($1000 bounty) How effective are marginal vaccine doses against the covid delta variant?

Tl;dr: I think we can vaguely guess a 1.5-2x risk reduction from a third shot, maybe pushing 2-2.5x if you get it close to 6 months after the first; for those with less immunity after two shots, it might be more like expected 3-6x, though it’s unclear if one can discern who these are. If you do get a third shot, you should definitely try to get a different kind, as others have said, though unclear if you want to sacrifice efficacy for this.

---------

First, to establish an expected effect size, how good are the first two shots? The first Pfizer/Moderna takes you to about 80% protection, or a 4/5 risk reduction. The second takes you to 95% efficacy, a 4x risk reduction. To compare, Delta variant takes Pfizer from 95% back to 80, so is a 4x risk increase. So most likely, the third vaccine is not going to be as efficient as the second vaccine. So we expect less than a 4x effect; realistically, maybe I expect something like 2x at the outset since secondary boosters are common in immunology and third boosters aren't. I wouldn't be that surprised at seeing 3x or 1.25x though. (Note that this is all about transmission, for a shot of the same type as your first two; probably slightly stronger effect than this in mortality, and mixed vaccinations I'll get to at the end).

There's no released data, so we have to get creative fast. The best empirical evidence I know of comes from this study on how much vaccinations boost your neutralizing antibody (NAb) titers after already having COVID. (NAbs are by no means all of your immunity, but they’ve been shown to correlate surprisingly well with vaccine effectiveness. Hilariously, there are nonetheless papers claiming we have some evidence that NAbs can CAUSE infections to be worse—while this may be true, all I can do is plead for them to make a better case that this isn’t bog-standard Treatments Correlated with Harm [LW · GW].)

Anyways, this study shows limited effect from a second Pfizer vaccine dose if the patient had already contracted COVID! It’s useful because very roughly speaking, getting COVID gives similar titers as a vaccine dose—and COVID plus one dose is similar to two doses, so looking at COVID plus two doses is a good approximation for a third dose. This gives us some limited evidence that a third dose won’t be that effective.

Notably though, it did significantly increase the titers of the least-protected! This seems like a decent reason to get a third vaccine if you are vulnerable. To put very rough numbers on it, my guess is that for the vulnerable it would cut risk by a factor of 3, whereas the group as a whole only gets a factor of about 1.5.

• Calculation: Getting a second dose cuts risk by 4 as we’ve already said; on this graph the first to second shot gives a jump in titers by about 3 logs, whereas it looks like the vulnerable have their titers go up about 2.5 logs from the third shot, so I’ll give this a ~3-fold reduction. The group as a whole only goes up about 1 log on the third-dose-equivalent, which is maybe a factor of 1.5 protection. Both of these figures match well with our original bound of finding less than a 4x average reduction, and quite well with my prior of about 2x!

As a second creative line of evidence, could we figure out anything from the claims from Pfizer that their third dose increases neutralizing antibody (NAb) titers by by 5-10x?

Is 5-10x a lot? Unfortunately, a major hurdle is that every vaccine appears to measure NAbs using different assays, with no translation to efficacy. I really hope I’m mistaken here, because otherwise it seems like every report of increasing titers by X-fold is almost completely meaningless, e.g. this otherwise-nice comparison. But I don’t think I am.

I tried going through other Pfizer studies to see if maybe they at least used a single assay across the board. One Pfizer study says a factor of 1.5 difference in unnamed geometric titers for a strain meant roughly nothing. Another shows about a 4x factor in log2PRNT50 difference between Delta and wild-type, which corresponds to about a 4x decreased efficacy as we know from elsewhere. But this can’t be the same as the above-stated titer, because that would mean a 5-10x reduced risk from a third shot, which is way higher than our bounds. It’s possible that if we escape the log though, this becomes a 16x titer difference for 4x risk reduction, which is at least in the ballpark and would imply a 2x risk reduction from the third shot. I think this is on shaky ground though and I don’t place much stock at all in it, even if it happens to pop out the magic number.

