Covid blood testing in Telluride, expert projections + more
I posted a link to a story last week saying that Telluride Colorado (San Miguel County) is doing some of the first wide scale blood antibody testing. Early results are in, they’ve tested about 1000 of 8000 residents.
‘Of the almost 1,000 people tested for COVID-19 in San Miguel County, eight have come back positive and another 23 have either indeterminate or borderline results.” (About 1% to 3%, but unclear if testing is being done randomly or there is an attempt to focus on those who are likely to have been exposed first)
In other news, linked is the weekly survey of experts being done at U-Mass that 538 has been highlighting. Myself and others have posted earlier iterations (I haven’t seen this one yet.When there is this widespread disagreement amongst experts, you can see why some of the debates here seem like they are based on completely different sets of facts One tidbit:
“They believe there were between 289,000 and 12.8 million infections, with 1.1 million being the consensus estimate, implying that the experts think that only about 12 percent of all infections have been reported. This is in line with what they’ve reported for the past three weeks, when the share of reported infections has ranged between 9 percent and 12 percent.”
Also from 538, an interesting piece explaining some of the challenges with modeling Coronavirus.
‘So, imagine a simple mathematical model to predict coronavirus outcomes. It’s relatively easy to put together — the sort of thing people on our staff do while buzzed on a socially isolated conference call after work. The number of people who will die is a function of how many people could become infected, how the virus spreads and how many people the virus is capable of killing.
See? Easy. But then you start trying to fill in the blanks. That’s when you discover that there isn’t a single number to plug into … anything. Every variable is dependent on a number of choices and knowledge gaps. And if every individual piece of a model is wobbly, then the model is going to have as much trouble standing on its own as a data journalist who has spent too long on a conference call while socially isolated after work.”
https://fivethirtyeight.com/features/why-its-so-freaking-hard-to-make-a-good-covid-19-model/
Finally - a few quick thoughts about exponential growth.
5 doubles of 1 get you to 32. 10 doublings get you to 1024. 15 doublings to about 32,000. Doesn’t seem to bad does it? But 20 doubles gets you to over 1 million. And from there, things really take off.
Should have noted this in the original post, but antibody testing lets us know who’s had the virus even if they are fully over it (or never displayed symptoms at all)
Beat me to it, I was just about to post before I saw this. TBH, the CO Public Radio article (they're great, I'm out in Boulder several times a year) was very shaky in it's handling of the science in that article. Someone who has tested positive in an antibody test is not necessarily a "carrier" and they're not testing for COVID-19, they're testing for anti-SARS-COV-2 antibodies. Many if not most people testing positive for the latter will never develop the former.
EDIT: There is no serum antibody test AFAIK that establishes current infection, and you're unlikely to see one because a serum test won't catch low viral titers, but an RT-PCR test or the new Abbott ID NOW platform will do it. The article would have done well to clarify a little more.
I've been seeing some very (seemingly) reputable sources (Mayo Clinic, other University-based MDs and MPHs) posting YouTube videos on coronavirus infection biology with clear mistakes in the immunology aspects. Saying all this simply to make a point: the analogy between the pandemic and war is apt in several ways, one of which is the "fog of war" that descends and causes confusion and misinformation, usually with no bad motive.
Always appreciate your insights. Would you mind explaining your points a bit more (or if you have a good cite, I’m happy to read for myself).
Sorry, I was editing even as you were posting this reply.
The serum test is an exposure test, in other words, "has the SARS-COV-2 virus ever introduced itself to you." If it has, your immune system (assuming you have a healthy one) will expand a clonal subpopulation of B-cells (immune cells that produce antibodies) specifically targeting the virus. Those will circulate around in your blood in case it shows up again, well after the virus is long gone. That's what the serum test is looking for... it's basically a biological footprint that the virus stopped by at some point, but doesn't tell you it's still hanging out. If the virus is not actively producing more virus in your body, you're not a "carrier" even if you've been exposed.
Worth bearing in mind that coronaviruses do not have a "latent" phase in their life cycle, meaning there is no period where they're just laying low and silent like, e.g., HIV or Herpes Simplex Virus, which can go radio silent for many years. Coronaviruses are either making noise or they're absent.
That makes sense. Is it safe to assume that if you test positive you were at one point a carrier even if you were asymptomatic? That is what I was assuming.
Yup. Spot on.
We may not be aware the virus has arrived and we're (possibly) infectious, but the immune system knows it's there, and remembers when it's gone. Every bug that stops by gets its mugshot taken and filed.
You have a talent for explaining complicated stuff to the layperson like me. Thank you.
