So a little while ago, I came across a post on Nextdoor by a neighbor questioning why ICU beds are in short supply when the last times there were surges, it wasn’t this dire.
HINT: It was, but now is exceptionally bad.
So let’s start by looking at the graph he provided me.
Source: https://covid-texas.csullender.com/
First let’s breakdown what this chart tells us. Let’s note the x and y axis. The x axis is just a simple time series going from May 2020 to Now. Ok, that’s straight-forward. Now, the Y axis is a percentage of beds used which is a reasonable metric, except for one thing. See, the number of ICU beds is NOT constant. It can change from day to day based on how many nurses can work, number of doctors, mechanical issue, etc. We can see that during the previous peaks, the beds used did go up but not quite the same as we are seeing at the Aug 2021 points. Each point is a single day, so we can assume that this is an extensive data set. In this state we can see that there is a correlation between the amount of covid cases and the number of ICU beds occupied. The percentage of beds occupied by covid patients goes up, so does the total number of ICU beds being used. This makes sense as other incidents that land people in the ICU (car wrecks, strokes, and so on) don’t go away with when COVID surges happen. Now, some of that did occur when we were in lockdown as outside factors for injuries were reduced. For example, with less driving there’s less chance for anyone person to happen to experience a car crash.
At this point, many commenters assumed that the number of beds stayed constant, and I argued that this wasn’t guaranteed. As a result, I took the time to look into the ACTUAL numbers. At the very bottom of the page is a link to this individual’s github with all the data and code he used for this project.
Link: https://github.com/shiruken/covid-texas
I always start with the README file. This file should provide general descriptions of the data source, how it was transformed, and anything else of note. Read it here: https://github.com/shiruken/covid-texas#readme
In this python script, we can see what dates were manually removed. https://github.com/shiruken/covid-texas/blob/master/fetch.py
I’ll go back in a bit and take a look at these values he manually removed, what I am really concerned with is looking at the actual numbers he used, versus the percentage.
I’ll start by cloning the repository from github and reading that file into a jupyter notebook and then taking the data into a separate notebook while I reference the original code. *Note, for some reason I can’t find that notebook… lame.
Well, we can look at the column names and see what we have. The things I am interested in is the ICU beds at the statewide level. Let’s filter to just the beds at the State level.
Looking at the columns, we need to add the available beds and the occupied beds to ascertain the total number of beds. Let’s do this now and call it ‘total_icu_beds’.
Here’s what the bed count looks like for the state of Texas when looking at just the number of ICU beds we have.
Oof, it looks like there was a weird outlier for some reason last year. I’ll add that to a list to investigate at the end. Clearly, the number of beds has a significant variation over the past year ranging from ~6000 to ~9000 beds. [let’s ignore the weird outlier for now].
Ok, next we want to see the number of beds occupied and how many of those beds were occupied by covid patients and add them to the chart.
All right, so we can see that there’s the weird blip in July/August 2020 but other than that the data looks like it is expected. We find out answer in the README.txt, “All data is incomplete between July 23 to July 28 due to a transition in hospital reporting to comply with new federal requirements.” That aberration checks out and it’s going to be pursed any further.
Now the chart is still wonky since the orange and green lines should be additive. At this point they are laid atop one another. Let’s change those to area charts or stacked bar charts.
Ok, so now we have something useful for the total occupancy of ICU beds in Texas. Clearly, there’s a lag in tracking the entire number of beds for a small amount of time, but that picks up at the end of April. Again, we have that weird outlier set from the transition in hospital reporting. As we can see, the total number of beds changes overtime and it not static. The amount of those beds taken by COVID-19 patients rises significantly during surges and displace other individuals who might need ICU treatment, but again not everyone can be displaced. Here’s the real point of concern. Look at the green section at the August 2021 position. That’s horrifying. In the entire pandemic, Texas has not been this close on the margin, the entire time.
BONUS
Let’s look at some cities! Note here, the y axis on all three of these are different because the cities are all very different in size.
Then Austin
and lastly, Longview/Tyler
So, overall I think the original graphic is good, I’m not knocking this persons skills. They did a great job and had AMAZING documentation. However, it’s really easy to forget that percentages can obscure some really meaningful information.