“Infections in vaccinated Americans are rare, compared with those in unvaccinated people . . . But when they occur, vaccinated people may spread the virus just as easily.”

Posted on by

Dean Eckles writes:

Thought you might like this example from the leaked CDC slides. One of the big claims being repeated in the media is that “Infections in vaccinated Americans are rare, compared with those in unvaccinated people, the document said. But when they occur, vaccinated people may spread the virus just as easily.” (NYT) That is, this focuses on possible equivalence (vs. not) within some subpopulation who get infected. And, of course, the vaccine affects who gets infected and whether it gets reported and included in the sample.

This is apparently based on the results on this slide:

The first bullet is a comparison within vaccinated people who have reported breakthrough cases. Based on 19 such cases with Delta, this suggests ~10 times increase in viral load associated with Delta. (One widely reported comparison, cited by Dr. Fauci earlier this week, in viral load for Delta is ~1000 times, so this would actually be much lower than that.)

The second bullet is a comparison — for one particular outbreak — of vaccinated and unvaccinated cases in a cluster associated with Provincetown’s extensive July 4th parties. It seems like an interesting question here is whether this conditioning on a known infection makes sense.

(Other outlets focus on a different dichotomization of these results, saying for example, “New data suggests vaccinated people could transmit delta variant” as if this is new information at all!)

To me, this all gets at how valuable it is to think about things in degrees (if not fully quantitatively) and comparatively rather than reducing everything to 0 or non-zero.

Obviously, I am not an infectious disease biologist, but this seems like a nice example of dichotomization, conditioning on post-treatment variables (which can sometimes make sense — does it here?), and science communication.

Interesting point about conditioning on a known infection. The implicit causal model in the comparison is that the infection is something that just can happen to you, but I take Eckles’s point to be that, if you know that a vaccinated person was infected, that fact tells us something about that person—some combination of behavioral and biological information that we would expect to be relevant to the rate at which they spread the virus. Thus, it could be true that infected vaccinated people spread the virus as easily as infected non-vaccinated people, but that statement could be rephrased from a latent-variable perspective as “the sorts of vaccinated people who are likely to get infected are the sorts of people who are more likely to spread the virus,” without necessarily implying that the effect of being infected on spreading the virus is the same among vaccinated and unvaccinated people. I agree with Eckles that these question can get very tangled.

49 thoughts on ““Infections in vaccinated Americans are rare, compared with those in unvaccinated people . . . But when they occur, vaccinated people may spread the virus just as easily.””

  1. The thing that I think people are really overlooking with the second bullet (vaccinated vs. unvaccinated viral loads) is that, if you look at the full MMWR, of the identified vaccinated cases 79% were symptomatic, while of the unvaccinated 58% were symptomatic.

    This makes sense to me as the vaccinated are probably less likely to get tested unless they have strong reason to believe they got infected, like symptoms. We KNOW from other data they’re actually less likely to get symptomatic infections, so I suspect there’s a really really strong selection effect here.

    But the real point here is this suggests comparing the viral loads in these identified cases is totally invalid for making any causal claims about the vaccine. You can’t compare a 79% symptomatic vaccinated group with a 58% symptomatic unvaccinated group and claim that their viral loads being similar is evidence that the vaccine doesn’t effect it EVEN CONDITIONAL ON BEING INFECTED AND IDENTIFIED because we know symptoms are correlated with viral loads.

    I’m just completely baffled how the CDC interpreted this as evidence of even the relatively modest claim that vaxed and unvaxed infections have similar viral loads in “comparable” infections. It’s like they looked at death rates in crashes with seatbelts at 50 mph and no seatbelts at 30 mph and somehow came to the conclusion that seatbelts may not work as well as we thought.

    Reply ↓

    • Zach:

      Yes, it’s frustrating. I think part of the challenge is that it’s easy to see the problems with direct data analysis but it’s not so easy to come up with a good adjustment. I mean, sure, that’s what professional epidemiologists are supposed to do—it’s part of their job—but saying this doesn’t make it easy. Bob Carpenter and I have been working with a group of public health analysts in the U.K. and we’ve been struggling with how to deal with selection bias in who gets tested. It’s clear that selection bias is a big issue but not so clear what to do about it. Throwing up our hands and giving up isn’t a good option.