-----

A very important thing here that a few people mentioned: combating waning vaccine efficacy is important. Many studies show this effect, but I’ll cite my favorite that I linked earlier. It predicts about a 2x risk increase (or effectiveness decrease) over 6 months. Another study shows variable waning from natural infection that roughly accords with this, at least insofar as NAb titers from this or similar papers can be interpreted.

There are two questions here: first, would a third vaccine reverse this waning effect? There aren’t data to confirm this yet, but surely the answer is yes—this is how other vaccine boosters work, and all theory points to this being the case. The second question is whether the third vaccine’s natural effect, as described early to be ~~1.5x normally or ~~3x for the vulnerable, would fully stack on top of a waning-reversal. I think the answer is probably no, but with partial effect, closer to fully stacking than not. Part of this answer comes from the idea that the longer you wait between doses, the more effective they are. In total then, I expect a booster after 3 months to give you about 2x risk reduction, and 4x for the vulnerable.

Last, I completely agree with the other people bringing up that getting a different third dose seems better than the same (though the same shot seems still worth it for many, as above). I wish I had gotten my second dose of a different kind—this is significantly because of “original antigenic sin”, the tendency to get antibody lock-in toward whatever strains you experience early. I expect different vaccines as boosters will help somewhat with this problem, though I really can hardly speculate quantitatively about this—all I’ll say is I would be unsurprised to see that taking this strategy with the right vaccine might get you 4x instead of 2x from a third shot. (However, you’d need both an effective vaccine and for it to be substantially different than the first one or two—and I don’t know if Moderna/Pfizer fit that bill. Perhaps I can research this in the future, or how much effectiveness to sacrifice for difference, though it sounds hard and by default I won’t since I don’t think it’s all that decision-relevant.)

Two last thoughts: if Elizabeth is right in the comments about a possible low limit for mRNA vaccines, that’s a major downside. If it weren’t for that, it would almost surely be worth it to get a 3rd shot for most people by the calculus, since Delta is growing so fast and has so little morbidity for the vaccinated that there isn’t much incentive to keep it out of young and healthy circles. (Cf about 4 days of expected lost life from Delta while vaccinated, and getting a third shot would save you 2 expected days at a cost of two days lightly sick. But if there's much cost in the vein of what Elizabeth mentions, that washes out.)

EDITED: Connor's comment below [LW(p) · GW(p)] landed directly in a place where I was ignorant and I've updated to think the right answer is "defer to him on the details of immunology until I read a lot more". 

I feel like what I wrote was basically adequately hedged given my ignorance, and maybe it is interested to read for having some of the picture, but not all of the picture, but if you're just saving your attention for super high value things you can probably just skip it? I get kind of rant-y towards the end about the lack of a proper public health system. Maybe if you like the rant-y bit this is still worth reading for that?

...

My current personal mechanistic model is that for "a given challenge" (that occurs in a window roughly 6-21 days large, with variation between people and also between challenges for the same person in different states of health), people either have a germinal center response, or not. 

If not, then on a second attempt they might have a response again, or not. This is "in general". I've seen studies where they did really good followup with a population for a measles vaccine, and some patients managed to play along with followup 5 times, and only the 6th vaccine caused their immune system to finally begin testing positive for the desired antibodies.

However, my current personal mechanistic model has huge error bars, and numerous asterisks, with warnings about how immunology is a blacker art that computer security, because it basically IS computer security, but for programs written in nucleic acid that are billions of years old, with millions of new revisions spammed into existence every second by a mad god. I can't read the code (at least not most of it, and not easily). I just expect it to contain devilish tricks because: it would be surprising if it didn't.

Under the simple "response or not" model, the reason they did "two tries without bothering to check in between for a response" for the current Pfizer/Moderna design is that this "double tap" protocol ups the official "chance of having the response from the one official protocol" and basically jukes their numbers, so they can brag about a 90% efficacy "if you do it right" and this makes the narrative and the medical protocols and everything simpler. 