Excellent stuff, thank you
The serum anti body test is what we need to get us back to work. It will also help those 65% of people that are either asymptomatic or presenting with atypical symptoms know if they can end this crap and get back to life.
Moore’s Law?
Interesting stuff, thanks for sharing. Some additional info that might be relevant:
per the state, 9 people from San Miguel have tested positive for C19 https://covid19.colorado.gov/case-data
the county has ~8k people (2018) so they’ve got a ways to go on this study.
Yes, +1. The actual confirmed data is relevant. The confirmed infection rate is ~ 0.1% (9/8191).
The serology testing suggest an infection rate from ~ 0.8% (8/986) to ~ 3.1% (31/986).
So, the rough range on the "multipler" is 8-31x.
Most of my conjectures have been that the multiplier is 10-20x.
------
Another somewhat relevant tangent --- San Miguel County probably has some "extra" population right now. Many of the vacation towns in America do, folk who have non-primary houses or condos there. In fact, James Franklin and family have spent the past several weeks there.
Per Colorado protocol, if Franklin tested positive, he would show up in San Miguel County, CO stats. Which, in that first equation, he'd be in the numerator but not the denominator. A dynamic like that may inflate the confirmed infection rate a bit.
This is exactly the problem that anyone who has been messing around in the data and building models from it (including me, Nittany Fan, and others) encounters: it's all about the assumptions you make. One of the early insights I had was that the numbers of reported/confirmed cases only works if one of the following are true: there is at least a large multiple more people - maybe an order of magnitude - infected who don't know it; or this isn't nearly as transmissible (which implies R0 is far lower) than is commonly understood.
Just in Oakland County, I built models using two methods based on data from Health Weather to get to the confirmed number of cases. They arrive at different answers for where the peak will be, but the worst case is about 11,000 by next weekend. That seems like a huge number, but is still a small fraction of the number of infections that is almost certainly out there.
One piece of good news, is that most people in the county aren't being stupid. I had to venture out to Walmart this morning for a few supplies that were running low. 90% of people were already wearing masks. That will help keep community spread contained.
Yeah I follow Nate Silver on Twitter and people have been pissed when someone says "Trump is going to kill us all" and he says the models are actually all over the place and we really don't know. And I'm someone who thinks (knows?) Trump has been horrendous in this. But there's so much we don't know. How easily is it transmitted? How many people are social distancing? How long are you contagious for? How many people have it? How does population density affect things? How deadly is it really? What underlying health issues contribute most to hospitalizations or death? How many people are asymptomatic? How much does warmer weather help?
The amount of unknowns there are makes Trump's incompetence through this whole thing even worse. There's so much we don't know, which makes his not taking this seriously even worse
Yeah, I mean anitbody tests are good, and will help shed light on the number of cases and especially asymptomatic ones, but that tells us about what already happened with the coronavirus. It still seems the US isn't even talking about any kind of testing and tracing. Not even gonna go into all the other issues.
Fauci and many others have mentioned multiple times that this is the key to opening the economy back up.
The economy will get restarted by the late spring, assuming a few things: 1) everybody wears masks in public places, including offices, stores, etc.; 2) we have testing for the virus and its antibodies at scale; and 3) we continue to limit exposure by only going places when and where it is absolutely necessary to reduce our risks.
Hopefully employers who have large numbers of employees who can work from home understand this and the key tole they can play by not just allowing but aggressively encouraging as much telework as possible.
We’ll also need to think about what can be done for high risk working folks who have significantly less practical options for occupational social distancing. I suspect this number is larger than many assume.
That won't work very well for bars or restaurants which are a huge part of the economy....much bigger than people realize.
Totally agree. Same goes for tourism.
How big of a sample do you need for the serological test to say that’s there’s enough herd immunity to protect the population? Or are you going to be able to go back to work only if you test positive for the antibodies? Compulsory testing? If that’s the case, I guess some of the people on the blog won’t be going back to work for a while.
I shouldn't have been as vague as what I said. Of course some people have mentioned it. But some countries have been doing it for months already. The US is still lagging with testing, even MD's Republican governor is complaining. Has Fauci mentioned anything specific about a nationwide tracing system or how that would work? That still sounds like it's months away from happening.
Unfortunately most of the mentions I have seen from Fauci and others have been more about the need, not the detailed plan. The good thing is, 1 month ago we didn’t have the technology. Now we do. We have rapid real time testing providing results in 15 minutes and we have the ability to do antibody testing. Those are big steps forward. Now we need to scale up that capacity as fast as we can.
We also need to put in place mechanisms to do massive amounts of contact tracing, a very important and very labor intensive piece of the solution. One of the big challenges is that our best public health minds are quite rightly more focused on the immediate challenges than on next steps.
