Posts
Comments
This stuff is super hard.
I'd recommend (with reservations) Consent Academy, who do a lot of training on incident response, accountability processes, etc. They're good folks who have figured out a lot of really useful things about doing this kind of work.
Their classes can sometimes get pretty rambling and theoretical, but I've learned a lot from them.
Strong work: thank you!
I believe there's a small mistake: in the first table (after "In equilibrium, we see the following amounts of sars-cov-2 relative to no filtration:"), I believe the second column should be labeled "presence", not "reduction".
Lots of cool data here—thank you!
(Edited to remove a comment based on misremembering the Most Penetrating Particle Size)
I'm overdue for making another pass through the latest data, so my opinions on this are weakly held. But briefly: my current thinking is that many people (including Zvi and me) have made the mistake of conflating a number of different phenomena into the single category of "long covid". I believe Zvi is correct that if a large number of people were suffering long-term debilitating impact, we'd know it.
I suspect that after I plow through the data again, I'll update significantly in the direction of believing that:
- "Long covid" is a debilitating phenomenon that affects a very small number of people for a long time.
- "Post-acute covid" is significantly impactful and impacts a non-trivial number of people moderately for weeks or maybe a few months.
Anecdata: I don't know anyone who's been profoundly impacted by covid for a very long time. I know multiple people who've suffered significant impairment for weeks / months.
The impact of long covid is (small incidence #) x (large impact #), and the impact of post-acute covid is (medium incidence #) x (medium impact #). I think for most people, the total expected impact of getting covid will be somewhere between a day and a few weeks of useful live lost, with large error bars and much of the impact being in low-likelihood events.
Eliezer, back in 2009:
Yet there is, I think, more absent than present in this "art of rationality"—defeating akrasia and coordinating groups are two of the deficits I feel most keenly.
This is not a small project, and I'm too new here to have a clear sense of how it might happen. But this feels important.
To more directly address your initial question: to my mind, Zvi's analysis isn't obviously wrong, but it's pretty far to the optimistic end of what I see as the reasonable range.
My best model suggests that for me (55 but very healthy), 1,000 µCoV of risk has an expected life cost of about 15 minutes.
Based on that, my approach to risk is very situational. Is eating in a restaurant worth 75 minutes of lying in bed with flu wishing I was dead (based on today's numbers)? No, it isn't. Is going to a friend's wedding worth that? Yes, it probably is.
I'd love to see a more structured approach to the kinds of questions you're raising here. LW does a good job of creating a space for smart people to share their thoughts about individual topics, but isn't so good at building toward a coherent synthesis of all those pieces.
The original microCOVID white paper did a good job of summarizing a lot of relevant evidence back in the day, but (like the rest of the site) has been only sporadically updated.
Put me down as tentatively interested in being part of some larger project, if one comes together.
Also: may I humbly request that if this ever takes off, it be named LessSick?
That all makes complete sense.
And yes, the specifics of the population make a huge difference. Honestly, I think that accounts for the breadth of my estimate range more than uncertainty about abstract test performance does.
I think it's important to emphasize that antigen+ people are much more contagious than antigen-. It's hard to quantify that, but based on typical differences in Ct value, it's probably a very substantial difference (factor of 10+?).
You're absolutely right that the reference class is the key issue (if there's one thing I've learned from hanging out with epidemiologists, it's that they're always grumpy about people using the wrong denominator).
In a perfect world, where everyone with any symptoms whatsoever stayed home and was scrupulous about following what the CDC exit guidance ought to be, antigen tests would be significantly less useful. But in the real world, people absolutely go out when they have mild symptoms. That's advocated for in the comments right below this, which are from people who are presumably much more conscientious than average.
IMHO, the biggest value of antigen tests is in catching people who are mildly symptomatic but think it's just allergies / they had a negative test last week so it can't be covid / they're probably over the worst of it. Within my (not enormous) extended social circle, I'm aware of two very recent cases when antigen tests flagged as infectious people who would otherwise have been out and about despite having mild symptoms.
Let me give you two answers for the price of one:
-
FDA and others have been very clear about this: you should use the tests as directed.
-
I (a decades-long amateur epidemiologist who's done a deep dive on antigen test research), my partner (a medical epidemiologist who works full-time on Covid), and several other epidemiologists I'm aware of, all use throat + nasal swabs.
I wouldn't worry at all about false positives: they really haven't been an issue with antigen tests. If I got a positive from a throat + nasal swab, I'd follow it up with a nasal-only swab or a PCR, just to be sure.
