“There are three kinds of lies: lies, damned lies, and statistics.“
Now that the video is back up, Part 2 is in progress.
Yesterday a friend sent me the following video and asked two things; would I write about it, and would I try to make it short! The second skill is not really in my wheelhouse, and it is a very, very long video, clocking in at 52 minutes; I am currently writing a 2 part essay on a video that is less than 5 minutes long.
I’ve chosen the “live tweet” format (I don’t know what else to call it) in order to keep my comments brief and in-line, chronologically, with the video itself; I am sure I will have some additional closing remarks, however.
While most of what I try to address on this blog falls into the first two categories of ‘lies’ and ‘damned lies’, Dr. Erickson’s analysis belongs primarily to the final category. Dishonest statistics are extremely difficult to dispel because those who don’t have a background or training in interpreting them are apt to chalk up disagreements to a mere difference of opinion about what the numbers mean. They are often right. However, in this case Dr. Erickson is actually creating false statistics out of thin air, and then framing his arguments with these imaginary numbers.
Edit 4/28/2020: The video is available again here: https://www.facebook.com/watch/?v=537566680274166
(Note on time: with the original video removed from youtube, these time stamps are going to be a bit off. The facebook video above is about 12 seconds ahead of the original video; so 0:22 becomes 0:10, 0:27 becomes 0:15, etc. Sorry for the inconvenience.)
0:22 Kern County California.
0:27 This is my first yellow flag; “ER Physician/Entrepreneur perspective.” Most doctors wouldn’t describe themselves in that terminology even if they run their own practice, so I’m listening very carefully for what the “entrepreneur” angle is.
Over and over again with these misinformation videos, we have seen that the creating of false information has some direct link to attainment of money, power, or fame for the person in the video.
0:45 “If that still makes sense.” This is the question on every person’s mind, and rightfully so. For medical people, clinicians and nurses, it’s a definitive and resounding “yes,” so I’m interested to hear his perspective.
1:00 Already this video is different from most of what’s going around, because these guys are actual doctors.
1:34 Here we reach the “entrepreneur” piece; my understanding is that Dr. Erickson is an owner or partner of Accelerated Urgent Care, a group of 5 Urgent Care centers around Bakersfield CA.
Two things about this: First, we do need to recognize that while Urgent Care centers can and do provide services that help take the pressure off of over-utilized hospital emergency departments, they are NOT emergency rooms, and so unless Dr. Erickson is also working in a hospital context it is not quite accurate to treat him as a practicing ER Physician; he is likely ER trained, but not currently working in that context.
Second, Urgent Care centers are indeed entrepreneurial ventures; they are for profit, like so many fixtures of our broken healthcare system. During this entire video we are going to have to ask ourselves how the pandemic is affecting his business, and how that is implicitly affecting his understanding of the situation and statistics.
1:44 See above.
1:58 I don’t know what “furloughing patients” means, but otherwise this is the exact situation in Waco; we’ll get into this in more detail later because I think it’s an important topic.
One note for now; do not fall into the trap of thinking that “empty ICU’s” means that the pandemic is not real. Cancelled elective cases and alternative delivery of care is part of containment measures in areas where COVID-19 has not yet surged, like Waco or Kern County California. The worst is yet to come.
2:03 Make note of this. Everything else that is said in this video needs to be understood in the context that even Dr. Erickson recognizes that this virus can overwhelm healthcare infrastructures; it’s doing it in New York right now.
2:30-3:02 He’s absolutely right, in a way. As I’ve written before, every single clinic I know of is working hard to make sure that their patients with chronic medical and mental health needs are still receiving the best care possible under the circumstances.
But there is another side to ‘secondary effects’ of COVID-19 as it relates to chronic conditions, and it’s this; as deadly as this virus is for people with the very conditions he is listing (in other words, their fear or caution is not unfounded), an overwhelmed healthcare system is also dangerous even apart from the virus. When patients who have heart failure or diabetes, or depression, or any other medical or mental health condition cannot get care because the healthcare system is overwhelmed with a pandemic, that is no less dangerous than not getting seen for other reasons; and probably much more dangerous in many cases because at least with the ‘minimum capacity’ healthcare usage he is discussing they could still get timely treatment in a true emergency, which is not a guarantee when the local ER’s are overwhelmed. These are difficult decisions that every clinic, hospital, and system is weighing carefully; and the quality of that decision making depends on reliable COVID-19 data, as we will see shortly.