Do you think contact tracing 330mm people is actually doable or is that pie in the sky? Just seems like an impossible task. My feeling is that as more data continues to come in and continues to show how deadly this is at the upper range we start to consider longer term quarantines for the elderly and infirm and start the contract tracing and testing for everyone else while we restart our lives.
We would remove 40-50mm in the short term from the testing protocols and make it more feasible. We've done about a 1mm to date so we need to compartmentalize this somehow if we want to get back to life at some point.
I don’t think the idea is that you necessarily test everybody (even though there is obvious value to knowing who has had it). The idea is that you jump on new infections quickly, aggressively track down the people who have been in contact with that person and test them. This is where rapid testing becomes important. You no longer have to quarantine everyone they’ve contacted for 14 days. I suspect you’d couple it with other things like periodic testing of medical personnel so as to identify asymptomatic people. Particularly for these folks, antibody testing would be good as it limits the numbers you have to keep checking.
Maybe they also try to do larger scale antibody testing to reduce people who have to be tested as part of the contact tracing effort?
Yes, exactly. That's what's been going on in Korea.
When you say the US is lagging in testing, what do you mean? I know we were slow to get started, but it seems we have very quickly ramped up testing. According to this source, the US leads the world in most Covid tests completed (about double the next country, Italy). Normalized for tests per 1 million citizens, we rank 4th in the world (and the countries above us are much smaller geographically and demographically). Since about mid-March, we have been leading the world in most tests completed per day.
Could and should testing be better? Yes, I think so. I would like to see more testing, and the data I referenced above does not invalidate the push from Maryland's governor; the de-centralized nature of America means that there are going to be some states/locales that are further along than others.
But I think it's worth acknowledging that while the US was slow to start, we have been ramping up very quickly. And I don't think the idea that we are "lagging" applies anymore.
The forward-looking importance with antibody testing is also potential immunity. As we learn more about how long those who have had C19 and recovered are immune (there are promising indications that it could be on the ~months-yrs timescale), those individuals can go about their lives without risk of getting infected or infecting others. They can also serve important societal roles to protect the most vulnerable (example: caretaker in nursing home). Antibody testing absolutely has to be part of the plan to safely increment society back to some normalcy.
Earlier in the week I started playing around with some data, mostly because I wanted to see if there was any way that this was widespread in the US in January. One thing that quickly became apparent to me, is that the lag times in data and assumptions about doubling time had a huge impact.
For instance - deaths and the morbidity rate are related to the number of actual cases 3 or 4 weeks ago, not today. The number of new daily cases should be related to the number of new infections about a week ago, but because of testing lag it is greater. Therefore if you make assumptions about the undercounting of cases based on positive tests today (e.g. assume it is between 9 and 10 times higher) and use that as your denominator, with today’s deaths as your numerator, you will almost definitely under estimate the morbidity.
This is also the conclusion I drew. It seems as though the mean lag in confirmed cases is 10-14 days from infection (again, using the data in Health Weather as a proxy), and deaths is about 21 days' lag from infection.
The silver lining (if there is one) is that it means we are likely within a week of the peak number of confirmed new cases per day. The rest of April should see declines in new cases day over day, though the death toll will continue to climb throughout most of the month.
As always, your insights are appreciated.
Are you talking about for Michigan, the US as a whole? Another thing that is clear is that the national numbers are driven so much by NY (and NYC) in particular. Once NY peaks, the national numbers very well may go down even as other areas just start to see their biggest problems.
I'm kind of mixing (Big) Apples and bowling balls; but, yes, the data in the U.S. is heavy skewed by NY, which is beginning to see some flattening in the confirmed cases per day. Regarding MI, if all of the data from all of the experts works out (I'm not counting myself among them - I'm just a guy who can math for fun) then the county, if not the state itself, should hit a peak in a week or so,
Also, many thanks for the kudos. All of those forecast models I built to predict the demand for USAF spare parts are starting to become useful again.
Just as the US is driven by NYC (in terms of peak etc), So to is Michigan driven by the Detroit metro area. I’m in Grand Rapids and the hospitals here are running their own data. Based on their modeling we are weeks behind the east side of the state. They don’t think we will peak over here until May-June.
Things at the hospital have been on lockdown for 3 weeks in preparation but we just aren’t seeing the numbers here yet. So Detroit will just be coming down right when GR and west Michigan start to ramp up.
Using similar data and methodology they estimate 15% of the population of Spain has been infected already and 10% for Italy. If that's true, the end of the tunnel is closer than we think.
Except you need 60-70% of the population to have had it to keep us from having Covid 19 or Covid 20 this winter. With the same lockdown and the perhaps similar mortality rate this winter.