If I was rich enough to have a private doctor (I'm not, and I don't even have "a doctor" as such because of moves and paperwork and dealing with insurance paperwork constantly being so so so exhausting) then I could totally see just paying "one competent personal doctor" to acquire and administer one vaccine dose, test for response, then doing another dose only if the antibody test is negative... but also doing a third if the antibody test after the second didn't show that it worked, and so on.

However, under the "response or not" model this binary response is for a specific epitope (or epitopeS if several antibodies form for different surface bits of different proteins... I'm not personally clear on how the immune system picks exactly what folded protein surface part(s) to pick as a target, or why) and so if you have DIFFERENT vaccines that target DIFFERENT surface patches on the same set of proteins (or variants of proteins) then for a specific later challenge by an environmental exposure probably the best fitting antibody for that exposure reacts the strongest and leads charge by the immune system to fight off the invading virus particles and their initial first couple generations of viral babies [LW(p) · GW(p)].

My current working model might be wrong. If it is right, then the best vaccines would target lots of different epitopes selected from numerous possible genetic variants and we would generate and ship them in maybe close to realtime, as a sort of "human culture and institution mediated meta immune system". 

This is how a properly designed public health system would actually work. But the US does not have a public health system basically? It just has a pharmaceutical and medical licensing monopoly maintenance and legal challenge immunization system, that prevents incumbents in the medical industry from being competed into low margins and high consumer health surpluses :-(

If our country was well run, we would already have a delta booster designed and available that could trigger the formation of these new kinds of antibodies, and the reason we don't is because the government is "either evil or incompetent" and the private industry isn't allowed to innovate around this barrier to technologically solving technologically solvable problems. ("FDA delenda est", but that's neither here not there.)

(Maaaaybe someone HAS already designed such boosters, but the FDA and/or CDC threatened them with angry looks and being put on some kind of professional blacklist if they ever admitted this in public on the second go-around because the PR was so bad for them when it came out that the first mRNA vaccine was designed in early 2020?)

Basically, my model is that RaDVaC or other novel vaccine designs COULD improve your response to VARIANTS by increasing the breadth of covid epitopes that your body can recognize and respond to, but that if a given vaccine design triggered a response in you already then another one is unlikely to help much more...

...however I have genuine model uncertainty. Maybe a second exposure causes the specific form of specific immune responses to somehow be stronger or different in a way that I've never seen a clean and clear explication of? Some anecdotal evidence suggests that double-dosed people are safer than single-dosed people, but I've never seen anyone report anything like this with a coherent followup about how or why such anecdotes might occur more (or less) in "different worlds" where (1) single-dosed people are less likely to have had a germinal response at all vs (2) the second dose materially changes the nature of the immune response.

I'm interested in learning a better model, if my model is wrong.

Another thing to point out is that having more antibodies is called "being allergic to things" when the thing you have antibodies for something that isn't actually bad. And it is called an "autoimmune disease" when the antibodies recognize something that naturally is already part of you. Acquiring new antibodies is not totally risk free. All biology is half black magic and full of exceptions because evolution is deep and twisty. Please please please be careful!

If you want to give me the $1000 for this writeup, please give it to johnswentworth [LW · GW] instead.

↑ comment by Connor_Flexman · 2021-07-25T11:06:14.641Z · LW(p) · GW(p)

I think almost all the evidence points in the opposite direction, unless I’m drastically misunderstanding something, which does occasionally happen.

First and foremost, the idea of having a binary seroconversion dependent on germinal center response seems highly contraindicated at best. There are a billion studies showing that single doses of vaccines give some antibody response, and then a second dose gives far more (often 2-3 OOM). For example, 1, 2, 3, 4, etc. This is the point Lanrian seemed to be making, which I think pretty immediately disproves the hypothesis. 

This isn’t just COVID either—many vaccines have this pattern of giving boosters to increase antibody count. And not only does it increase count, secondary responses vastly increase antibody affinity and produce different antibody types, e.g. the primary response is more IgM whereas secondary response produces more IgG and IgA (the latter aiding especially in mucosal immunity). [Citations for this can be found on pgs 413-414 of the Janeway immunobiology book, and I can maybe link pictures.]