      Reply ↓

      • I agree and have a ton of sympathy for that. We need to do the best we can and bend the rules more in emergent situations. But I think making a claim like this about the vaccines from that outbreak’s data really wasn’t justified and has caused a bit of a panic (at least on social media, maybe not among normal people). It just wasn’t somewhere we needed to go or that was supported by this particular data. But that’s something different people can come to different conclusions about, I suppose.

        Reply ↓

      • And for my specific issue I actually do have a proposal for a simple solution: stratify each group by symptomatic and asymptomatic and calculate standardized Ct values (that is, adjust for presence of symptoms and see if the Ct values are still similar).

        Reply ↓

      • RE: “It’s clear that selection bias is a big issue but not so clear what to do about it. Throwing up our hands and giving up isn’t a good option.”

        Surely there is a certain class of problems for which “giving up” is the principled response, though. I mean “giving up” in the sense of admitting that no amount of sophisticated statistical manipulation will turn a dataset with extremely biased selection into the functional equivalent of a random sample.

        It’s perhaps an open question whether COVID test results data falls into that class but it sure looks like a possibility to me.

        I would find it entirely plausible to think that it is not possible to sufficiently understand population rates of infection, viral loads, transmissibility without actual population sampling and testing. Not that such a thing is ever going to happen in today’s politicized world.

        Reply ↓

      • “Throwing up our hands and giving up isn’t a good option.”

        It’s not?

        We’ve seen time and time again that data analysis has failed in almost every application and respect throughout this pandemic. FFS science can’t even say if surgical masks are effective or not!!! Giving up might not be a good option for statisticians, but giving up on data analysis and resorting to intuitive assessment of the evidence is the blatantly obvious option for conducting public policy. Unfortunately the science establishment has been selling and drinking data koolaid for so long they’re overcome with credulity at anything that comes out of a computer but no longer capable of using non-quantitative evidence effectively.

        The vax makers in science have been wildly successful but the contribution of university scientists to resolving the pandemic is embarrassing.

        Reply ↓

        • Jim –

          > We’ve seen time and time again that data analysis has failed in almost every application and respect throughout this pandemic.

          I dunno. That’s true if you take a binary approach: You cindyct perfect analysis that nets clear and unarguable conclusions (even with those who are ideologically committed towards a particular answer) or you “failed.”

          > FFS science can’t even say if surgical masks are effective or not!!!

          No – but people study yhe issue abd provide evidence that helps inform us about probabilities. In a world filled with uncertainties that’s often better than nothing.

          > Giving up might not be a good option for statisticians, but giving up on data analysis and resorting to intuitive assessment of the evidence is the blatantly obvious option for conducting public policy.

          Is there really such a mutual exclusivity?

          > Unfortunately the science establishment has been selling and drinking data koolaid for so long they’re overcome with credulity at anything that comes out of a computer but no longer capable of using non-quantitative evidence effectively.

          I might say that people having unrealistic expectations for so long has led us to a state where people are so overcome with binary expectations, that collective action to address imminent threats had become damn near impossible.

          The problem with the practice of avoiding unintended outcomes by not taking action – unless outcomes are positive with 100% certainty – is that it can have unintended consequences.

        • When decisions need to be taken, urgently, there *is* something which is or used to be called “leadership”. Leadership is different from “infallibility”. There are pretenders who cannot or will not lead because they suppose they must be infallible; and there are pretenders to infallibility who may have a devilish knack for leading! The former dress up their pusillanimity in the guise of disinterested quest for absolute truth. The former dress up their recklessness and stupidity as courage. And then there are the non-pretenders; who do not pretend to certainty that cannot be had; but who take seriously the burden of sailing the boat even if the way ahead looks grisly.