One more note; this absolutely is being talked about, and extensively. Don’t fall for the “why are the higher ups keeping quiet” argument about very complex medical systems and situations; these conversations are being had on every level and have been for months (I have yet another Zoom meeting this afternoon about this very issue).
3:17 I think this is a really misleading way to frame the amount of data we had 1-2 months ago, and at the beginning of our social/physical distancing measures. Cases began to rise outside of China in early to mid February, and We already had 100,000 confirmed cases worldwide by March 7th. It was officially declared a pandemic on March 11th. So those (not) early (enough) decisions to begin social (physical) distancing measures were made based on data, not in the absence of it.
3:33-3:50 This is a false equivalence, and actually rather silly. What would it look like to quarantine the healthy because of ‘normal’ infectious diseases? “Sorry Billy, no school today; somebody at your school has pink eye so everyone is staying home.” “We can’t go to Church today kids; the pastor’s daughter had Hand, Foot, and Mouth Disease.” Pretty ridiculous, right?
But our template for COVID-19 is not pink eye, or strep throat, or even the seasonal flu; it is the 1918 Spanish Flu pandemic, smallpox, and the freaking Black Death. He is acting as though he didn’t study these diseases and periods of history in pre-med and Medical School.
In a Pandemic, social (physical) distancing, what he is calling ‘quarantining the healthy’, absolutely saves lives. If you don’t believe me, read this article. Or go play the Plague, Inc flash game and try not to throw your phone across the room when Madagascar shuts down it’s seaports.
4:21 I didn’t realize what he was trying to say here right at first, but it’s worth pointing it out here instead of 10 minutes later when it finally hit me, since this is actually his main thesis throughout this video.
- Kern County:
- People tested: 5,213, Positive Cases: 340
- Dr. Erickson: “That’s 6.5 percent of the population.”
- Wait, no, it isn’t!
- “Which would indicate that there’s a widespread viral infection.”
- No, it doesn’t.
You see, this is where the statistical bungling really begins; he’s saying that since 6.5% of the people tested were positive for COVID-19, we can conclude that 6.5% of the entire population has it. But that’s an absolutely erroneous conclusion, because the testing wasn’t random. This testing was done, especially early on, primarily on patients who had symptoms of upper respiratory illness and fever, had known medical conditions that made them high risk of complications from COVID-19, and who had some degree of known exposure to the virus.
Do you remember how just a couple of weeks ago so many people were upset that they couldn’t be tested because the criteria for testing was so strict? The fact that only 6.5% of even these patients had positive tests shows that the virus is not yet widespread in Kern County California, just like it isn’t here in Waco, or in any city that hasn’t yet hit a surge in COVID-19 cases yet.
This data cannot be “extrapolated” to the general population to determine the prevalence of the virus because the testing, so far, has not been random or representative. His methodology sounds reasonable enough on the surface, but it is actually leading him to wildly inaccurate numbers and conclusions that are the exact opposite of the case.
“We think it’s kind of ubiquitous throughout California. We are going to go over the numbers a little bit to help you see how widespread COVID is.”This should properly be understood as Dr. Erickson’s thesis for this video.
- 4:40 California:
- 280,900 Tested.
- 33,865 Positive for COVID-19.
- *dubious math*
- “That means that 12% of Californias were positive for COVID”
- Except it doesn’t, because you can’t get data on the number of cases in the state from non-random testing of symptomatic individuals with known exposures.
- It actually shows the opposite; even in patients who met the until recently very strict testing criteria, only 12% of those patients tested positive; California has NOT hit it’s peak yet. https://www.latimes.com/projects/california-coronavirus-cases-tracking-outbreak
5:08 These projections were based on what would happen without social/physical distancing, shelter in place orders, and other mitigation strategies. The fact that it “hasn’t materialized” is evidence that mitigation is working. We have been saying since day 1 that as soon as these strategies started to show success, people would say they weren’t necessary.