On this note: higher doses of vaccine straightforwardly give higher antibody levels from a nonzero baseline, inconsistent with a binary response [1].

Further, amount of NAb present in the infected scales with disease severity—it isn’t a binary, which you’d expect to see if the main correlate of immunity was a single threshold of GC response. [1 shows correlation with severity, 2 “great titer variability, 3 amazing paper showing different neutralization levels translates well to efficacy in vaccines, plus waning titer and loss of efficacy over time.] 

A different important departure from a binary condition: a substantial number of cases exhibit rapid waning of NAb levels over 6 months to a negligible concentration (and cases fall all over the spectrum for how fast they wane) [this great Lancet study]. Presumably this could be easily overcome with another vaccination, as asked.

Another datapoint against is that it doesn’t explain partial vaccine efficacy. How would a vaccine protect you from mortality but not from symptoms, unless it mattered where on a spectrum you landed? (Obviously there are ways, but they all strain credulity.) It also wouldn’t fit with the fact that vaccines can result in NAb titers that are universally high in a group, but still leave them getting occasionally infected [I’ve lost this source but it was obviously relevant for Delta strain, though many people produced no or little NAb for Delta which was an important distinction].

Last, even if we were totally dependent on a binary GC response, we could probably still modulate that with more introduction of antigen! My understanding of GC responses is that they get initialized in part through other cells like CD4+ T cells, the concentration of which also correlates with different doses of vaccine, presumably causally.

Replies from: JenniferRM, Connor_Flexman

↑ comment by JenniferRM · 2021-07-30T04:36:42.787Z · LW(p) · GW(p)

I just read this tonight and it is fantastic. You know more immunology than me by a lot! <3

I'll be editing my answer to leave the substance of what I wrote so posterity can see me being dumb but willing to make an educated guess, but defer to your answer at the top with an explanation :-)

↑ comment by Connor_Flexman · 2021-07-25T11:55:28.431Z · LW(p) · GW(p)

The relevant graphics on booster shots:

↑ comment by Lukas Finnveden (Lanrian) · 2021-07-22T20:14:26.182Z · LW(p) · GW(p)

Huh, I'm pretty surprised by this model. Why do you think it's correct?

Here's an image of some measure of people's antibody responses from page 9 of this paper, where the first set of points is people's response 5-6 wks after dose 1, and the second set of points is people's response 2-3 wks after dose 2.

It looks like people who get an antibody response to the first dose still get a much improved response from a second dose. And there's no sign of a bimodal responses to any of the doses. Is that consistent with your model?

Also, the way vaccines can protect from severe disease without protecting from infection seems to suggest that there's more than a binary question of response/not-response.

Replies from: JenniferRM

↑ comment by JenniferRM · 2021-07-23T02:07:18.093Z · LW(p) · GW(p)

You found a neat paper! Thank you!

"MY WHY" (for my admittedly simply model)... Is... Basically... Uh... a thing I experience often is that I have a sort of "models inside of models" expectation that seems rarely to be applied in practice even by so-called experts? 

For example during Feb/Mar 2020 there were people talking about how "X% of patients are asymptomatic" but they were making these assertions based on single snapshots of infection cohorts in whom the epidemic was actively moving. So some of the people (like on a cruise ship) might have been PRE-symptomatic (because maybe checking for symptoms 14 days later would turn something up) rather than A-symptomatic (as a general property of their entire disease course). People were using the terminology willy nilly, and no one was tracking any of it precisely... almost no one was thinking of it like they were getting momentary glimpses of a sort of state machine or advertising funnel or something, where each person might get a slightly different ride along a slightly different path AND ALSO might be at a different step along whatever ride they will end up having taken.

Another very common failing is that people think that a group mean implies a group median. If there are bimodal responses to something, and then a group summary is given... is the group's denominator over "all treated" or "all who were treated and had a followup confirming an adequate response"?

