This past week has been another doozy for the state of Georgia and how they report data. In late april the big debacle was in the new case counts and how misleading it was.
Then the issue surrounded a non chronological graph showing case loads by county that was super confusing to...well...everyone. (we didn't write about it because we didn't think it was *that* big of a deal)
Now we have a testing issue. We'll try and explain for the casual bystander:
Right now there are two primary tests that are being conducted for COVID-19:
Want more information? The CDC has a good section on this.
Now that we know about the two kinds of tests, we can explain what happened this week. It was revealed that the big number on the Georgia Department of Public Health website that lists "total tests" includes both antibody tests and viral tests. Prior to this week it was generally understood by the epi and data community that this number only included viral tests.
This week we learned that starting in mid April, the state started lumping antibody tests in with the overall number. You can almost see exactly when it started to happen. Roughly around April 19th, the number of tests conducted seemed to grow by 25-50%. The community rejoiced thinking this was the much called for expansion of testing. However, it's now clear this was just testing for who might have had it in the past but no longer has the virus.
One of the more important metrics that we care about is percent positive. How many people have been tested are positive. WHO says that number should be less than 5% over the previous 14 day period. It's a simple enough number to calculate. How many people did you test is the denominator. And the number of people who came back positive is the numerator.
The challenge here is that if you're lumping two testing methodologies together, your denominator will always be inflated.
Here's another way to look at it. You have two high school classes: one math, one science. The math class has 50 students and the science class has 25 students. Each class takes a test and you want to know how many kids in the math class pass the test. Easy, you take # of students who get a 70 or better and divide by 50. Let's say 40 of the 50 get a passing grade so 80% of the class passes, right?
Well if you were the state of Georgia, instead of diving by 50 (the number of students in the math class), you would divide by 75 (the total number of students). Now your pass rate went from 80% to 53%.
Conflating these two numbers is "ok" by CDC standards (in fact, they are the ones who started doing it this way) but the epi and data community largely agree that these two numbers should be separated for clarity. As long as these two numbers are lumped together, we cannot accurately
A lot of people are removing their testing numbers from their own dashboards. And I understand why: we don't currently know what the testing number *is*. However, our objective from the very beginning is to graphically represent and map the state's data. This means we don't change it. Or alter it. We just...present it. Flaws and all. So we're going to leave the testing number on our site. When the state "fixes" their data, we'll fix ours as well.
However, we are removing the percent positive per day and the overall percent positive graphs there were at the bottom of our dashboard. These were interpretations of the state data and in fairness, should not have been on our site from the beginning. We have a spot for "theoretical" dashboards and that's probably where we should have put these measurements. So they are removed.
We look forward to seeing the state break out the antibody and viral tests to make it clearer to the public what's happening with COVID-19.