    • Zach –

      > I’m just completely baffled how the CDC interpreted this as evidence of even the relatively modest claim that vaxed and unvaxed infections have similar viral loads in “comparable” infections.

      Your point is well taken – but are you sure that the data from that one preliminary report was the only information that they used to draw their conclusions?

      Reply ↓

      • I’m not sure at all. But I haven’t seen anything else cites in the media or among the experts I follow. I’d love to be pointed to any other evidence anyone else is aware of, though!

        Reply ↓

      • It can be backed out from the numbers presented. There are 469 (total cases) – 346 (vaccinated cases) = 123 cases who are not vaccinated. There are 346 symptomatic cases of which 274 were vaccinated, leaving 72 symptomatic cases in the unvaccinated group. 72 / 123 = 58.5%.

        Reply ↓

    • Are you frustrated by the causal interpretation, the lack of mentioned normalizations/conditioning (DENOMINATORS), or both?

      I’m a little confused about the analogy to car crashes though. We know that seatbelts always work up to a certain speed. But for a disease, isn’t it possible that there is a subpopulation that is infectious regardless of their vaccination status?

      I agree that this doesn’t mean that vaccine doesn’t work. But you analogy suggests (to me) that the infectious people with the vaccine must have a worse infection (aka driving 50mph). Why can’t it be that they are just, for some reason, unable to benefit from the vaccine?

      Of course, I don’t know if this matters in the end, because infectious is infectious.

      Reply ↓

  2. I looked at this study this morning. I was hoping to look at the data. Surprise! It’s not there. Given the importance of this study, I find it bewildering (baffling, puzzling, infuriating,…) that the data is not released. What is the excuse?

    I think this case perfectly illustrates the issues with making data available. I can easily think of reasons to withhold the data, and I can just as easily think of reasons why it is important to do so. For me, the latter dominates the former by a wide margin, given the way this study is being used to inform policy.

    Reply ↓

  3. Is the big interest on that bullet not more about the implications on what measures make sense? I.e. if the vaccinated can get infected (it just happens less often), but would spread the virus much less even once you condition on them having been infected, then that would put in question why one would ask the vaccinated to mask (it would then be more about their own personal risk). On the other hand, if they can still infect others as much as other infected people (or at least to with a pretty high rate), then asking them to mask (or take other precautions) makes more sense.

    Reply ↓

    • I’ve done several back of the envelope calculations based on multiple sources (israel, UK, LA county stats), and it’s clear to me that the vaccine effectiveness against initial infection is not super high. I don’t know what the number is exactly but a prior of between 30 and 60% seems reasonable which is a lot lower than against Alpha variant where it was probably in the 80%+ range. So the late spring/early summer summer plunge was caused by vaccination being highly effective at cutting R down, while the current boom is at least in part because vaccinated people are acting like they are fully immune but they’re not, and are spreading the disease because they’re out and about with masks off doing things they think are safe. (and of course also because unvaxxed people are out and about spreading it as well, but we’re seeing booms even in highly vaccinated communities so it can’t be just unvaxed). The overall picture is that even with just documented prior infections + vaccination rates… we should be at or near herd immunity in many places that are experiencing booms. So it can’t be the case that immunity to initial infection and spread is super high.

      So far among my various contacts I know of several cases of fully vaxxed people getting COVID with symptoms. That suggests there are probably several others who had COVID without symptoms. Right now in LA county there are something like 22 cases per 100k people per day, and of course there are many more infections, esp. asymptomatic ones.

      So far, no one I know of who was fully vaxxed needed hospital care, or even oxygen at home. most of them have had around 7 to 10 days of fever, chills, cough, aches, lying in bed, etc. There’s no question in my mind that the symptomatic were likely infectious for multiple days.

      What concerns me most is that where I live, the Pasadena Unified School District is planning a return to 100% in person schools (with independent study, different from the video conference classes of before, as an option that they expect “a few hundred students” for). Right now, if you assume vaccine effectiveness of about 50% against infection, and around avg 5-7 days of mean transmission time, you get an R0 of 3.5 to 4 or so for Delta variant in the general LA community. Since there is a fair amount of immunity out there, the current reproduction number in LA county is actually something like 2, but as soon as you pack a school with 100% unvaccinated under 12 year olds, you’ll get something closer to the R0 value of 3.5-4. With mitigation measures you will probably reduce that, but I still get that with R0 of 2 you’ll swamp schools with hundreds of infections before the end of the first month.