BOTH of these errors have a common cause of "assuming homogeneous efficacy and assuming competent followup at the clinical level" and in my experience neither of these assumptions are strongly justified. They constantly fail, and people are constantly acting surprised about it. 

Failures of followup are ALSO why we couldn't get people properly quarantined at the beginning of this disaster. 

Often patients DO NOT WANT to have "the system" FOLLOW them.

The study you linked to seems to have somewhat solved "the standard problems with the 'lost to followup' state" that is the bane of so many time course studies. The design certainly seems to take the followup process very much into account (and I couldn't find drop out rates from skimming or ^f and so maybe literally no one dropped out):

Something I'd like to call attention to here... in the paper you link the Extended Track had Bleed1 data from the extended cohort group, where they got (in some sense) to see how many people seroconverted from just one dose by week 5 or so...

Within the extended interval cohort, antibodies were detectable in 91% (62/68) at the first timepoint, at 5-6 weeks after the first vaccine, but this rose to 100% 2-3 weeks after the vaccine boost.

Recall that up in the abstract the paper summarizes the key result thusly:

Results: In donors without evidence of previous infection the peak antibody response was 3.5-fold higher in donors who had undergone delayed interval vaccination.

Suppose, hypothetically, that instead of 91% of people having "a seroconverting response" on the first shot it was only 28% of them?  

(This would be almost understandable. The youngest person in that study was 80 years old! The whole study is on a group whose immune systems should be assumed to be decrepit and fragile from the raw fact of great age.)

Then if the second exposure brought this up to 100% seroconversion "somehow", and the seroconverted "antibody levels" were gaussian (log normal?) among the seroconverted and 0 among the rest...

...Then that bimodal response could directly and cleanly justify claiming "antibody response was 3.5-fold higher" in some very fuzzy and general way (because 28% x 3.5 = 98%)

MY NORMAL EXPECTATION is for people to communicate in a fuzzy and general way :-(

The graph you included as a supporting claim was, I think, just the B panel from the totality of Figure 2 which is nice in many ways. Color coded! The horizontal axes are mostly aligned! Nice!

Note that in Panel A the two timepoints give basically the same levels of antibody response, with maybe some hint of a slow decline, but also overlap with Panel B's separated ranges. Some in Panel A went up?? Weird. Probably stuff goes down (and sometimes up?) over time, in general?

The data in Panel A therefore seems consistent to me that "eventually" there is some roughly normal and acceptable level of "vaccinated at all, in an essentially bimodal way" that two doses reaches faster than typical? 

This is what the two dose shot is designed to do in my mind: get ALMOST ALL of the patients (because of herd immunity benefits) to the state of "CLEANLY SEROCONVERTED" with the LEAST amount of measurement and need for followup (because followup is really hard).

Bleed2 of the standard group is "10 weeks post standard dose2".  There is no Bleed3 for either group out all the way at week 21.  That third data collection event would be "10 weeks after dose2 for the extended group" and thus sorta comparable to the standard group's Bleed1?

My hunch is that extended Bleed3 would show a decline from the extended Bleed2 measurement... 

...maybe this prediction is a crux?

I could also imagine those slow risers in the standard group would STILL be going up by week 21?

Basically, I suspect that antibody levels eventually go down EVENTUALLY (over months and years), but also have some "sensitivity to dynamics over a timecourse" (which is probably not showing up here, not because it didn't happen, but because it wasn't measured).

I don't know. My error bars are wide.

...

Also it would have been great to measure antibody levels for everyone on week 3, as Bleed0? More of the dynamics would be visible I think, and it would help characterize (and separate?) various members of the standard group in terms of seroconversion status before the second dose?

...

Basically, immunological science is more of an art than a science. It has a gazillion moving parts created under extreme adversity. Also, humans underestimate the difficulties of just doing the thing over and over.

