Finally. Somebody stating what is obvious to me. From The Hill:
John Ioannidis, a Stanford epidemiologist who is famous for debunking bad research, has been pushing for it. He told me that random sampling is needed and could be done with a couple of thousand tests. When I told him that I previously worked in the polling industry, he put it in terms that resonated with me. He said, “Random representative testing is like polling. We run thousands of opinion polls in this country. We should similarly get a representative sample of the population and get them tested. It is just so easy.â€
A recent television interview with Anthony Fauci, director of the National Institute of Allergy and Infectious Diseases and member of the White House Coronavirus Task Force, underscores the need. After estimating that 100,000 to 200,000 Americans could die of the coronavirus, he said that projections are a “moving target†and that models are “only as good and as accurate as your assumptions.†But how good are models if the data is insufficient?
Ioannidis warned of a potential evidence fiasco in a recent op-ed for Stat. He wrote, “The data collected so far on how many people are infected and how the epidemic is evolving are utterly unreliable. Given the limited testing to date, some deaths and probably the vast majority of infections due to SARS-CoV-2 are being missed.â€
Eran Bendavid and Jay Bhattacharya, also professors at Stanford, echoed that concern in The Wall Street Journal, writing, “The true fatality rate is the portion of those infected who die, not the deaths from identified positive cases.†They speculate that due to how infectious the coronavirus appears to be, and because tens of thousands of people traveled from Wuhan to America in December, millions of Americans could have been infected.
Random sampling will tell us what percentage of the population has the coronavirus and its lethality. Only testing the very sick skews mortality rates and leaves us in the dark about how many Americans are unknowingly walking around asymptomatic or with mild symptoms. Looking at other countries’ data also has its challenges; age structures, climate differences, quality of health care systems and testing all vary.
New York State has administered nearly a quarter million COVID-19 tests, by far the largest number per 1M population of any state. Nearly two-thirds of its cases are in the counties surrounding New York City. Administering just a few thousand of those tests to a random sample of people in the Bronx, Queens, Manhattan, Kings County, etc. would have probably been more productive than the vast majority of those tests. Epidemiological testing is long overdue.
I have a question for physicians. In the absence of a treatment for COVID-19, how does testing a patient with the symptoms of COVID-19, change how you treat that patient? I doubt that it actually does. I think the same supportive care would be provided, as available, whether the test was positive or negative.
We have cut back on testing and just assume that if they have the symptoms they have the disease. We do test health workers and others in essential areas so that we can potentially get them back to work sooner. We are doing this largely because we are still limited on testing supplies. Would we do the same if we had adequate supplies? Probably. Why waste the money?
Steve
Isn’t there some risk tough of introducing a coinfection if the person requires hospitalization but did not actually have Covid-19?
“In the absence of a treatment for COVID-19, how does testing a patient with the symptoms of COVID-19, change how you treat that patient.”
My point for weeks. Test your ass off. If a patient presents, they will be treated as the physician sees fit.
Testing has been a red herring. You don’t need to test to know NYC is a mess, and non-compliance is the dominating variable.
If you mean we have a really sick person we thought had Covid but will miss the real diagnosis we still do the other tests if clinically indicated. Had a pt 12 days ago with a pneumovirus and Covid. If we arent sure then we still test for Covid but if any doubt they get the usual panel of tests. Where we have really stopped testing is for outpatients. If you have a fever, cough, fatigue but dont need admission we dont test. We do follow them in case either the Covid gets worse or it blossoms into something else.
Steve
That makes sense, Steve, thanks
I’ve been following John Ioannidis for a while, reading and referring to one of his earlier sensible columns. Supposedly, he was enlisted to join the CV Task Force too. However, his presence, if there, has not been visible like that of Doctors Fauci and Birx. Random testing, though, has always seemed like it would provide a helpful data point, in the midst of all the other data collected from a myriad of models, some which have proven to be highly inaccurate, doing nothing more than fueling panic.
I also think CS brings up a good point about the liabilities of indiscriminately mixing non-infected patients in with those who are infected with the CV.
Drew- Not testing when the incidence is probably high makes some sense. When you don’t know the incidence, you need to test. If we had the ability to test as early as other countries we could have known that the rate of infection was increasing in NYC much earlier and locked down sooner. Same for New Orleans. Same for everywhere. Early on it would also have been useful clinically if tests had been available. We had at least one pt who tested positive for a severe viral pneumonia that we did not test for Covid since we had so few tests and thought we had another explanation. If had testing had been available we would have tested and a bunch of our staff would not have had to spend a couple weeks at home not touching their kids.