I was somewhat hopeful the undercounting was high enough in Italy and Spain the infection rate was higher than that. God knows they haven't bothered to test anyone who didn't end up in the ER in either country.
Respectfully, can I ask what qualifications you (and others on this board) have that makes you feel that sharing the results of your models in this format is productive? I'm not a infectious disease expert, but I am PhD physiologist with numerous friends who do this for a living. It is complicated stuff and even the best models of past outbreaks don't always replicate reality. Additionally without reporting what types of models you are using, readers have no clue what assumptions you are making or what your model includes.
I think everyone should play around with the data. It is a good quantitative exercise, and you may be able to see some patterns. However good science isn't just one person playing with the data. It is being transparent about your work so others can examine it. It is spending years understanding nuanced differences between models. It is understanding that models alone shouldn't shape behavior, as they are only as a good as your original data (and Health Weather is AWFUL data). And it is understanding that the best models should also agree with other sources of data.
Frankly, I feel that sharing results of your models on an internet sports forum can be harmful. For better or for worse (it is for worse) many people will base their decisions on what they read on a sports blog. Saying things like " there is at least a large multiple more people - maybe an order of magnitude - infected who don't know it" or "They arrive at different answers for where the peak will be, but the worst case is about 11,000 by next weekend." will likely influence other people's behaviors. Are you ok with this?
I'm an aerospace engineer by degree (and spent the first 8 years of my career designing and testing rockets and aircraft), and a supply chain professional for the past 20+ years, in which I built forecast models of demand for the spare parts of USAF aircraft and ground support equipment. Those models are at least as sophisticated - and difficult to get right - as anything associated with infectious diseases.
The important thing isn't what you're modeling. All models, whether statistical or deterministic, follow the same principles, and they help take emotion out of the discussion. To quote a line from the movie, Margin Call, they're all just numbers, really. All you're changing is what you're adding up.
As far as the rest of your remarks, I agree that it's important to be clear about the assumptions I'm making. But this is a blog, not a peer review, and the forum isn't really suitable to upload Excel spreadsheets, SQL databases, and the like.
And with all due respect to you and your degree, based upon what I've seen today at the local Walmart, few people, if any, are basing life and death decisions for themselves and their families on what they read on sports blogs. Most are treating this with the seriousness it deserves. This virus, and the illness it can cause, should be respected; I sure do.
We agree on a lot regarding models and their underlying data. I'm sure the ones you built previously are advanced stuff. I'd however argue that these are not the same thing conceptually, especially regarding the spatio-temporal components of human disease spread and social networks. However without knowing what models you used who knows, I could be wrong here. I do know however that your original post has a ton of upvotes yet multiple parts of it send off red flags based on my training and what is in the primary literature.
The amount of misinformation going around during this pandemic has been killing me. Everyone is an expert on everything, media headlines are terrible, a vaccine is discovered every day, and suddenly people think everyone already has covid-19 because they were sick once in Dec or Jan. I've been keeping quiet myself except to point to a study here or there because I realize I'm not a world expert in this field and frankly these are life and death times. I'm not saying your model is misinformation. I'm also not saying it is good information.
You are right that this isn't the right forum for the nuance these topics deserve. I think that was my point to begin with and what you saw at Walmart today doesn't change my view.
Stay safe!
Every epidemic is really two epidemics. There's the microbe, and there's the misinformation.
I think the only point on which we disagree is that the models aren't the same conceptually. The models I've worked with in the past (though admittedly haven't accounted for directly in the models I've built to understand this pandemic) do add variables and/or constants that account for seemingly random behaviors and events. In your profession, you would probably think of those as the factor associated with the dumb things humans do just because. In my world, it's about the factors that drive the failures of parts and components. In the models I've been playing with lately, I've also attempted to build in temporal effects, but not spatial effects.
But in no way am I am expert on infectious disease or epidemiology. I am just using my math skills to try to better understand the nature of this pandemic and how bad it might get, which helps prepare me for the dark days ahead. I am hopeful that by sharing it, a few folks might be less worried when the numbers grow, but realize that it wasn't unexpected. For others, certainly, acknowledging that the cases and death toll will be staggering is simply too much to wrap their heads around.
What the data from COVID-19 tests show at the University of Washington Virology Dashboard and Michigan's Dashboard, is that only 10% and 25%, respectively, test positive for the virus. Given the scarcity of test kits and supplies, this implies that most patients presented with COVID-like symptoms and didn't have it. In other words, there are a lot of bad bugs going around that can cause serious complications. Respiratory viruses are no laughing matter at any time, not just now.