5:20 You cannot extrapolate prevalence data from testing of symptomatic individuals. We will explore how you could get this data later on, but for now, each time he ‘extrapolates the data’ you need to realize that the number that results doesn’t actually mean anything.
5:32 “That equates to 4.7 million cases in the state of California.” (No epidemiologist believes this; this is a nonsense number.)
“We’ve had 1,327 (now 1,651) deaths in the State of California with a possible prevalence of 4.7 million.”
“That means you have a 0.03 chance of dying from COVID-19 in the State of California.”Dr. Erickson
Do you see what he’s done here? He’s multiplied the percentage of tested cases that were positive by the population of the entire state and called that number, 4.7 million, “prevalence.” He’s then divided the number of deaths by that gigantic made up number in order to make the death rate seem incredibly small.
You are supposed to think, “wait, I heard something like a 3-4% death rate, but he’s saying it’s 0.03%. They’ve blown this whole thing out of proportion!” But the number he is deriving is incredibly small because the fake denominator he has come up with is gigantic; and that is going to be the case for any location regardless of whether they have yet been hit hard by COVID-19, because while he is multiplying the percent of positive tests by the entire population, the number of deaths stays the same. He is comparing known COVID-19 deaths not to known cases, but to a wildly inflated ‘guess’ at the number of cases that is not based on sound epidemiology statistics principles.
In fact, while he isn’t really calculating anything, what he’s closest to deriving by comparing number of deaths to population is what’s called the mortality rate, and since most people don’t die in any given year, this number is always going to be small compared with the general population; any number of deaths looks small compared with 328 million people. This is the reason we talk about mortality and attributable mortality rates in terms of ‘per 100,000 people’, because most of us (myself included) can’t conceptualize the significance of very, very small numbers. If I told you that the mortality rate of heart disease is 0.122% and the mortality rate of cancer is 0.049%, that’s going to be much less helpful than the more typically reported figures of 165 deaths per 100,000 vs. 37 deaths per 100,000, respectively.
So, what he’s giving us is an erroneously calculated ‘death rate’ that is so impressively tiny it cannot be conceptualized and compared well, in place of the commonly discussed and oft debated case fatality rate, which is the chance of dying if you do get the virus.
6:10 “I also wanted to mention that 96% of people in California who get COVID recover.”
Here he has tipped his hat; this is the case fatality rate. You see, the opposite of ‘recovering’ is ‘not recovering’, i.e. dying. He’s sharing the actual case fatality rate, what laypeople call the death rate, but in a form that is unrecognizable.
This is a classic spin technique; flip the statistic so it suddenly sounds like a good thing. “96% is really high! Recovery is good! See, the good thing has a high number, so we are fine!” But if 96% recover it means that 4% die, and that number is astronomical for a case fatality rate, far closer to the Spanish Flu epidemic (2.5%) than to the seasonal flu; and this is just in an area where the healthcare system is otherwise slow due to COVID-19 concerns; in places where hospitals are overwhelmed, the death rate (case fatality rate) is much higher.
6:12 “With almost no significant continuing medical problems (sequelae)”
It is way, way too early to know what the long term sequelae from surviving this virus are going to be.
6:28 “This is our own data, this isn’t data filtered through someone.”
Like, for instance, an epidemiologist who could help make sense of it for you? Sorry, I’m getting snarky again.
6:42 This is exactly backwards; the more the prevalence data goes up, the more positive tests you will get; but because it’s the real prevalence and not the erroneous prevalence he has calculated, that increasing prevalence will be accompanied by increased hospitalizations and increased deaths.
6:47 He’s just admitted to the calculation error I was talking about earlier. Incredible.
6:53 “Millions of cases, small amount of death”.
He says this over and over again; it may as well be the title of the video. Except it isn’t true; there isn’t any evidence that there are millions and millions of cases in California (41,000 confirmed at this point), and the number of deaths is anything but small. By the end of this week we will likely have passed the deaths from the worst flu season I’ve ever experienced, 2017-2018 (62,000 deaths), and epidemiologists believe we are underestimating the number of deaths from COVID-19. Moreover, this hasn’t peaked yet in most areas of the country; if we stop mitigation efforts, this could blow anything in our lifetimes right out of the water.