Sampling sounds good and it is a common method for handling issues that affect large populations. However, I would point out this method has been questioned and mocked by those living in the right wing sewers online.
” a myriad of models, some which have proven to be highly inaccurate, ”
Which ones were those?
Steve
As I understand it, the CDC uses sample testing to determine the extent of the annual flu epidemic – the annual numbers are not an actual count of cases, but the result of a model backed by sampling data.
The same could be done for Covid assuming we can get a good random sample technique and a skillful and validated model. How long will that take?
Sure have noticed the large number of celebrities , A-listers, and other elites, including royalty, positive for co-19. Maybe they do get around more, but I have another theory. They have no problem getting tested.
Looking more to me if we could take an infection snapshot at any moment in time, there are most likely millions already infected. That’s not necessarily bad news because it would mean most infections aren’t severe.
“How long will it takeâ€
Extremely quickly if someone actually decides to do it.
The mathematical model / sampling techniques all exist — they are employed already for example to determine how often / what to sample for diseased livestock imports, impurities in drug imports. And if we wanted to improve or create a new model; some of history’s best mathematicians work in this country — I think they would love to be useful.
The number of tests needed is small — 180 tests were enough to convince San Francisco to lock down. The US is running 140000 / day.
Another example, in Italy, they tested the blood of 60 blood donors in one of the hard hit towns, 40 came back with sign of past infection which is enough data to suggest the town has herd immunity.
Give the man a cee-gar.
Looks like the CDC is getting on it.
https://www.statnews.com/2020/04/04/cdc-launches-studies-to-get-more-precise-count-of-undetected-covid-19-cases/
On the soapbox – why is the CDC so slow?
By the summer — some projections have this wave of outbreak over by the summer.
To ask the question is to answer it. It is a bureaucracy. Most of its leaders are career bureaucrats, in their 60s and 70s. The words “bureaucracy” and “urgency” are at the opposite ends of the dictionary.
Would like to see it done to see how high the rate of infection peaks at, but in terms of management probably more important to monitor on way down.
Steve
The most highly touted model, early in the game, was the Imperial College London. It predicted the possibility of 2.2 million deaths here and 500,000 in the U.K. That has since been walked back, absent in the deluge of media chatter, and replaced by the IMHE from the University of WA, which is predominately referred to by Fauci & Birx. As of 4/5, that model for NY offered a projected number of 69K hospital beds, while the actual number needed is 16.5K. ICU occupancy was projected to be 12,346, whereas actual number is 4,376. Today’s updates for that area are reporting discharges are outpacing hospitalizations, with Cuomo positing that the apex might have been reached.
That’s why I think models predicting future numbers should be viewed gingerly, with equal amounts of respect and skepticism.
A quote derived from the popular X Files series might be applicable to today ——> “Control the disease by controlling the information.â€
The 2.2 million number assumed that there was no attempt at mitigation. When numbers in the UK started to climb they abandoned their laissez faire approach and went to mitigation like most everyone else. They then came out with a prediction that roughly matches what the Trump admin is using.
The concept behind most of the models is not that complicated. We have good data on looking at how many people are hospitalized, what percentage of those will need ICU and what percentage of those will need a vent. How long on average a pt will be on the vent. What we have poor data on is penetrance into the community. That is where the epidemiological testing Dave keeps talking about would help. However we didnt have the testing available when we needed it so we are extrapolating from what is happening elsewhere.
” while the actual number needed is 16.5K. ICU occupancy was projected to be 12,346, whereas actual number is 4,376. ”
So you have inside information that the current numbers are going to be our maximum?
Steve
Nate Silver wrote an article that explains why running a lot of tests may not be giving much useful information on the course of the epidemic.
https://fivethirtyeight.com/features/coronavirus-case-counts-are-meaningless/
I’ve lost track of where I was finding the info, but as of a few days ago Florida was getting positive results from a little under 11% of tests. The population being tested are mostly people that have some reason to believe they have been exposed, or have symptoms. There is also probably a segment getting tested and re-tested. So from that it is hard to believe that a high percentage of people in Florida already have the virus. Just an FYI.
Here’s the data on testing in Florida. Running at ~10.8% positive at this point.