State of Georgia likely undercounting COVID19 Deaths by around 30%

Bold headline – better have some numbers to back it up. OK: Two different calculations ended up with the same conclusion that the US State of Georgia, which yesterday decided to start relaxing mitigation guidelines, is very likely undercounting or under-reporting COVID-19 deaths by on the order of 30%.  Let’s see why this outrageous statement was made. I will keep the math and statistics light here; may try to do another post with the detailed calculations for those who are curious when I get more time to write it up.

First, a disclaimer: the data for the State of Georgia is problematic (wretched, abysmal, horrific, consult your thesaurus and pick your favorite adjective).  Consider:  For the week ending April 4th (that was now over two weeks ago), Georgia has only managed to report 66.8% of its death data to the NCHS Pneumonia and Influenza surveillance system.  Three weeks ago, that number is up to 87.7%.  I’m doing comparisons with the state of Washington in this analysis, since the claimed curves seem similar, and the populations are of similar size (7.8 Million for Washington, 10 million for Georgia).  Now, that’s really unfair, because by most health care quality of life metrics, Washington State should be “better”, but on the other hand, Georgia’s CFR is 3.993% vs Washington’s 5.308%, and Georgia says it’s ready to re-open whereas Washington is being more cautious.

Oh, and what are the Washington reporting statistics?  100% for the week of March 28.  100% for the week of April 4th.  Hmmm …

So, with that background, lets try to back out what Georgia’s numbers should be.

Method 1:  Case Fatality Rates

As noted, I really don’t like CFR’s. They are really noisy given the testing environment. But sometimes it’s all you have.   At similar points in the curve (Washington State should be “ahead” of Georgia by at least 10 days), we should be seeing similar CFR’s, all things being equal (and they probably aren’t, given difference in demographics, aggressive early action, etc, all of which are in Washington State’s favor, so this give a huge “advantage” to Georgia).  The Washington CFR is 5.0% at the point in the curve where Georgia is today, yet Georgia is reporting only 3.9%, or a 28% lower rate.  Applying that to the most recent report, Georgia should be showing 994 deaths rather than the 775 it is in fact reporting.  If we use the smoothed end point CFR, it’s even larger, 1030 vs 775 (32.9%).

Method 2: Pneumonia Death Rates

This method is a bit more complex.  It involves analyzing the NCHS Pneumonia and Influenza (P&I) death surveillance system and computing “excess” mortality.  A key problem is that Georgia takes a long time to report death data under normal circumstances.  COVID-19 is moving so fast, using data from a couple of weeks ago is making a noisy situation even more noisy. But, with that yuge caveat, we can dive in.  I’ll spare you a lot of math here, but the bottom line is adjusting for where we are in the season, the reported P&I and Influenza deaths, etc. the Washington State data matches their reported COVID19 deaths pretty well – 127 reported, 129 predicted.  Georgia dosen’t.  The reported is 132, the predicted 170 – a 29% undercount.

Conclusion and Opinion

It is widely suspected that the excess mortality from COVID-19 has been underestimated.  When you dig in to cases and fatalities for the US state of Georgia, they just seem anomalous, even given the normal noise in these kinds of data sets. In comparison to other states and countries, the numbers just don’t add up.  Even so, based on that data, there is no evidence that Georgia has reached a point in the curve where mitigation measures should be reassessed.  Testing is still problematic, and the reporting is highly questionable (likely a result of both the well publicized problems in testing and long standing delays in public heath reporting statistics).  Maybe things are better than the data shows.  But there is no way to know that – it’s just guessing and hope.  But Hope isn’t a Plan. The Governor is doing a massive experiment with the lives of the people of the state.  I hope he’s right; I’m afraid he’s wrong.

Data sources:

8 thoughts on “State of Georgia likely undercounting COVID19 Deaths by around 30%

  1. Do you have the data the governor used to make the decision to re-open businesses? What I see reported on the news is a steady increase in the number of cases and deaths, not a decrease.

    • I really don’t know what “data” they are using. The official data sources are what I am using, so I assume it’s the same. I think it’s a matter of interpretation and priorities.

  2. I have to believe the government has more information than publicly available. Didn’t you just blog some anecdotal information regarding local hospital preparedness? It’s more than just data. They probably have data AND communications that made them comfortable making this big decision. The initial goal was to flatten the curve – not put it at Y=0 intercept. You’ve been right before on data driven things but this is more than data to me.

    • Unfortunately, that’s rarely the case (that government has more information). I’ve worked in and around government my whole professional life. In general, while it often has access to raw data the public does not, in most cases a lot of that data is noise and requires interpretation. And that is where the problem comes in. In recent years (starting in the mid 1990’s, which I guess isn’t all that recent any more!), the politicization of interpretation and analysis started creeping lower and lower into government. And availability of “raw” data started creeping upward. These two factors have created a situation where people who are biased, and/or don’t really understand all of the nuances of the raw data and interpret it through the filters of their political beliefs rather than an objective analysis. It’s ok to have different plans, rooted in politics, based on solid information derived from the raw data. It’s not ok to create plans based on politically driven filtering of the raw data. Yet that’s what both political parties in the US have become experts in doing.

    • Never a stupid question, only a stupid answer; I’ll try to avoid that!

      CFR is the Case Fatality Rate. It is the number of deaths divided by the number of “cases” of a particular condition. In this situation, it’s an almost worthless number because we really don’t have a good idea of the number of “cases” – in other words, how many people are really infected with the SARS-COV-2 virus, because apparently a huge number of people have either very mild or non-existent symptoms.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.