There is non-zero risk that you'd get false negatives, by some unknown mechanism. That seems unlikely given that some countries like the UK use throat swabs, but it's possible. It's my well-informed but not data-supported belief that the benefit of swabbing your throat probably exceeds the downside.
Why do I respect Michael Mina? Weak but deep answer: because in my experience he’s been consistently smart, and insightful about Covid and especially about testing (and he’s a professional epidemiologist / immunologist). Strong but shallow answer: because my partner, who is a medical epidemiologist working full-time on Covid, thinks highly of him.
If you’re not already familiar with her, you might also be interested in Katelyn Jetelina (Your Local Epidemiologist). IMHO, she produces by far the best deep research summaries for laypeople. Here’s a recent piece of hers on antigen tests.
In the interest of staying focused on truth-finding, here’s my understanding of the crux of our disagreement—does this look right to you?
I believe that using antigen tests before social gatherings substantially reduces the amount of transmission at those gatherings. It’s very hard to put a number on this—if I had to guess, I’d say a 70% reduction, but probably somewhere between 25% and 90%. If I’m understanding you correctly, you'd pick a very low number: less than 10%?
Let me try to explain my thinking, which I believe reflects the current medical / scientific consensus (though I think most scientists would balk at the rationalist proclivity for picking best-guess numbers).
There’s a massive body of evidence that antigen tests can detect all strains of Covid, including Omicron. Antigen tests are much less sensitive than PCR tests, meaning that they will consistently return false negatives when viral levels are low, but they have excellent sensitivity when viral levels are high.
The standard interpretation of that data is that antigen tests are an unreliable way to tell if you have Covid early on in an infection, but they are quite good at detecting Covid when viral levels are high (and therefore when you’re infectious).
The Soni et. al. chart you included is an example of this in action. Antigen tests gave nearly universal false negatives during the first two days that PCR tests were positive. Viral levels (and therefore infectiousness) tend to be low during the first couple of days, especially among vaccinated people (which most of the Soni subjects were). So what we’re seeing there is that antigen tests would consistently have missed people early in their infections, when they were minimally infectious.
From day 3 onward, however, antigen tests were extremely accurate. This corresponds to them consistently detecting people during their period of maximum infectiousness.
So there’s a huge amount of evidence that antigen tests are highly sensitive during periods of peak viral load / infectiousness. That’s easy to measure, and I think it’s pretty definitively established at this point. The question we’re really asking, however, is how that affects infectiousness. Unfortunately, there’s no really clear way to answer that.
We believe most transmission happens during periods of high viral load, and we know antigen tests are very accurate during that time. But we don’t know exactly how viral levels impact transmission, and figuring that out would require complex, expensive studies that would likely not be approved for ethical reasons.
I'm wondering if you can explain a bit more about your thinking here.
From my perspective, there's a strong prior that antigen tests work well for Covid screening:
-
There are numerous peer-reviewed studies to that effect. Here are two recent ones, but there are many others Soni et. al., Jüni et. al..
-
Multiple experts in the field continue to assert that antigen tests work very well for Covid screening. Michael Mina is extremely knowledgeable on this topic and particularly vocal about it.
It's important to note here that PCR / antigen discordance early in an infection is not evidence that antigen tests aren't working. Because antigen tests are less sensitive than PCR, they are good at detecting people who are infectious, but not people who are infected but not infectious. Mina has an excellent explanation of why symptoms often start several days before people become infectious.
This post presents new data, which is interesting (especially from a perspective of real-world failure scenarios), but this is weak data: it isn't peer reviewed, there's no study protocol, and the results are inconsistent.
It seems to me the correct interpretation is to update slightly in the direction of antigen tests working less well than previously believed, but to continue to believe they are useful and effective (but far from perfect). Am I missing something?
This is a really good start, and I look forward to the inevitable improvements in the quality of discourse. But to fully leverage the potential of this exciting new system, I think you should create a futures market so we can bet on (or against) specific individuals writing good posts in future.
Also: from now on, I vow to ignore any and all ideas that aren't supported by next-level puns.
When I had an acute bout of insomnia, one of the things I found most helpful was listening to sleep-focused bedtime stories. The key part wasn't the sleepy imagery, but rather just having something boring and inconsequential that my mind could latch onto, to replace the busy inner dialog.
I particularly like the stories from the Headspace app—they're slightly randomized each night, which prevents you from using the story progress as a timer ("Oh no, we're up to the skunk already and I'm still not asleep!")