      Needless to say my family will be choosing the independent study option for our under 12 children. I’ll let others choose to experiment on their kids :-(

      Reply ↓

      • > I don’t know what the number is exactly but a prior of between 30 and 60% seems reasonable which is a lot lower than against Alpha variant where it was probably in the 80%+ range

        It was only about 30% vs alpha. Compare routine testing vs. routine testing, symptoms vs. symptoms, and contact vs. contact. This is the only paper that provided data on that:

        https://pubmed.ncbi.nlm.nih.gov/34282839/

        The most basic assumption of a test negative design was violated by dozens of studies, which misleadingly overstated the VE by 2-4x and lead to very misleading info given to policy makers.

        Reply ↓

        • I’ll have to take a look later but 30% against Alpha doesn’t seem consistent with the large drops in R between Feb and Jun. Unless you mean that infection occurred but it was so mild as to produce no symptoms and have very low chance of reproduction.

          But I’ll look at the paper later today thanks for reference

        • From Table 2. Reasons for testing.

          We must compare the results due to a routine test in vaccinated vs. not, then tests due to symptoms, etc. A test-negative design assumes vaccination has no effect on the rate of, and reason for, testing. As seen in that table, this assumption is not even close to correct.

          This was well-known and uncontroversial prior to the use of this design for Covid-19:

          > For example, if clinicians order the diagnostic test for those with severe ILI and those who did not receive the vaccine, the proportion of non-vaccinees among cases is likely to increase, resulting in overestimation of VE. This translates to selection bias and it is impossible to estimate its extent or direction once such a bias is introduced.

          https://www.sciencedirect.com/science/article/pii/S0264410X1730899X

        • Thanks. I agree that selection bias is an issue. (By the way, the numbers I quoted are for “fully vaccinated” two weeks before while that table considers vaccinated those with just one dose.)

  4. @Zach:

    “I’m just completely baffled how the CDC interpreted this as evidence…”

    … jt’s baffling because you assume incorrectly that official ‘evidence’ is factually accurate– and that the CDC is a highly competent and unbiased analyst of the ‘evidence’ it chooses to collect.

    The public is flooded daily with a mass of supposed statistics and evidence-based conclusions about COVID/Vaccines — most all is very suspect for accuracy.
    It is extremely difficult to accurately measure COVID infection incidence in mass populations.
    It is extremele difficult to accurately measure COVID Vaccines effectiveness in mass populations.

    The CDC has been a highly politicized agency for decades, demonstrating strong bias on diverse issues and clearly pushing partisan agendas via disinformation.
    CDC stopped closely tracking “Breakthrough’ infections last May and had previous stopped openly reporting Breakthroughs — in biased effort to promote COVID vaccination rates, by concealing negative information about the vaccines.

    Whom do you trust for COVUD statistics?

    Reply ↓

  5. As a guess, I would think that conducting antibody tests for the Provincetown individuals who were vaccinated and infected. It would have been helpful toward discerning how effective or not the vaccine is.

    One my friends told me that several vaccinated young men have come down with the Delta variant, in the last week. Two of them had been infected with the Alpha variant in fall 2020. They were vaccinated in March or April of 2021. One of them has quite severe symptoms, requiring steroids and other prescriptions.

    Reply ↓

  7. What’s the relationship between viral load and transmissibility? I’m sure there’s a correlation but I don’t know how strong that correlation is. Is viral load the only non-environmental factor that contributes to transmission? I remember earlier in the pandemic I attended a talk where the speaker noted that there was an anomaly of children reported as having equal or higher viral loads compared to adults but were observed to be less transmissible. Not sure if that anomaly has been resolved since then but it just made me wonder how to interpret the “may” in “But when they occur, vaccinated people may spread the virus just as easily”, even if you were to assume all the estimates/evidence did not have any issues.

