVID really is secondary, but it also has some medical relevance.  Those would be cases where the presence of mild respiratory illness tipped the balance in favor of hospitalization.

For example, in Medicare hospital data, you will frequently see Medicare beneficiaries with congestive heart failure (CHF, more or less, a weak heart) who are hospitalized as a consequence of a bout of pneumonia.  What gets coded as principal versus secondary diagnosis (CHF or pneumonia) is a matter of judgment.

But in any case, the mild respiratory illness was medically relevant.  Even if the primary reason for the hospitalization is CHF, that particular hospitalization would not have occurred absent the respiratory illness on top of the heart failure.

For COVID, I think those cases are relatively infrequent — cases where COVID wasn’t the main source of illness, but where a mild, non-life-threating COVID infection merely tipped the balance toward hospitalization.  Accordingly, I just ignored those for the discussion above.