Florida Isn't Hiding COVID Deaths in Pneumonia Numbers (In Excruciating Detail)
TOO MANY CHARTS!
I’m a chart guy and I think I’ve gone overboard with the charts on this one. If this seems long, know that I’ve cut far more than I’ve written. But everything is just so FASCINATING!
The Conspiracy That Wasn’t - what the Florida “pneumonia” conspiracy is all about
What is Actually Happening? - wherein we try to give the doubters the benefit of the doubt
COVID (hearts) Pneumonia - a deeper dive into how the CDC charts things
Deciphering the Code - an even deeper dive into CDC coding requirements and how that messes with the data
Subscribe for more charts and more words. I have all the best words.
The Conspiracy That Wasn’t
If you are extremely unlucky, you have been exposed to the idea that Florida has been hiding their COVID data and/or mis-classifying COVID deaths as pneumonia deaths.
Here is, in a really heart-breaking and disgraceful performance, self-described “Obama health care head” Andy Slavitt spreading this concept.
There are a couple of possible interpretations to this and I want to spell them out so we can be sure which ones we’re addressing.
Florida is intentionally and maliciously mis-classifying COVID deaths as pneumonia deaths.
Florida is accidentally mis-classifying COVID deaths as pneumonia deaths.
One of these is obviously worse than the other, but if either were true we would have an inaccurate assessment of the true state of COVID in Florida.
What is Actually Happening?
Let’s start with the available data. We’ll look at pneumonia data and COVID data and search for patterns. We will then dig into detail about how the details of medical coding (how doctors code patient deaths before they are reported to the state or the CDC) impact the data we see .
You can review the CDC’s “Pneumonia and Influenza Mortality” over here. This is the data set that most people are pointing to when they say there is a “pneumonia crisis”. We’ll look at the first 20 weeks of this year.
I want to start by trying to understand how someone could honestly believe that Florida is “hiding” their COVID numbers as pneumonia deaths. We must try to treat this view with sympathy and honesty. They see this chart of Florida pneumonia deaths and they see the obvious stick out.
Holy cow! Something obviously happened in 2020. It’s obviously COVID that is driving this surge in pneumonia deaths.
Right?
Well. Yes. And we see this in every state. Here is New York, Georgia, New Jersey, Massachusetts, Michigan, and Washington all showing the same pattern.
So let’s assume for the sake of charity the the people spreading this rumor only checked Florida and forgot to check any other state. That’s an unfortunate reality in the media atmosphere in which we live. This should be enough to show us that Florida isn’t maliciously hiding anything… they are seeing the same pattern of increased pneumonia deaths that every state is seeing.
If that’s enough for you, feel free to skip the rest of this. If you’re more curious about the details of what is *really* happening with COVID and pneumonia numbers, let’s ask some more questions.
Why do we see this surge in pneumonia? Are we missing COVID cases in this crush of pneumonia cases? What is going on?
COVID (hearts) Pneumonia
Pneumonia is not really a specific disease per se. It’s an inflammation of the lungs caused by influenza, bird flu, swine flu, SARS, MERS, rheumatic fever, or dozens of other conditions. It could be due to bacteria, amoebas, viruses, fungi, or parasites. Basically, pneumonia means something went wrong with your lungs.
COVID can cause pneumonia. The charts above may contain cases that are both COVID *and* pneumonia. I really wish there was a way to sort out pneumonia deaths involving COVID-positive patients.
But wait! There is!
The CDC publishes exactly this information, which is unfortunately hidden to Twitter pundits who aren’t careful enough to go looking for it before jumping to conclusions.
In the data at the link, the CDC has a column specifically titled “Pneumonia and COVID-19 Deaths”. Most of these COVID pneumonia deaths are included in the charts above (though not all of them for complicated reasons that I will explain in the next section).
So what happens if we subtract the COVID/pneumonia deaths from the pneumonia deaths above? That should give us an more clear indication of states that have been missing COVID deaths by mis-identifying them as simple pneumonia. Well.. let’s try that.