Related: I also found it extremely helpful to get rid of my bedside clock and to use a smart watch for sleep tracking rather than keeping track of my sleep manually. Worrying about sleep makes your sleep worse, and keeping track of how you're doing tends to feed the sleep anxiety.
I'm excited about this sequence, and look forward to the rest of it.
Having just read this introduction, it almost feels to me like the tail is wagging the dog. I completely agree that relinquishing options is a critically important part of civility. But my instinct is that the relinquishment is in service of a greater (and defining) goal, not the goal itself. So, something like "civility is prioritizing cooperation over autonomy, which in many cases requires relinquishing physically possible options". But I assume it will all become clear as the sequence proceeds.
Thank you for this.
I want to start by echoing your gratitude to the microCOVID team: they've done amazing work and the tool they've produced has been incredibly valuable to me and to many others. And I agree with your assessment that microCOVID is much less useful than it has been in the past. I'll add to your points:
- Their calculation of prevalence strikes me as far too clever. I understand what they're trying to do, and it makes sense in theory. But Covid surveillance is something I know a lot about, and I believe they're over-driving the data. Test positivity seems like a really valuable signal, until you understand what goes into it—once you do, it seems much less useful. In my own modeling, I use (current case rate) * (under-reporting constant), with a manually determined adjustment for trend. But I'd rather use the current level than the fancy extrapolated level used by microCOVID.
- They're using significantly inaccurate constants in some places that matter. For example: their household Secondary Attack Rate (SAR) for a fully boosted person is 15%, but a better value would be 25% paper 1, paper 2.
I don't know the answer here. We need a tool like microCOVID, and I understand how hard it is to maintain a volunteer-based tool.
Thank you for the reminder to explain and not scold—I shall strive to do so.
I'd caution you against spending too much time diving down infinite crank rabbit holes: true believers will always find some new detail or theory for you to rebut. At some point, if someone is committed to denying the clear scientific consensus, there's no point trying to get through to them.
At a high level, we have a pretty deep understanding of how covid vaccines work and how they perform over time, and there's absolutely nothing in there to suggest that, unlike every other vaccine ever, covid vaccines display the bizarre transition from positive protection to negative protection that you're asking about.
Vaccine effectiveness declines over time because (in large part) antibody levels wane over time. That's very well understood and in no way unique to covid vaccines.
Protection has shifted from protection against infection to infection against severe outcomes because of antigenic drift: the vaccines are most closely targeted to the ancestral strain. That match is most important for antibody protection: since antibodies are critical to protection against infection, the vaccines produce significantly less protection against infection as the virus drifts further from the ancestral type. T cell immunity is less affected by antigenic drift, so their protection against severe disease isn't as attenuated.
ADE is a real thing, and it was a concern early on. In particular, there was a feline coronavirus vaccine some years ago that triggered ADE, so there was concern that covid might have similar issues. But we've seen no sign of that.
Original antigenic sin is also a real thing, but wouldn't produce the effect you're asking about.
Two pieces of advice:
- Get your next vaccine. It's incredibly safe and incredibly effective.
- Spend less time engaging with lunatics on the internet. Too much time listening to cranks is bad for your epistemic health.
To a greater or lesser extent, I think that's true for many of us here. Which is a good thing in some ways, but can make it challenging to fully understand and engage with people who are more hive-oriented.
The key thing is that it's low-commitment / low-guilt. I was inspired to start it by a friend who started a book club during the pandemic, fell catastrophically behind on the reading, and ultimately ended up ghosting her own book club.
I've noticed that book clubs tend to become machines for making people feel guilty / overloaded, so I tried hard to avoid that. We do a book every 2 - 3 months, and the default expectation is that people won't attend unless that specific book is interesting to them.
Shortly before the discussion, I send out a summary of the book (which was my motivation for writing this), so that people can attend and participate without needing to finish (or even start) the book.
It's still a fairly new endeavor, but it seems to be working so far.
As a side note, I run a thing that's like a book club but different and we're talking about The Righteous Mind on Saturday 1/22. We have room for a few more thoughtful people—feel free to message me if you're interested.
Vaccine-required zones seem unworkable to me: ours is a highly connected society and it's common for a single household to have members who have jobs / school separated by many miles. Self-sufficiency is completely impossible in the modern world—the closest example is probably North Korea, but that's probably not a model we want to pursue.
