Post #1362: The CDC’s sharply revised Omicron numbers.

Posted on December 21, 2021

The CDC’s revised Omicron numbers.

I’m an economist by trade.  Economists take a lot of criticism for routinely making bad predictions.  But we can’t hold a candle to epidemiologists in that regard.

The CDC’s announcement yesterday — that Omicron already accounts for more than 73% of U.S. cases — was unexpected.  By about a factor of four.  (I was expecting something around 12%, based on last weeks announcement).

In this section, let me try to work through all of that.

1:  What is that number, really?  What is Nowcast?

First, the 73%-Omicron pronouncement is from the CDC’s “Nowcast” model.  It’s a prediction, not a measurement. 

I don’t think I have ever seen a complete writeup of what, exactly, CDC does to make that “Nowcast” prediction.  Here’s what I have pieced together.

Nowcast makes a projection for the most recent past week, using data that are from a much earlier time period, and an estimated rate of growth.  In the case of these variants, CDC variously says that it takes between two and three weeks for the data to come in.  So CDC is projecting the current situation, based on what they think they observed for level and growth of Omicron roughly two weeks earlier.

You can clearly see that if you go to the CDC COVID data tracker website and turn off Nowcast.

Nowcast off, most recent data is week ending 12/4/2021. Omicron was estimated to be about 0.7% of all new cases that week.  This is the actual data.


Nowcast on, CDC projects the two-week old data to the week ending 12/18/2021.  Omicron was projected to be 73% of all new cases in that week.  This is the CDC’s forward projection of the actual data.

When something is growing that fast, even a modest mis-estimate of the growth rate, compounded over two weeks, can lead to substantial error in the projection.  I’m fairly sure the CDC understands that.  In any case, you can see the column labeled “95%PI”.  I’m pretty sure that flags that as an estimate of the overall error of the entire data-and-model framework.  (That is, it’s not just a standard statistical sampling error of the underlying data.)

In any case, note the range there.  What the CDC actually said yesterday is that they are fairly sure that, as of last week, the Omicron accounted for somewhere between 34% and 95% of all new U.S. COVID-19 cases.  This is not a slam on the CDC, it’s just an acknowledgement of the wide confidence interval associated with that 73% projection.

It’s 73%, plus or minus.  Plus or minus quite a bit.

In addition, the CDC revises the historical data as more tests get counted.  That appears to have been a primary driver of the change between last week’s estimate of 2.9%, and this week’s estimate of 73%.  Between the two weeks, CDC revised last weeks estimate upward to about 13%.  So what we thought was 2.9%, going into this latest revision, is now said to be 13%.

2:  Count your time lags as carefully as you can.

Now let’s assume that the CDC’s 73% estimate is perfectly accurate.  (We might as well, because as far as I can tell from press reporting, just about everybody else does).

What does that imply for the rate today (12/21/2021).  And is this in sync with the data reporting for the count of cases?

Today’s rate.  The CDC Nowcast is an average for the week ending 12/18/2021.  Given the rapid rate of growth, my best guess is that CDC’s projection actually hit that level around 12/15/2021.

So as of today, it’s six days after the U.S. passed that 73% mark.  Nowcast shows the percentage increasing more than five-fold in the past week.  And while I can’t do the detailed math, I’m certain that puts the estimate for today well into the high 90%s.

In other words, based on the CDC’s projection for last week, and the growth rate, the CDC’s 73% estimate from last week means that virtually 100% of cases tested today should show up as Omicron.

There is a caveat.  I’m pretty sure the CDC is tabulating all of that by sample (test) date.  That’s the only firm date that can attach to the set of swabs they used for their full genetic breakdowns of the virus.

Roughly a week elapses between test date and the cases are counted fully in the seven-day moving average of new cases.  Best guess, about four days elapses, on average, between the time you get tested, and the time that test is reported by your state public health agency.  The raw daily counts should run about four days behind the test dates.  And then we take a seven-day moving average, which in a crude sense amounts to an average 3.5 days additional time lag.  The upshot of all of that is that the seven-day moving average of new cases ought to be about a week behind the CDC data.

And so, as far as that goes, it’s a case of very nearly offsetting errors.  The CDC’s 73% really is for swabs that were taken six days ago, on average, or circa 12/16/2021.  Today’s new case count (for 12/20/2021 data) comes from swabs that were taken circa 12/13/2021.  And so there should be something like a three-day difference between the CDC’s 73%, and whatever is driving today’s increase in cases.

Given the rapid rate of growth, this means that, even if the CDC 73% is absolutely correct, today’s seven-day moving average of new cases actually reflects a substantially lower percentage.  Given the growth rate, plausibly half that or less.

So the seven-day average of new case growth reported on 12/20/2021 (today’s data) reflects something like 40-ish percent Omicron in circulation.  If the CDC’s 73% estimate is correct.

It’ll all be moot in a week anyway, given the growth rate.

3:  Does the regional variation in Omicron match the regional change in new case growth?

Near as I can tell, the match between where CDC says Omicron now dominates, and where the rapid case growth is, is only so-so.  First, here’s the CDC’s chart, via Nowcast:

Source:  CDC COVID data tracker, accessed 12/21/2021.

Based on the CDC estimates, Region 10 (Pacific Northwest, mostly) should be hardest hit by the Omicron wave.  Followed by Region 2 (mid-Atlantic) and Region 4 (South Atlantic)

Now case growth, based the same NY Times data source as above, just split into the 10 U.S. DHHS regions.

The only thing that really stands out is Region 2, the mid-Atlantic region.  Region 10 has seen more-or-less no excess case growth in recent weeks.

I judge that to be a mediocre correlation at best.  Perhaps the estimates of the Omicron share are that bad.  Perhaps other factors (e.g., weather) are driving case growth.

4:  Note if the CDC is right, most of what Omicron has done so far is to displace Delta.

I remain fascinated by this phenomenon.  Once again, the main impact of the newer, more-easily-spread variant is to kill off the prior variant.  And spread a bit faster.  How, exactly, Omicron manages to kill of Delta, I cannot quite fathom.  But this is just a repeat of what Delta did to Alpha, and what Alpha did to the native (Wuhan) strain in the U.S.

I still remain hopeful that case growth will remain low enough, and top out early enough, that Omicron will be a net bonus for the U.S.  As of the last time I checked, more than one percent of diagnosed Delta cases were still ending in death in the U.S.  More than 6% were ending up hospitalized.  As long as Omicron’s case rates are well below that — and I believe that, on balance, the data pretty strongly suggest that — I think the Omicron wave in the U.S. will end up reducing the harm from COVID, compared to letting a Delta-based winter wave play out.