7:05–8:56 “So I want to look at New York State.”
- 25,272 Positive Cases
- 649,325 Tests
- 19,410 Deaths (not sure where he got this number from)
“That’s 39% of New Yorkers tested positive for COVID-19”
At this point one of the reporters clarifies that it is not 39% of New Yorkers, but only 39% of people who were tested in New York State, and how if it were 39% of New York’s population that would be nearly 10 million cases of COVID-19 in that state alone. This is an incredibly important distinction. Dr. Erickson acknowledge this but fails to understand the implication; he is still insisting that you can “extrapolate” data from the testing that has been done.
An explanation of why we can’t extrapolate the information he thinks we can, and how we could get that data.
This data can’t be used for the purposes he is trying to use them for, for at least three very compelling reasons. First, it’s the wrong testing strategy. He keeps saying you can extrapolate the test data we have to the general population, but the people who were tested do not represent the general population. They have self selected due to exposure or illness and, especially early on, had to meet very strict criteria (or be an NBA player or celebrity) to even get tested in the first place because of the shortage of tests; these tests were done on the people who were already the most likely people to have COVID-19, and so their percentage of positive tests (39% in New York, 12% in California per Dr. Erickson) is going to be far higher than any other group. Even accounting for asymptomatic carriers, there is no reason to believe that asymptomatic people would have the virus at anywhere near the rate of people who have symptoms of the virus. This is… pretty common sense stuff, actually. For testing to be used to extrapolate to large numbers that give us population level data, it has to be random, and this is the opposite of random. So it’s the wrong strategy for the conclusions he is drawing.
But even if it were random, it simply isn’t the right sort of test for that. The current tests detect COVID-19 (SARS-CoV-2) antigen; circulating proteins specific to the virus; it is detecting the virus itself. It can do this before the patient is symptomatic if the virus is replicating inside them, but not once the virus has been eradicated from the body. Because of this, it’s actually the wrong test for the job; a person can test negative once they have recovered, so they would be miscategorized as a ‘negative’ test even though they had already had the virus. At best, a sufficiently large number of (random) tests done on the same day could give you a snapshot of how many people have the virus at any given time; this is called point prevalence. If this were at all possible, it would indeed be helpful for knowing the current risk of being exposed to the virus (though it would change quickly and require serial rounds of testing). But you can’t use it to determine a death rate; for that we need period prevalence, the total number of cases throughout the time period of the pandemic, and for that we need to know who has had the virus, not just who has it now. So, it’s the wrong test.
But it’s also the wrong time. If we want to know the final, true case fatality rate for COVID-19, which we all expect to end up being very high but much, much lower than the astronomical numbers we are seeing now, we are going to need that period prevalence for the entire period of time of the Pandemic. Even if Dr. Erickson’s calculations were correct up till now (and they are so, so not), it would still be the wrong time to rely on them because many of the regions he is discussing, including his home state of California, have not yet hit their surge. We don’t know what the death rate in California will be because the virus hasn’t come and gone yet; their healthcare system, doctors, and nurses are yet to be tried. It is the same in Waco; we are still in the long calm before the storm, hoping that something will give (a vaccine, a brilliant epidemiological strategy, a radical new treatment being discovered, seasonal decrease in transmission, etc) and we won’t have a surge at all.
So, what would an ideal testing strategy look like if we really wanted good quality case fatality data? It would use antibody testing (which tells us if the person has ever been exposed and had an immune response to the virus, not just if they have it right now), would be random, and would be done after or at least at the tale end of the pandemic. This would take into account asymptomatic and minimally symptomatic cases, and people who had symptoms but never got tested at the time. With a sufficient number of tests it could be used to extrapolate data for the entire population with a good degree of reliability. He’s probably right that we won’t ever do testing quite like that; but since there are potentially lots of other uses for antibody testing, and some of it involves testing people who aren’t actively ill, it is likely that we will get data that can at least be legitimately used to derive some idea of prevalence and true case fatality rate.