There are also immense transaction costs here: there's no area where everyone wants (or doesn't want) to be vaccinated, so implementing this would require massive migration, with immense costs.
It seems to me you've hit on one of the most interesting and challenging things about Covid policy (at both a government and a household level): many of the usual libertarian-ish solutions don't work here, because of the difficulty of keeping one person's choices from impacting everyone around them.
I'm afraid I only have time for a short, partial response today. Short version: Covid surveillance is hard, and there's lots of noise in the data. But there are lots of smart people working hard on this, and in the aggregate we actually have a pretty good idea what's going on.
I'll address one of the questions you asked specifically:
So where are these numbers for variant spread coming from? Maybe hospitals do have special genetic tests and reliably do those? But then isn't there going to be a pretty strong bias based on the fact that these are only for people who are getting hospitalized?
In Washington, much of the variant prevalence data comes from UW, which sequences a subset of the samples they receive. This is a bit complicated: some samples are fully sequenced, and some are tested for S-Gene Target Failure, which is a faster, easier test that is a fairly good (but not perfect) proxy for Omicron vs Delta. The UW sequencing is a good but not perfect sample of what's actually happening in Washington. For details on this project, the person to follow is Pavitra Roychoudhury. Details vary, but there are multiple other institutions with largely similar programs.
More general answer: you're asking good questions. They are all important, and they're obvious to any smart person who thinks about the issue for a moment. Although I don't have time to answer them all, I assure you that the smart people working on Covid have thought of every single one of your questions, and have good answers to every single one. Many of the answers are in Zvi's excellent series of Omicron updates.
Thank you. This helped me think more clearly about something we do often.
Zvi's Omicron summary is probably your best source of information:
Omicron probably milder than Delta (~50%) so baseline IFR likely ~0.3% unless hospitals overload, lower for vaccinated or reinfected.
Good questions—thank you for starting this conversation.
Your assumptions about testing seem reasonable, and hopefully we'll have confirmatory data soon.
I have with great regret stopped using microCOVID. A factor of 2 - 3 x risk multiplier seems reasonable, but I no longer entirely trust their transmission model. It's probably still more or less valid, but Omicron is a very different disease. There's some interesting data about it preferring the upper respiratory tract to the lungs, and about how Omicron particles behave differently in aerosols, that make me worry that transmission patterns may have changed in ways that are more complicated than a simple multiplier.
I'm hoping to see more data soon (and especially hoping that the microCOVID team will update for Omicron, although that seems somewhat in question).
Excellent question!
My best guess: for detecting people who are infectious but asymptomatic, antigen tests will likely perform approximately as well with Omicron as they have with Delta. Because Omicron infections ramp up so fast, however, I'm reducing my guess for how long you can trust the results from 12 hours to 6 hours. (That is to say, if you tested negative this morning, you shouldn't assume that you aren't infectious this evening).
In addition to the data you cite, Abbott claims their testing shows no decrease in BinaxNOW effectiveness against Omicron. That would also be my prior. I'm curious what the FDA has found, although they've been coy about releasing details. I assume we'll see more data soon.
We could certainly have done much better (both before and during the pandemic), but unfortunately it isn't as simple as just giving IBM $100M. Any solution needs to fit into the vast array of other existing systems used for reporting lab results, managing medical records in hospitals, etc.
The US delivers health care in a very patchwork way, which has made the deployment of electronic medical records very slow and difficult. Strong, smart leadership at the top would help a great deal, but even in the best possible case, really fixing this problem would take many years.
microCOVID has been a game changer for me and many people around me: the ability to get quantitative risk assessments radically improved our ability to efficiently spend risk. We recently stopped using it because of Omicron, and I'm very sad about it.
To me, one of the coolest things about microCOVID has been the proof of concept that a group of smart civilians can put together a useful tool that significantly shifts the efficient frontier for navigating Covid. That alone seems valuable to me, and I'd love to see the project keep going as a testbed for how to make similar projects succeed in future.
But, like most volunteer projects, it seems to be slowly sinking beneath the waves. I don't know what, if anything, could change that. A $50,000 grant from ACX? Providing an easier on-ramp for new volunteers? Some kind of Y Combinator for rationalist projects?
I can't speak to San Francisco specifically. But if it's anything like many other locations in the US, the problem isn't malice or indifference: it's that generating this data is vastly harder than you realize. The politicians get the data the same time you do: as soon as it's ready.