People don't get that "seroconversion" IS A THING. Sometimes it just doesn't happen. That's often the most important practical fact. Many bad vaccine designs end up with a "vaccine" whose seroconversion rate is non-zero but so low as to be impractical.  This meta-analysis for measles (trying to find a relationship between age and seroconversion) shows numerous things that were tried in the clinic that had less than 50% rates for some kids. 

The Pfizer/Moderna/mRNA design is weirdly effective from my perspective here. To take a second dose of a weirdly effective thing... uh... I guess? Sure. If empirically that works for this disease maybe it could or would help somehow in a way that eventually could be made sense of.

So if I have to pick ONE THING to assume about vaccine status at the INDIVIDUAL level, it will be "seroconverted or not" and then after knowing the answer to that, I assume "the rest will be very very complicated".

I freely admit this is a simplified model. I just also think that any specific additional mechanistic second step to the modeling effort is likely to explode the combinatorial space of patient states to worry about, and yet also be a TINY epicycle, and the first of MANY epicycles.

The thing I'm asking for is: what's the best second epicycle to add? What is the mechanism? If someone is already seroconverted, what would you measure to detect "that their mechanistic biological state is not ALREADY in the configuration that you'd be hoping to cause to improve via the administration of a third dose"?

And in the meantime: DELTA MIGHT BE MUTATING AROUND THE CURRENT VACCINE DESIGN and so a new design aimed at the new epitopes just obviously seems like it would address the main uncertainty in a central way. 

RadVac for Delta is something I might pay for, and maybe something I might want to try to make on my own? Searching a bit: it looks like mixed third dose trials are already starting, so... <3!

A third dose that's "the same as the first two" doesn't interest me. The second one doing what it does seems to be empirically real, but I don't know why the hell it empirically gives the results it does.

Replies from: Lanrian

↑ comment by Lukas Finnveden (Lanrian) · 2021-07-23T20:13:09.935Z · LW(p) · GW(p)

On your "WHY", you seem to be presenting reasons why other people not believing your model shouldn't count as strong evidence against it. Which is all fair. But I'm still curious for positive evidence to believe your model in the first place. Maybe this would be obvious if I knew more biology, but as it is, I don't know why I should place higher credence in your model than any other model (e.g. the one at the bottom of this comment, if that counts).

...Then that bimodal response could directly and cleanly justify claiming "antibody response was 3.5-fold higher" in some very fuzzy and general way (because 28% x 3.5 = 98%)

As far as I can tell, "antibody response was 3.5-fold higher" just means that, on average, people in the extended dosing schedule had 3.5x more antibodies. I can't tell whether you interpret it in some other way, or if you think this is a misleading way to describe things, or if you're making some other point...?

The graph you included as a supporting claim was, I think, just the B panel from the totality of Figure 2 which is nice in many ways.

Yup!

The data in Panel A therefore seems consistent to me that "eventually" there is some roughly normal and acceptable level of "vaccinated at all, in an essentially bimodal way" that two doses reaches faster than typical? 

Ok now I'm confused.

• Do you think that all people on these graphs have reached a "normal and acceptable level of 'vaccinated at all, in an essentially bimodal way' "?
• If so, do you not think that there's any important immunity difference between a single-vaccinated person around 1-10 on the graph, or a doubly-vaccinated person around 1000-10000?
• Or if you think that only some of the people on this graph are immune, where do you think the line between immune and not-immune should be drawn on these graphs? (The distribution seems to be fairly continuous everywhere, to me, so it seems arbitrary to draw the line anywhere.)
• Or if you think the important immunity difference isn't captured by antibody-levels, what is it about?
• And re "that two doses reaches faster than typical"; are you implying that the single-dosed people's antibody response would've kept increasing beyond the 5-6 week mark and eventually gotten as high as the doubly-vaccinated people? That seems unlikely to me. (Other than maybe the few people where their antibodies did increase, but I'm happy to ignore them until I understand the most normal response curve better.)

My hunch is that extended Bleed3 would show a decline from the extended Bleed2 measurement... 

Agreed.