While we are discussing New York and possible testing strategies, it is important to note that there is some preliminary data about the actual prevalence coming out based on the antibody testing we discussed earlier, and the news is indeed hopeful; but even the most optimistic numbers so far only get the case fatality rate down to about 0.5% in New York, when you include asymptomatic carriers, assuming the sample is representative; 5 times higher than the number Dr. Erickson has landed on, and still incredibly dangerous. This is a number most of my colleagues would believe sooner than something apocalyptic like the 8-12% in overwhelmed healthcare systems across the globe, and Physicians and Epidemiologists have anticipated and said from the beginning that these numbers would drop significantly once broad-based testing and antibody testing were available. But unlike Dr. Erickson, most doctors I know are not comfortable making that kind of stuff up and would prefer to wait for data that actually has a logical connection to the questions we are asking.
But even as more random antibody testing is done and death rates for COVID-19 hopefully trend down away from the utterly incomprehensible numbers they are at now, please remember; it isn’t just the case fatality rate that makes a disease dangerous, it’s also the degree of infectivity. Even if COVID-19 settles out to be less deadly per case than the bubonic plague or ebola or the Spanish Flu Pandemic of 1918, it can still kill incredible numbers of people if it makes up the difference by being highly contagious… Unless our mitigation strategies can prevent it from spreading.
8:12 Reporter: “Those models were based off if we did no social distancing.”
Dr. Erickson hand waves this off, but it’s an important point for understanding the timeline of this pandemic and understanding that those models are still a real possibility if we stop mitigation efforts.
It’s also an important opportunity for demonstrating some intellectual integrity, since the reporter is correct that those models were for scenarios where social distancing wasn’t followed, and Dr. Erickson has been dismissing them as ‘wildly inaccurate’. Sadly he fails to rise to the occasion and acknowledge this.
8:54 “We extrapolate out and use the data we have, because it’s the most accurate we have, versus the predictive models that have been nowhere in the ballpark.”“
This is a blatant false dichotomy. The predictive models were done to show the range of possibilities of the impending danger if no action was taken; the antigen testing strategy to identify and isolate cases. Neither can be used to establish actual prevalence, but he wants us to think we have to accept his calculations, based on erroneous assumptions, because it’s the only option.
8:59 “So how many deaths do they have? 19,410, out of 19 million people. Which is a 0.1% chance of dying from COVID in the state of New York. And they have a 92% recovery rate! (Edit: That’s an incredibly high known case fatality rate of 8%!) Millions of cases, small amount of death. Millions of cases, small amount of death.“
I want to be as generous as possible here. I really believe that this could be me, were the circumstances different, going on youtube and sharing these false statistics. Yes, Dr. Erickson has financial interests at stake here, but so far I’ve been inclined to think that he really believes his numbers. When you are pouring over data like this for hours or days and you think you’ve hit on some vital statistic that nobody else is picking up on, and it confirms what you already really, really want to believe, it can be so easy to get tunnel vision and not check your math against the backdrop of reality.
But New York should have been the “Aha!” moment for him; the point where he sees the house of cards he’s built collapse so he can start over from scratch with all of his equations. 19,000 deaths; 19,000 deaths in one state, in one month. Overwhelmed hospitals, too few ventilators, nurses and doctors collapsing at work. These stories from the front lines should be enough to make him question the conclusions he is drawing.
If you are calculating a pediatric dose of antibiotics and arrive at instructions that tell the parents to give 28 teaspoons three times per day, you’ve made a mistake somewhere; it doesn’t matter if your math was perfect, something must have gone wrong because those numbers don’t mesh with reality. If you are trying to figure out how long it will take you to drive from San Antonio to Waco and google maps tells you it’s 22 hours, something went wrong; it doesn’t matter how good their calculations and traffic algorithms are if the app thought you meant Waco, Montana instead of Waco, Texas. And if you are trying to derive real-life mortality data from numbers available on google and discover that a virus that is killing tens of thousands in a short amount of time, overwhelming hospital systems, and leaving your colleagues in New York with post traumatic stress disorder is actually not that dangerous, you’ve probably made some flawed assumptions before you even fired up your calculator. Your mathematical conclusions have to line up with reality, and his don’t.