    Reply ↓

    • I think viral load should be highly correlated with transmission at least for the first 5-10 days of infection. Once a lot of neutralizing antibodies are present the viral load could be high but coated in antibodies so it can’t infect.

      My assumption for the mechanism of transmission is basically viral load -> respiratory droplets -> aerosol spread -> inhalation -> infection

      when you inhale the same number of droplets but the number of virions is higher, you have higher chance of infection. There’s going to be some “infectious dose” and when concentrations are higher, then fewer droplets are needed to infect.

      Reply ↓

      • > Once a lot of neutralizing antibodies are present the viral load could be high but coated in antibodies so it can’t infect.

        Vaccinated people produce the antibodies way more quickly than un-vaccinated, right? So, again taking that CDC slide at face value, it’s possible that the even if the observed viral loads are equal between vaxxed and un-vaxxed, the “effective” viral load in the former could be substantially lower because of the antibodies. Am I understanding that correctly?

        Reply ↓

        • It has been unclear to me whether the antibodies produced from the vaccine are as potent or not as potent as the antibodies from natural infection. I have been trying to locate a Twitter thread which raised this query. My recall from the Twitter thread is that natural infection could bestow broader immunity for mutating coronaviruses.

          If antibodies wane after some period, then the risks increase, I infer from some experts: although, subsets suggest that T-cells can kick in to kill the virus. It would be informative for us to access credible virologists to weigh in.

        • Yes vaccinated people should have sorter duration infections and shorter duration of infectivity. How much less infective is not clear. If most people spread pre symptom anyway… It might not budge the needle on R very much to be vaxxed. It will however dramatically move the needle on severity. My guess is that lots of people will get Delta even after vax. That’s what happened in UK after all. But very few hospitalizations or deaths in the fully vaxxed community. So the model should be “get vaxxed because you’re GOING to get the virus now…

        • Daniel, Yes, my guess is that lots of people will get Delta as in the UK.
          One can hypothesize that the majority who contracted COV-2 Alpha variant did not get hospitalized or die either As you know, there has been a steep age gradient for the Alpha variant. Is there a steep gradient for the Delta? I gather that the CDC is tracking spread in nearly all counties. It’s now making public some data.

          Even moderate symptoms can disable someone: that is those symptoms that are short of requiring hospitalization.

      • > Once a lot of neutralizing antibodies are present the viral load could be high but coated in antibodies so it can’t infect.

        Incorrect, it has been known since the 1960s that intramuscular vaccines do not trigger mucosal immunity. There are only very low levels of neutralizing antibodies, if any, in the mucosa:

        https://science.sciencemag.org/content/373/6553/397/

        As expected, SARS-2 is the same:

        https://www.nature.com/articles/s41591-021-01318-5

        Reply ↓

        • You are misinterpreting my point. I didn’t mean from original vaccination but rather the rebuilding of neutralizing antibodies during infection. A vaxxed person who is infected presumably produces neutralizing antibodies at some point. Let’s pretend it’s 8 days after infection. The unvaxxed person maybe has them by 15 days post infection. So the vaxxed person may be infectious for a shorter time.

        • Sorry, I see what you meant now. I would also assume that infection can still trigger mucosal immunity in the vaccinated. Probably sooner as well. There is no data on this I have seen though.

  8. At the public level, what we’re seeing is a rebound from the simplistic dichotomy put forward by official sources (the CDC, the White House, compliant media) that the pandemic was “over” for vaccinated people: no meaningful risk of infection (or infections drawing on health care resources) and no meaningful risk of transmission. This was pushed as messaging to encourage vaccination. It was always BS even if, at one time it might have been only somewhat wrong.

    Now Delta has made it impossible to stick to this message. In its attempt to justify retreating from an always-untenable position, the CDC has hyped the difference between Delta and its precursors. Data point: today’s Oregonian, the newspaper for my own town of Portland, has a headline reading, “Study: Vaccination Status Has Little Effect on Variant Spread”. So we go from one overhyped extreme to another. Which is not to say that Delta isn’t a lot more transmissible, just that it moves the needle on what was always a continuum.