The thing I'm asking for is: what's the best second epicycle to add? What is the mechanism? If someone is already seroconverted, what would you measure to detect "that their mechanistic biological state is not ALREADY in the configuration that you'd be hoping to cause to improve via the administration of a third dose"?

Here's one suggestion:

1. The more antibodies you have, the less probability of getting sick, the less probability of getting severe disease, etc.

2. More vaccines increases the number of antibodies you have.

3. Therefore you want to have more vaccines.

I would've thought (1) to be fairly uncontroversial? And the linked study seems to provide good evidence for (2) when going from 1 to 2 doses, increasing antibodies by roughly a factor of 100. And of course adding more vaccines will eventually stop adding more antibodies. But right now I don't have any reason to believe in a big difference between going from 1->2 vaccines vs going from 2->3 vaccines (other than 2 vaccines being the general standard). So I wouldn't be surprised if taking a 3rd vaccine could increase your antibodies by another order of magnitude.

Maybe you think this doesn't provide enough of a "mechanism"? Biology being complicated, I'm very happy to take empirical data for what it is, and make extrapolations even if I don't know what the mechanism is. Personally, I also don't feel like I have any more mechanism for "vaccine have a fixed probability of causing antibodies if you don't already have them, otherwise they don't do much" than "vaccine typically increases antibodies by a lot regardless of whether you have them or not". So when the evidence clearly indicates the latter, I will definitely believe it.

And yeah, also, if someone has the option, I agree that it seems probably better to get a different vaccine than the same vaccine again!

Replies from: JenniferRM

↑ comment by JenniferRM · 2021-07-30T04:33:07.472Z · LW(p) · GW(p)

I sort of tapped out because "very long posts with an explosion of quotes" is a smell for me, but I wanted to continue because other indicators suggest "teaching and/or learning in good faith" <3

Finally posting now because of a big update from elsewhere...

On your "WHY", you seem to be presenting reasons why other people not believing your model shouldn't count as strong evidence against it. Which is all fair. But I'm still curious for positive evidence to believe your model in the first place.

For me, evidence happens at the point of measurement. Then often measurements are summarized in language by people who don't think clearly, or worry about standard misinterpretations of simple measurements... so careful reading is sometimes required just to acquire evidence able to distinguish between models.

So for me, the default is to need to think about mechanistic timecourse evidence through the screen of "how it was confusingly explained to me" by people who often aren't worried about mechanistic timecourse dynamics. 

I kinda don't care if people don't believe my model, I just want my models to get better over time... and I'm happy to explain them to people, and I like teaching... but if people don't believe me, then it is their tragedy that they believe false things, not my tragedy. (Conversely, people teaching me things is awesome!)

But to make my models better I don't just import other people's posterior believes about how a mechanistic system works, but rather see if my own model can "round trip" through my best guess of the raw data that they observed in a specific situation.  If people have bad reasoning, then their posteriors are even less safe to import than otherwise...

FWIW, just tonight I got around to reading this cousin comment by Connor [LW(p) · GW(p)] and it swiftly tipped me over almost entirely. Three doses... might work? Sure.

I already thought there were empirical reasons to think it, so for me I think the key words in Connor's post started somewhere around:

And not only does it increase count, secondary responses vastly increase antibody affinity and produce different antibody types, e.g. the primary response is more IgM whereas secondary response produces more IgG and IgA (the latter aiding especially in mucosal immunity). [Citations for this can be found on pgs 413-414 of the Janeway immunobiology book, and I can maybe link pictures.]

The filter I have I think, is that I want to hear about mechanisms when it comes to biological theories. 

I'm not saying button mashing doesn't work. That plus "copy the winner" is how most actual technical innovation occurs and scales in practice most of the time. Its fine <3

But... a HUGE filter that avoids adding broken bits to my general reasoning capacities is whether someone can offer keywords that connects their proposed mechanism to ALL THE OTHER MECHANISMS in physics and chemistry and evolution and all of it.

Gimme a word like "IgA" and I can find my way to new and helpful parts of the truth mine! I can round trip it through general science, and so on.

