How much double counting of cases is going on?
Several of the doctors on this list have told me that when a patient is hospitalized for COVID, they are generally tested multiple times for COVID during their stay, sometimes as many as 5 positive tests before testing negative. I was curious about this, so spent the day researching various states in an attempt to figure out if these multiple tests were reported as multiple cases (even though it’s only a single individual) or whether the reporting is sophisticated enough to de-duplicate these tests so we don’t exaggerate the number of cases.
First, I found out pretty quickly that it’s not easy to get this information. I did find at least 13 states that report positive tests, rather than positive people. My confidence in my findings ranges from around 75-100%, so don’t take this to the bank, but I think these states are double counting cases: AL, AR, CA, GA, ME, MD, MA, MI, OK, PR, TX, VA, and WY. There may be many others, but I can’t verify that. In addition, I found 4 states (AR, PR, TX, and VA) that report positive anti-body tests as cases. Texas doesn’t do this entirely, as the public labs in Texas attempt to de-dup, but the public labs represent a small percentage of tests.
I’m convinced now, that as the volume of testing grows, so does the double counting. We currently have at least 40,000 people hospitalized for COVID currently (we don’t know the actual number, as many states don’t report this, or report it based on surveys only). If the average number of positive tests is 4 (I’m guessing from anecdotal evidence here), then that could mean as many as 10,000 to 20,000 extra cases per week if all states are doing this. Since I’m currently modeling less than 160,000 active cases in the U.S., this is a big distortion.
Antibody test reporting could be even more problematic. These tests (often referred to as Serum tests) are intended to reveal who was infected with COVID in the past, but now recovered. When these positive tests are reported as new cases, the new case count data is corrupted. The media is very interested in showing that we undercount cases and deaths, but has no interest in the distortions that cause over counting, so you’ll likely not read much about this.
I don’t think any of this is planned or intentional, as it’s probably not easy to track individuals verses tests, and I don’t think many foresaw the multiple tests per individual when testing was scarce. However, I’m now thinking that my model of active reported cases is an upper bound, and the actual number is probably 5-10% lower.
In any event, I don’t try to correct for any of these data distortions, simply because I don’t have reliable data to do so. I think it is important that the CDC quickly develop and enforce consistent protocols for test reporting. It’s hard to know how the disease will progress in the future if we don’t even know what it is doing today. Public response relies on whether the disease is growing or declining, so I think solid data is critical. I’d give the CDC a mediocre grade at best on this issue.
As always, feel free to send me your questions about my assumptions, methodology, or modeling in general.
- Likely date of active case peak (Chalke modeling): April 10
- Likely date of peak deaths (IHME): April 16 (last revision on May 20)
- Short term projection for active cases tomorrow: 154,000
- Total Test Results reported today: 342,418 (high)
- Total Pending tests reported today: 3,709 (very low)
- National reported case Growth Rate today: 1.5% (very low)
Shane Chalke Interviews
Here is the national picture of active cases – small decline today but relatively flat for a week now. We’re down 28% from the peak.
Here are the new reported cases nationally. Long, slow decline. No new news here. It could be lower – see discussion above.
Here are the daily reported tests. We’re reporting a million tests every 3 days.
On to the states. Both VA and NC are in some stage of reopening (actually all 50 states are now). North Carolina is moved on to Phase 2 today, while VA maintains heavier restrictions. Looking at the 2 states side by side, you can see about 50% more COVID in VA, even though VA has only 85% of the population of NC. IHME updated their model again yesterday, and projected peak deaths in NC is still June 6th. If that’s the case, then I expect peak active cases about 2 weeks earlier, or tomorrow.
No surprises with Washington, now down about 61% from their peak on April 3rd.
Florida is worth keeping an eye on. It has been rising for 7 days now, but still small for the size of the state.
I’ve added NJ and combined it with New York. NY is now down 82% from the peak, which is remarkable. NJ had a long plateau at the top, and is now down 66% from the peak. Surprisingly, I expect that NJ will have more COVID than NY in a week or so. The two states look to be converging.
California is drifting upwards, but they are one of the states that reports each positive test as a new case, so caveats here. Not much COVID presence in the state outside of Los Angeles.
Massachusetts has had a very rapid recovery, now down 59% from the peak on April 27th. The northeastern states were hit harder than most, but are recovering faster as well.
Georgia is flat today, still on a long, gradually decline.
Another drop in Michigan today. Michigan is down 52% from the peak on April 6th.
Pennsylvania continues on trend – steady declines since the peak on April 11.
Here is Texas, with another significant drop today. We may have hit the peak here on May 19th. Texas is already down 30% from that high point in just 3 days. Texas is another state with a small COVID presence per capita.
And finally, here is Colorado. Last Friday, Colorado reduced their cumulative death count by 24%. https://www.foxnews.com/us/colorado-lowers-coronavirus-death-count They were coding deaths as COVID where COVID was present, but where there was clearly another cause of death. They revised the COVID deaths downward, but then added another category called ”deaths among people with COVID-19”, and continue to report these as COVID deaths on their website. This was a red flag to me, so I thought it would be interesting to see how they are progressing, and if we see an unusual pattern it might be cause for further research. In any event, it looks like Colorado is about 52% below the peak on April 29th.
So that’s it for today. Please remember that although the number of active cases is declining, it’s declining slowly (some areas quite fast, though). We’re on the backside of the curve, in some states still near the top, where the probabilty of coming in contact with an infected person is relatively higher than it will be in another week or two. Everyone be careful.
–Shane Chalke, FSA