Post #997: Deaths data, the more you look, the odder it gets

Posted on February 6, 2021

Edit 2/7/2021:  The cat’s now out of the bag.  Today’s Washington Post showed a huge spike in deaths.  A footnote reads:  “The spike on February 4 is due to Indiana’s inclusion of 1,507 historical deaths that were identified through an audit of death records and positive test results.”  I had a reader email me to point out the same thing. 

Using Indiana as the template, I can look at the graphs of state mortality data published in the 2/7/2021 Washinhton Post and spot similar obvious reporting artifacts in other states, including at least DE, SC, IA, NE).  And those are just the ones that are obvious.  That would not count any revisions that were subtle.

Bottom line:  For whatever reason, it seems to take state Vital Statistics departments a month to look over their death certificate accounts and make any adjustments.  And what we’re seeing with the ongoing high death counts.  The resolution of high deaths vs falling new cases and hospitalizations is that the deaths aren’t new.  They’re accounting adjustments.  Give it another week, and the reported daily deaths should plummet.

Original post follows:

The seven-day moving average of reported COVID-19 daily new cases peaked in the U.S. on 1/8/2021.  Every day, I look at the COVID deaths data.  It should have peaked by now, and started following the new cases down.  But it hasn’t.  This caught my eye because I’ve been expecting a decline in deaths, as a simple and direct consequence in the decline in new cases.

As the new cases / day graph has gotten cheerier and cheerier, newspapers have essentially stopped reporting it.  Instead, all you will see in (e.g.) the Washington Post is a graph of deaths.  I looked at that early on and said “well, that buys them a couple of weeks of fear-based jouralism”, because it takes about two weeks for the average COVID decedent to die.  And, historically, COVID deaths lagged COVID new cases by about two weeks.  Given the peak in new cases, the media should have gotten another couple of weeks of play out of the deaths.  Then deaths, too, would begin a steep decline.

Only, it’s not working out that way.

There remain plenty of opportunities for fear-based journalism.  The new and more infectious/virulent strains of COVID are real.  The media can live off speculation about the forcoming disaster from those new COVID variants, if nothing else.

And, of course, our media continue to be the ringleaders in American’s favorite COVID passtime, bitching about vaccinations.  Anybody who can read the news knows that there isn’t enough vaccine to go around.  But somehow, there’s story after story that boils down to people being upset that they can’t get their vaccine shot yet.

(Just FYI, our bungled vaccine effort places us is fifth in the world in terms of percent of population vaccinated.  That’s far ahead of (e.g.) China (by about 5-to-1), Russia (by about 10-to-1), or all of Europe outside of Great Britain.  Israel leads the world in COVID vaccinations per capita by a wide margin, but they have the obvious advantage in being able to perform mass vaccinations via space laser.*

* Despite this huge success, they apparently are having difficult recruiting volunteers for clinical trials of their proposed space-laser-based bris.  And if I have to mark those last two sentences as sarcasm, I’ve clearly misjudged my readership.

But the last I heard, they were having considerable difficult obtain

But in the here and now, current deaths is the last big piece of real, on-the-ground pessimism that fear-based journalism can latch onto.  It’s worth looking at the data for that reason alone.

Why are there still so many daily deaths from COVID?


U.S. data 6/1/2020 through 2/5/2021

Source:  NY Times Github COVID data repository, data reported though 2/5/2021

Above you see (seven-day moving average of) new cases and deaths, plotted on the same chart.   (I am only showing the second and third waves of COVID, as the deaths in the first wave were hugely disproportionate to the reported case counts.)   So what you see is the period where reporting of the number of new infections was consistent with reporting of deaths.

As you can plainly see (I think), in general, deaths lag behind new cases.  The orange line “sits to the right of” the blue line.

Let me narrate it, left to right.  We start with the second wave of COVID, where new cases were rising.  A couple of weeks later, deaths began to rise in proportion.  Labeled, the peak of new cases occurred about two weeks before the peak of deaths.

Continuing:  New cases then began to fall.  A couple of weeks later, deaths began to fall until they made a broad, shallow bottom between the first and second waves.  As new cases began to rise, deaths followed, roughly in proportion to new cases.

Then Thanksgiving came — first oval.  Both series showed a sharp drop related to reporting.  Simply put, offices were closed a lot over that period.  There was a backlog in getting the counts reported.

But while there was a permanent decline in total reported cases (discussed in an earlier post), the mortality data appear to show a very high degree of “catch-up” reporting.  Note the size of the gap between blue and orange lines at the start of the Thanksgiving dip, and at the end.  While some people may not have gone and gotten tested over the holiday, deaths probably continued apace.  And the catch-up reporting drove up the deaths number just after Thanksgiving.

Move on to the Christmas/New Year’s holiday, second oval.  Again, both series show a sharp dip as a consequence of offices being closed and data reporting being slowed.  Likely, both series show some rebound as state health departments caught up with the backlog of unrecorded cases that accumulated over the holiday.

Then, new cases peaked on 1/8/2021 and began to decline sharply.  But deaths kind-of-maybe peaked about two weeks later.  And have not yet begun to mirror the decline in new cases.

And the upshot is what you see at the right edge of the graph.  We now have the widest gap between the blue (new case) counts and deaths (orange line) ever observed in second or third waves of the U.S. COVID pandemic.

What’s going on?

Well, first, wouldn’t you like to see the count of newly hospitalized individuals?  That turns out to be a remarkably hard thing to compile, given how the states report the data.  You can get some version of new hospitalizations from The COVID Tracking Project at The Atlantic.

Source:  Cases from NY Times Github Data Repository.   New COVID hospitalizations from The COVID Tracking Project at The Atlantic.

Plus or minus some data reporting glitches (the boxy spikes early on), you can see that new hospitalizations is just about coincident with new cases.  (In theory, it should precede newly reported cases by a couple of days, but it’s all a question of the lags in hospital reporting versus the lags in test reporting.)

In the main, the new hospitalizations data drop off just as sharply as the new cases data, at the right side of the graph.  )And yet, where there’s a bump up in deaths during that declining period, there’s also a (smaller) bump up in new COVID hospital admissions.  No idea how to justify that.)

But the bottom line is that, by the time you get to the right side of the graph, there’s a big gap between the new cases and new hospitalizations data (sharply declining since early January), and the deaths data (still near the early January level).

At this point, I can only speculate on some possible causes of that.

One, maybe it’s just reporting backlog.  As things slowed down, we saw states catch up with backlogs on cases.  Maybe they’re simply catching up on backlogs (and possibly audits and reviews) of death certificate data.  Typically, death certificates would be processed through state Vital Statistics department (vital = life events = births and deaths.)  So there is a separate data processing stream for death certificates, compared to newly diagnosed cases.

If so, once that process stops, there should be a sharp decline in deaths, as reporting catches up to the current level of new cases.  The projection there is that deaths should fall off a cliff at some point.

Two, maybe with better treatment, it’s taking the average decedent longer to die?  That seems completely improbable.  But if so, the lag between new cases and deaths would increase.  Under that scenario, deaths will then begin to decline to match the slope of decline of new cases.

Three, it’s tough to come up with any explanation of this as a “real” phenomenon of more deaths per diagnosed cases.  That’s because there would be more hospitalizations preceding those deaths.  (Last I looked, place of death of COVID decedents was something like 85% in-hospital, with most of the rest being in nursing homes.)

So, it’s just one more puzzler to keep an eye on.  There is no explanation that is obvious from looking at the U.S. totals.  Plausibly, if it’s a question of data reporting, that would become evident in the state data.  If this persists, I may take a deeper dive into the state-level data.