Wikipedia:Reference desk/Archives/Science/2020 November 23

Welcome to the Wikipedia Science Reference Desk Archives
The page you are currently viewing is a transcluded archive page. While you can leave answers for any questions shown below, please ask new questions on one of the current reference desk pages.

November 23

When less is more in a vaccine

From the recently released preliminary report regarding the Oxford vaccine:

'Prof Andrew Pollard, the trial's lead investigator, [...] protection was 90% in an analysis of around 3,000 people on the trial who were given a half-sized first dose and a full-sized second dose. Prof Pollard said the finding was "intriguing" and would mean "we would have a lot more doses to distribute." '

'Two full doses of the Oxford vaccine gave 62% protection, a half dose followed by a full dose was 90% and overall the trial showed 70% protection.'

according to [| the BBC].

Any plausible explanation for this? Any other vaccine that works like that (lower dose, higher protection)? --Bumptump (talk) 16:56, 23 November 2020 (UTC)

Sure - most of the realities of the dynamics of biological and physical sciences are well-modeled as nonlinear systems. The defining characteristic - the literal definition - is that the system's response is not proportional to the input.

So let's flip the question around - why in the world would anybody expect response to be proportional to the amount of dose? Exactly which model of immune response would predict that? When does the dosage linearly relate to treatment efficacy - over a wide range of dosages? I think you will find the answer is "almost never." Nimur (talk) 17:09, 23 November 2020 (UTC)

That's not entirely fair to the OP, and is attributing to them a question they didn't ask. They never said or implied it should be linearly proportional over a wide range of dosages. They're asking about it being inversely proportional (linearly or non-linearly), over some range. If we flip their question around to "why would we expect response to generally increase with increased dosage over parts of the range", the answer is still "sometimes it doesn't" (and possible reasons for that are given by Mikenorton below). But the question isn't as unreasonable as you make it sound. --Floquenbeam (talk) 17:23, 23 November 2020 (UTC)

(EC)The BBC story goes on to give two possible explanations: 1. "The immune system rejects the vaccine, which is built around a common cold virus, if it is given too big an initial dose", 2. "The low then high dose may better mimic a coronavirus infection and lead to a better immune response". No comment on the plausibility of these two alternatives. Mikenorton (talk) 17:11, 23 November 2020 (UTC)

I'll note most vaccines contain adjuvants, not just antigen, which means more moving parts. Biology is complex, and fascinating! Cells and organisms are full of both positive and negative feedback loops. For some reading material: the adaptive immune system has both humoral immunity and cell-mediated immunity "arms", and we want to appropriately twiddle both to get an ideal vaccine response. --47.152.93.24 (talk) 03:09, 24 November 2020 (UTC)

Plausible explanation: There were just 3 COVID cases in the vaccine group. The probability of having 3 or less if the expectation value was not different from the other group is not all that small (I think the expectation value would be about 7 cases in that group, and then the probability would be about 8%). A fully fledged statistical test would likely increase this probability due to taking into account all sources of statistical fluctuations. Count Iblis (talk) 08:18, 24 November 2020 (UTC)

I assume by "vaccine group" you mean the group that received first a half dose and then a full dose. The size of that group would be about 3000; a reduction of 101 cases per 10000 to 3 per 3000 would indeed indicate about 90% protection if you do not take a chance element in account. The number 3 is statistically speaking really low; it might as well have turned out to be 4. Based on the data, I don't think you can conclude with any confidence to a protection by the half-then-full regimen that is substantially better than 70%. The best that you can say is that it is likely not to offer less protection than twice full.  --Lambiam 15:07, 24 November 2020 (UTC)

• See also Data_dredging#Drawing_conclusions_from_data. The BBC article is a bit thin on info, you can find more facts in this Guardian piece but be cautious. The quotes from researchers give the impression that they are not aware that making hypotheses after you have seen the data is dangerous; maybe the researchers are bad statisticians, or maybe the journalist quoted selectively without realizing the article cut out some important info.

Notice that even you can compute some sort of statistical threshold to see a "statistically significant" difference between one-shot-and-a-half vs. two-shots, it is still not kosher to deduce it without a follow-up study - some low-probability "pattern" is bound to happen, because there are so many things you could try to see (for instance "bald people are better-protected" or "people whose sum of zipcode digits is a multiple of 7 are better-protected"). (Obligatory XKCD) Tigraan^{Click here to contact me} 13:39, 25 November 2020 (UTC)

Having several experimental groups next to the control group means, for a sanely conducted test, that the researchers also meant to compare the results of the experimental groups. In a two-sided test (the kosher choice), the difference is significant at the 5%-level but not at the 1%-level. Concluding with high confidence (better than 5%-level significance) to a one-sided higher efficacy does require a follow-up.  --Lambiam 11:47, 26 November 2020 (UTC)

The recent announcement that these two groups received different dosages due to an unplanned error means that a comparison between the results of the experimental groups may not have been part of the original plan. This does not invalidate any of my conclusions, including that we cannot conclude that one regimen is substantially better than the other.  --Lambiam 16:30, 26 November 2020 (UTC)