Here's one tiny true example, from one part of the pipeline in one particular location. A substantial amount of data enters the system as faxes. The faxes go to a room full of National Guard, who manually enter the data into computers, from whence it begins a complicated process of validation and de-duplication before it enters the main pipeline. You can imagine that this system doesn't scale particularly well as case counts rise.
At a broad scale, what's happening is that an immense amount of data is trying to enter a legacy system that was designed for less than one percent of its current load. Some of the data comes from sleek modern hospitals with state of the art medical informatics systems. And some comes from computer illiterate rural doctors, and some comes from nursing homes that had never reported lab results before Covid, and some comes from employers who test their employees, and some comes from private labs, and some comes from sovereign tribes that have complicated data sharing agreements with the state, and...
If I can find the time, I might write a post explaining in more detail how surveillance data is generated and processed. But for now, I assure you this problem is incredibly hard. Update: here's the post
Important disclaimer: my opinions are mine alone and I don't speak for any government agency.
Rapid antigen tests at the door reduce risk by about 75%, assuming people are asymptomatic and you test each day if it's a multi-day event. I did a deep dive on antigen tests recently, if you'd like to see the data.
PCR is probably similar: they're much more accurate, but the data is more stale, which especially with Delta is a significant issue.
That is precisely the question, and I confess that I don't know the answer for certain. I think, though, that both factors are important.
The issue you're talking about is definitely a thing: influenza evolves rapidly enough that any given vaccine will become less effective over time simply because the dominant strain of the virus has drifted.
However, I believe it is also the case that the immune response drops off fairly quickly. I haven't found a definitive source (I confess that I didn't look hard), but the closest I came is this article, with this quote:
"My informal sense of the literature [is] that the suggestion is strong enough that if people could reliably get vaccinated the week or two before the flu season starts, they'd be better protected," Marc Lipsitch, PhD, a professor of epidemiology at Harvard University, told CIDRAP News. Lipsitch also penned a commentary on this study. "The more complicated thing is the trade-off between putting it off and not doing it at all," he said.
My interpretation of that is that he's talking about a benefit from getting the identical vaccine closer to the start of flu season, so that flu season hits while the immune system is at maximum activation.
That makes sense—it's also true that the efficacy of the flu shot declines over time (maybe 8% - 10% per month?), so there is significant concern about getting it too early. I could certainly see making an argument for getting one as soon as possible and a booster shot in the mid to late season. That's a single shot with a booster, technically, not a two shot series.
Taking the "wait" argument to its logical extreme, it seems to me one could argue not only waiting for kids to get vaccinated, but waiting until COVID rates are minimal, so immune compromised people can safely attend.
I don't think it's necessary or appropriate to take everything to its logical extreme, but it seems to me that if one is going to advocate waiting for one group but not another, it's important to clearly articulate the moral principle behind that distinction.
I'm not a dancer, but my instinct is that a "reasonable accommodation" model is appropriate here: there's a moral imperative to make events as accessible as reasonably possible, but not to cancel any event that isn't 100% accessible to every person.
I would doubt it—different vaccines provoke different immune responses, and each has a dosing schedule based on empirical evidence about what produces the best response. The fact that two doses are needed for an optimal response from the covid vaccine doesn't tell you much about any other vaccine.
It's possible that two vaccines would produce a slightly better response but they decided the cost/benefit didn't pencil out, and I could imagine that for some immune compromised people, getting two would be appropriate. But I'd stick to the recommended schedule absent strong reasons for doing otherwise.
Caveat: I have a strong layperson's understanding of vaccines, but I haven't looked at data specifically for the flu vaccine.
Really solid analysis. Regarding rapid tests:
A pretty important downside in many cases is that they're logistically complicated at a large event. The tests need to lie flat on a table or other surface for 15 minutes. Are you gonna have a giant table covered in tests? Do people come in to test and then go back out? Do they take their test out and perform it in their cars? These are solvable problems, but they can add a lot of complexity and crowding to the checkin area, which is already a problem spot at many events.
With that said, I'm a huge fan of tests for smaller events. Rapid tests let you get four times as much socializing for the same level of risk—depending on your risk budget, your financial situation, and your social ambitions, tests might (or might not) be a game changer for you.
OK, did some digging. The relevant source is table 2 from Peng et. al., Practical Indicators for Risk of Airborne Transmission in Shared Indoor Environments and their Application to COVID-19 Outbreaks.