I apologize in advance that this post is going to be harsher than the last few.
Using math skills is not the same thing as dedicating your life to modeling infectious disease.
Here is a review paper on infectious disease modeling from 2013: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3710332/ Give it a read and let me know how your model compares. My understanding is after this paper was written the field has moved more towards complex network models "that are relaxing the hypothesis of the above stochastic models that the interactions between individuals are instantaneous and homogeneous". This isn't the same as just adding in some randomness.
Here is a whole journal article dedicated to papers on modeling infectious disease: https://blogs.plos.org/everyone/2020/02/20/mathematical-disease-dynamics/ Give it a read and let me know if a model is just still numbers.
The Dashboard data conflicts with one of your conclusions: that orders of magnitude more people have the virus than realize it. Did you mean people who haven't been recorded as positive? That is different than asymptomatic carriers.
I appreciate your love of the numbers and trying to predict the future. Perhaps it would be better to discuss models/predictions from experts rather than your own?
And now you've gone off the deep end. So be it.
I'd assume if you were posting to a public forum about the results of your disease model during the middle of a pandemic you would have already looked into disease modeling.
I think we've both made our points. Stay safe out there and good luck!
I have looked into disease modeling. That's what, in your arrogance, you refuse to see.
To put it bluntly: I have studied the models used by, among others, the University of Washington's Virology Dept, the University of Minnesota's Center for Infectious Disease and Public Policy, and the CDC. All of them have their advantages and disadvantages, assumptions, and best guesses; as all models do, regardless of the level of sophistication.
Dude, I know how math works. I've been doing it for my entire academic and professional career.
The point that he's making, and that you so far have not acknowledged, is that understanding modeling is not worth a lot if you don't have a nuanced knowledge of how those numbers should be sifted and applied. Without in depth study of infectious disease, it isn't possible to have that nuanced understanding. So while your models appear impressive - and maybe they are - they aren't worth much without having some credentials to back up your work.
The conclusion that many more people have covid than have tested for it is hardly revolutionary. The survey of experts piece that I noted in the OP suggests a range of 2 to 100 (with a best estimate average of between 8 and 9). The numbers that NJIa is suggesting are well within that range.
I’d be much more concerned about the way modeling from an untrained epidimiologist may have influenced early decisions by the White House (https://www.newyorker.com/news/q-and-a/the-contrarian-coronavirus-theory-that-informed-the-trump-administration) than I would about the discussions from an armchair epidemiological modeler on a random sportsblog.
I frankly think that throwing out numbers informed by actual analysis based on reading underlying expert analysis is way more responsible than some of the other comments that get thrown into these discussions on a regular basis.
This may seem nit picky, but there is a huge difference between "there is at least a large multiple more people - maybe an order of magnitude - infected who don't know it" and "many more people have covid than have tested for it". The former implies a ton of asymptomatic carriers (they don't know they are sick), leads to the dangerous theory that everyone already has covid, and has been circulating as a conspiracy theory and/or misinterpretation of scientific articles on social media. Testing data also shows it is completely false. This is what made me speak out about all this in the first place.
Regardless it seems clear that we are going to disagree about all this. I don't think it is arrogant to suggest that infectious disease modeling should be left to infectious disease modelers, regardless of people's quantitative skills and background. It isn't just the math that matters but understanding disease mechanics and social interactions.
Love and good health to both of you.
There are several parts of my background that are relevant. Mechanical engineer who has spent nearly 30 years working on environmental policy. In that time I’ve worked on environmental monitoring, run an economics modeling group focused on industrial economics and emissions, worked with states on implementing federal policies and overseen the development of national environmental regulations (which forces you to have an understanding of legal issues, health impacts modeling, air quality modeling, the economic impact of regulation etc.).
I am also a high risk individual who thus has a strong vested interest in this topic, as well as more experience than I’d like at understanding differing medical opinions on varying topics (I’ve had more than one dr ask me if I was a medical professional when evaluating treatment options - I believe it is important to be an educated patient).
I have found the discussions on this blog very helpful over the last several weeks. They’ve been both therapeutic and educational. They’ve exposed me to a much wider range of information than I would have necessarily uncovered myself.
I’d never make a health decision based solely on what I read on any blog (medical, sports or otherwise). I do however think the variety of opinions on this blog is likely exposing at least some people to a wider range of information than they would otherwise have. I feel pretty comfortable that is a good thing.
I am also a high risk individual who thus has a strong vested interest in this topic, as well as more experience than I’d like at understanding differing medical opinions on varying topics.
Same here. Like you, I have found that media reports of victims of this disease being described as "having an underlying condition," or "you are at higher risk if you have/are X" to be less than useless.