    The CDC should have given more nuanced advice about masking and distancing before Delta, and it should be giving more nuanced advice now. I agree with the party line on this blog that uncertainty is difficult to comprehend and communicate, but difficult is not impossible. It *is* impossible if you don’t even try.

    Reply ↓

      • Aaarrggghh..

        … quite incharitable given the uncertainty in the face of a constantly evolving situation.

        Further, no matter what the CDC et al. had communicated and no matter how they communicated it, they would have been criticized by an army of people lined up to find objections and exploit the inevitable sub-optimality (is that a word?).

        We’ve seen this with any scenario where “experts” are communicating with the public on politicized topics, particularly when there’s uncertainty and complex dynamics. Obviously, clinate change comes to mind.

        Many people spend a lot of time pointing fingers at science communicators where, imo, the real issue is the underlying polarization and tribalism.

        Reply ↓

      • It partially aligns. I agree that an amateurish fixation on messaging has been an abiding aspect of the problem, but I’d add two more issues. First, as I said above, at the heart of a lot of communication problems has been recognizing and conveying uncertainty. As we in this blog community know, it *is* hard to characterize uncertainty clearly for a general audience, but there is no ducking the need to do this. Climate scientists, after many years of skittishness, have taken the bull by the horns and, with much continuing internal discussion, have adopted the position that the uncertainty is itself a large part of what needs to be communicated. Public health people should do the same. (Actually, there is a longstanding discussion among some public health researchers on this question, and if you read independent assessments, like Osterholm’s, you will see this, but it hasn’t made it to the official disseminators.)

        Second and more speculatively, I think much of the public health establishment sees its primary mission as lightening the load on health care delivery — as a sort of protective ring. Now of course, achieving this is a good thing, but it should be an intermediate, not an end goal. Incuriosity about long Covid and breakthrough cases that don’t entail hospitalization joins earlier errors on masks and aerosol transmission, both of which had aspects of protecting hospitals. The dismissal of non-hospitalization breakthroughs in particular set up the CDC for this latest debacle.

        Reply ↓

        • Infinite “taking-pains”, compulsive circumspection, old-maidish intolerance for any jot or tiddle of superficial error, and irreducible alienation from the public that is ostensibly being advised …

        • (my off-the-cuff characterization of what I think is the state of mind that is behind the rudderless communications coming from the CDC).

    • How is “nuanced” advice to be given? It seems to me that it is indeed given quite often; and it is to be found in the weasely rhetorical style like “you may wish to consult your health-care provider … [if you are sick in bed but can still get up]” or “If this is a ‘true’ medical emergency please hang up now and dial 911 [if you have fallen down the stairs and cannot get up]”. This is the infuriating dilletantish rhetoric of the power-point smithies who make “nuanced” advices like it was what kept the world turning round its axis.

      Reply ↓

  9. The weasel word “may” in the quoted title is a cardinal symptom of this kind of infinitely hedged affirmative/negative “guidance”. Once you begin to pick up on the prevalence of this “may” (or “can”) you start to see it in the generic medical advice passed around (“you *may* want to consult your ‘medical provider’); you start to see in in the journal article rhetoric — where every single sentence is hedged with a “may” or a “might”. The syndrome is this: conversion of tottering uncertainty into statements ex-cathedra. Absence of confidence converted into expressions of supreme confidence. The producer of this rhetoric can fill all the available space, by qualifying “exactly” how little he is able to say. But the publication credit is the dessert!

    Reply ↓

  11. Andrew-
    Apologies that I’ve not studied this post and its responses fully but I’m most struck by the fact that CDC is at some level making broad national policy based on 20 people on Cape Cod with the sniffles. A tiny cohort that is self selected and showing a modest effect is shaping how everyone in the country is supposed to behave. It’s a problem.

    Reply ↓

Leave a Reply

Your email address will not be published. Required fields are marked *