They calculate the following relative risk rates:
Silent: 0.0012 (1x silent rate)
Speaking: 0.0058 (4.98x silent rate)
Shouting / singing: 0.0350 (29.91x silent rate)
Heavy exercise: 0.0817 (69.83x silent rate)
IMHO, you could may argue for a risk factor of 1/10 compared to heavy exercise (which is 7x the silent rate), but my gut is that 1/5 (14x the silent rate) would be more likely, and something like 1/3 (23x the silent rate) would be a better and more conservative choice
If you can afford them, the rapid tests are a great idea: microCOVID doesn't model them, but I believe they cut your risk by about a factor of 4.
Yes, I emphatically agree with this (as does my consultant, who is an epidemiologist who works on COVID full time).
I think one can reasonably argue about the details: loud vocalizations create different aerosol patterns than exertion, and off the top of my head I'm not aware of any really solid data on how the two would compare. But I think your numbers are low by at least a factor of 5, and a factor of 25 seems very plausible to me.
Also: you've selected surgical masks when doing the µCoV calculation. Will that actually be true? If most people wear cloth, thin or loose (which seems most typical here), that'll increase the risk by another factor of 4.
Excellent data: thank you! Two things to keep in mind:
- The comment on page 5: the study was "Not powered or designed to compare between the groups"
- They're only looking at antibody levels (because those are relatively easy to measure), but there's a good argument that some of the differences between strategies will involve activation of B cells & T cells.
See also the limitations on page 33.
Excellent question, and I think a lot of us are wishing we had more data on this—unfortunately, there is very little data so far. But here's my take:
- If you had J & J for your first shot, I think there's enough evidence now to say it's probably (p = 0.7?) better to get Pfizer or Moderna for your booster.
- If you had Pfizer / Moderna for your first two shots, my instinct is that J & J might be the better choice, because there's an argument from microbiology that mixing types might produce a more robust response.
- If you had Pfizer / Moderna and want an mRNA shot for your booster, I don't think it'll make much difference which one you get: they're very similar.
There's an argument to be made that absent strong reasons to do otherwise, it's best to follow standard practice (in this case, to get the same brand of booster as the original shots) simply because you'll be in a larger, better-studied cohort.
A couple of sources, such as they are:
“But something has really become clear: The mixing really is most impactful when you have a DNA/adenovirus vaccine first followed by the mRNA vaccine,” Gandhi said. WaPo
The study’s researchers warned against using the findings to conclude that any one combination of vaccines was better. The study “was not powered or designed to compare between groups,” said Dr. Kirsten E. Lyke, a professor at the University of Maryland School of Medicine, who presented the data. NYTimes
Yes, accuracy in antigen tests seems to correlate very strongly with viral load (and presumably therefore with infectivity). This paper found 100% agreement with PCR for Ct 13-19.9 (massive viral load), all the way down to 8% agreement for Ct 30-35.
Ct (cycle time) measures how many amplification cycles were needed to detect nucleic acid. Lower Ct values indicate exponentially more nucleic acid than higher values, although Ct values are not standardized and can't be directly compared between testing facilities.
Thank you for this! I have a few thoughts about antigen tests.
1: I'd recommend the BinaxNOW as the "standard" home antigen test in the US. Broadly speaking it's better studied, more accurate, cheaper, and more widely available than the others. Regarding data...
2: I think the best current source of general data on home antigen tests is this meta analysis from September. The results from multiple papers over the last year have been pretty consistent, but this adds a little more power to the numbers. They come up with:
Overall: sensitivity 68%, specificity 99-10%. Sensitivity for symptomatic individuals: 72%. Sensitivity for asymptomatic individuals: 52%
Sensitivity for Ct < 25: 94%, Ct > 3: 30%. (I'll be writing more about these results in a bit, but the short version is that this strongly supports the belief that test sensitivity depends strongly on viral load and will be highest during peak infectivity).
3: Two additional excellent papers are this one for subgroup analysis and this one for subgroup analysis and discussion of how user error affects accuracy.
4: Related to the above: accuracy seems strongly correlated with viral load, which strongly suggests multiple tests on the same individual at the same time would be highly correlated.
Yes, those are all excellent points.
I wrote this as a side reference for a deep dive on the BinaxNOW that's coming shortly, and it'll dig into the numerous, complex, and important issues affecting BinaxNOW accuracy. Short version: the accuracy varies substantially, largely based on viral load. And you're correct that repeated tests on the same individual will be strongly correlated.
And you've convinced me to change the example you cite: I'd gone with the first person for narrative consistency, but I'm shifting it to prioritize technical accuracy.