He has concluded that COVID-19 is no worse than the flu, which in any given year will kill between 10,000 and 60,000 people nation-wide over 3-5 months. But the deaths of 19,000 human beings, with friends and families, who wouldn’t have ‘died anyway’ at this time, many while their doctors and nurses looked on helplessly because they had not the time or lifesaving equipment to intervene, in one state in one month, should be a wake-up call even for him.
9:48 “We’ve tested 4 million people. Germany is at 2.” The population of the US is 330 million and the population of Germany is 83 million; their tests per capita is double ours. He hand waves this with ‘sure I realize their populations are lower, but…’ Don’t trust anyone with your statistical analysis who waves away the single most important statistical number for comparing countries, their respective populations.
And at this point, mercifully, the video has been removed from Youtube for spreading verifiably false information. This is a double-edged sword, because it inevitably means that copies of it will be spread elsewhere with the heading “BANNED FROM YOUTUBE!”, and even more people will click, watch, and be deceived (or more likely, further entrench the false narratives they have already chosen to believe before watching). If someone does have links to the video when it’s up again, please send it my way so I can finish the other (checks notes) 45 minutes of the video.
But some sanctions cannot be waived away by your being popular with conspiracy theorists. The American College of Emergency Physicians and the American Academy of Emergency Medicine today released a joint statement condemning the irresponsible and flawed information in the video. And while the parts that we have covered so far have been mainly bad statistical analysis disconnected from reality, there are statements made by these doctors later (which I cannot now quote verbatim) that much more flagrantly disregard the oath they took in medical school. I honestly hope these are played back for them the next time they are set to renew their board certifications, and indeed their medical licenses.
With the video down, I’ll have to conclude here for now, and considering the number of charts I need to close for clinic, I can’t thank YouTube enough for taking down the video when they did.
Over the next 10 minutes or so, Dr. Erickson applies his same flawed methodology to other countries, multiplying their positive test rate by their total population to come up with his fake prevalence numbers, and then dividing the number of deaths by that to show how not dangerous the virus actually is. “Millions of cases, very small deaths.” If the video ever comes back, you can watch him do it time and time again, as a tutorial of sorts, so that you too can enjoy creating your own fake statistics at home.
And this leads him to conclusions which, while obvious from his erroneous numbers, defy both our reason and the experience of our fellow human beings. He concludes, remarkably, that the COVID-19 virus has not been that bad even in Italy and Spain, where it decimated the healthcare infrastructure and killed tens of thousands. He concludes that the difference between Norway’s 200 deaths and Swedens’ 2000 deaths is statistically negligible, and therefore social (physical) distancing measures don’t actually matter. He does this because, again, he’s invented a sufficiently high denominator for his “prevalence” that literally any number of deaths is going to seem “insignificant,” at least statistically.
- Sweden’s Population: 10.2 million.
- Deaths in Sweden (without mitigation strategies): 1,765
- Norway’s Population: 5.4 million.
- Deaths in Norway (with mitigation strategies): 182
14:30 Dr. Erickson: “1,700 (deaths), 100 (deaths); these are statistically insignificant.”
I want you to stop and say that out loud a few times. Go ahead.
These lost lives are not insignificant; statistically or otherwise.
One more thing I remember specifically, because it was so shocking to me at the time. He goes on to talk about the way that the mortality data is being ‘manipulated’, even saying that a deceased patient with COPD (Chronic Obstructive Pulmonary Disease) who contracted COVID-19 has not actually died of COVID-19, but from 25 years of smoking… As though the medical vulnerabilities that predispose a patient to becoming a victim of this horrible virus and the pathology caused by the virus itself are mutually exclusive. As though tens of thousands of COPD patients who have been smoking for decades were suddenly going to go into respiratory distress in April 2020, apart form any exacerbating factors, and their happening to have the virus that is also killing people with heart disease, diabetes, compromised immune systems, and even the young and healthy is just some weird coincidence.
Bad at statistics is one thing. This is bad at being a Doctor.