I have paragraphs and paragraphs of text from my first attempt at a response, trying to explain "I don't know and neither do you (but politely and at length)".

They are deleted from this response. Maybe "two people debugging epistemics in the face of ignorance" is useful somehow for something, but I'm not attached to it. I could PM it maybe if you care?

Practical upshot: empirically more doses has worked, and now I have heard some "new magic mechanism words" from Connor, who seems to me to clearly knows his shit backwards and forwards and also seems to be in tentative favor of a third dose :-)

Maybe interesting: my main argument AGAINST a third dose is part of why I thought it might be smart to give single doses as fast as possible several months ago. Now that like... "mechanisms are mechanically different (giving more than just lots of IgM)" I feel like I learned enough to even notice errors in past thinking? 

But also... weirdly(?) this same body of empirical results says that the second reaction works BETTER after ... <missing mechanism that somehow is time dependent> has had 12 weeks to <do something> instead of just 3 weeks?

F-ing immunology, man. Its crazy.

According to one expert, the immune system essentially makes bets on how often it will face a given virus and how the virus will mutate in the future:

https://science.sciencemag.org/content/372/6549/1392

By that logic, being challenged more often means that the immune system should have a stronger and longer-lasting response:

The immune system treats any new exposure—be it infection or vaccination—with a cost-benefit threat analysis for the magnitude of immunological memory to generate and maintain. There are resource-commitment decisions: more cells and more protein throughout the body, potentially for decades. Although all of the calculus involved in these immunological cost-benefit analyses is not understood, a long-standing rule of thumb is that repeated exposures are recognized as an increased threat. Hence the success of vaccine regimens split into two or three immunizations.

The response becomes even stronger when challenging the immune system with different versions of the virus, in particular a vaccine and the virus itself (same link). 

Heightened response to repeated exposure is clearly at play in hybrid immunity, but it is not so simple, because the magnitude of the response to the second exposure (vaccination after infection) was much larger than after the second dose of vaccine in uninfected individuals. [...] Overall, hybrid immunity to SARS-CoV-2 appears to be impressively potent.

For SARS-CoV-2 this leads to a 25-100x stronger antibody response. It also comes with enhanced neutralizing breadth, and therefore likely some protection against future variants. 

Based on this, the article above recommends combining different vaccine modalities such as mRNA (Pfizer, Moderna) and vector (AZ) (see also here). 

Lastly, your question may be hard to answer without data, if we extrapolate from a similar question where the answer seems hard to predict in advance:

Additionally, the response to the second vaccine dose was minimal for previously infected persons, indicating an immunity plateau that is not simple to predict.

Depending on when you got your second mRNA dose, the Israeli data suggests there is significant vaccine decline in vaccine efficacy after 3 months (see  https://www.gov.il/BlobFolder/reports/vaccine-efficacy-safety-follow-up-committee/he/files_publications_corona_two-dose-vaccination-data.pdf ) This does not of course indicate whether a third dose would restore efficacy (presumably by reigniting the immune response in some way) - I suspect it would, however. No need for prize money if relevant - just resharing a link that was already in Forbes https://www.forbes.com/sites/roberthart/2021/07/23/pfizer-shot-just-39-effective-against-delta-infection-but-largely-prevents-severe-illness-israel-study-suggests/?sh=ca1146c584f1

It might be worth posting this question to Metaculus to see what the crowd-consensus view is there. I'd post a question myself there, but I'm not sure I have enough expertise to define the question and resolution properly (but it seems like you or others might).

My guess is that you have generated your own hypothesis (namely, more than two doses will work better than 2 will). I almost always leave medical hypothesis generation to the medical experts with the result that my job is to choose among the hypotheses and to tailor my choice to my situation. Of course I consider hypotheses by contrarians and alternative health-care providers, which number in the high dozens or hundreds of hypotheses for Covid.

(Experts have proposed a 3rd shot, but that would be a shot of a new vaccine, not one currently being given to people.)