We do see, on average, an increase in pneumonia deaths. But Florida is not an outlier in this. Everyone saw a surge in non-COVID pneumonia cases, but in the past few weeks most states have mostly managed to appropriately identify the COVID cases. The real outlier is New Jersey, which has clearly missed a substantial number of COVID cases as their pneumonia cases surged to more than twice a normal historic number.
So how could states miss COVID pneumonia cases? We need to dig a step deeper, might be better to just bail now.
But if you’re still here, let’s jump into the data slough.
Deciphering The Code
The first question that popped into my mind is why, when we remove COVID pneumonia from all pneumonia deaths, pneumonia deaths are actually *below* average. Among the states I checked (I have not checked them all) we see this in New York, Georgia, and Washington.
The embarrassing answer is that 1) medical coding is incredibly hard 2) the CDC has been playing catch-up on this thing for months 3) there is not a consistent answer or pattern so we have to make some guesses.
Here is something I know for certain. On February 20th, the CDC offered guidance for COVID-19 coding that suggested COVID-based pneumonia be coded as “J12.89, Other viral pneumonia”. This is a classification that would include COVID-based pneumonia into the initial pneumonia charts above.
This means that even if a person was positively identified as a COVID patient, if they died from lung inflammation, they would be included in the “pneumonia” chart.
But starting April 1st, the CDC changed their coding guidance to say “screw that, we’ve made a new code for COVID, use U07.1 but don’t forget to use J12.89 if they also have pneumonia”.
Starting in April, properly coded COVID-pneumonia deaths should be included as both COVID and pneumonia and should be tracked in both data sets. The CDC is trying to create this distinction in order to maintain cleaner data sets. This is a laudable goal. Some hospitals jumped on this re-classification and some did not.
All these things are things I know to be true.
Here is something I’m guessing at. In this chart, it looks like New York jumps on board with this classification and starts classifying 90% of all pneumatic patients as COVID-pneumonia, to the point where non-COVID pneumonia is actually at a 5 year low on April 11th (the first full week this coding recommendation was in place).
My guess is that other states and hospitals have not been as quick to adopt the updated CDC coding guidelines (or perhaps they aren’t able to confirm a COVID diagnosis), which means that we don’t see this pattern robustly across all states. It seems likely that states like Georgia and Washington are seeing non-COVID pneumonia deaths at more typical historic levels in part because they have adapted their medical coding to the new requirements over the last 6 weeks.
This is what I’m guessing at from looking at the data. I have only interviewed a few doctors, medical coders, and medical data experts, all of whom took time to impress upon me the messiness of this data and the very real possibility of innocent human error both on the individual and institutional level.
Ultimately what we need to recognize is that we’re dealing with an incredibly complex and evolving situation and that a single surge from a single state does not appropriately translate into a conspiracy theory of hidden data and evil intent.
There is no conspiracy. This is the most boring part of data collection, normally hidden from view. It’s weird and embarrassing when we data types realize that we need to re-align our metrics in the middle of a multi-week collection. We try to course-correct, but we can’t change the past so we just do what we can with what we have.
We would like to apologize to everyone who thinks that what we do is instantly perfect and that bumps in the data represent a hidden clue to the true nature of reality. They do not. Bumps in the data more often represent people messing up and trying to get it right while our hair is on fire and our bosses are scowling over our shoulders.
Disney Classics: The China Shop
This is not available on Disney+ at this moment, but I am utterly in love with this short. It begins with an elderly shop owner setting his china display as he leaves for the night.
As he locks the door, the china figures come alive in a sort of Toy Story scenario. The main characters to this are two classic lord-and-lady china figures. Their dance across a mirror is one of the earliest examples that I’ve seen of mirror animation and it’s really lovely.
When the lady is kidnapped and imprisoned by a devil china figure, the lord springs into action. We end up with a perfect distillation of early cinematic melodrama with the evil threat confronted by the hero and suitably handled, the lady love rescued, and everyone made whole in the end.
The music is fantastic and the character animation is near to perfect. I will never tire of this short.