What claims can legitimately be made about the Covid vaccines?
February 21, 2021 9:41 AM   Subscribe

Vaccines prevent severe illness, yay! But...(how to make sense of the numbers?)

The Moderna vaccine has an approximately 94% efficacy, meaning that, for every 100 people who received a placebo and got Covid, 6 people who received the vaccine got Covid.

Fine. We like this very much.

However, now, because other vaccines (Johnson & Johnson? AstraZeneca?) didn't show that level of efficacy in clinical trials, there have been many articles stating, "ok, but [name of vaccine] DID prevent SEVERE Covid! including hospitalization, and there were no deaths! And that's great!" (meaning that the people who received vaccines and got sick anyway got MILD Covid).

BUT...

when you look at the actual data from the trials, you find that, not only did very few people who received the vaccine get Covid, you also find that very few people who received the placebo got Covid, and NOBODY WHO RECEIVED A PLACEBO DIED EITHER.

So they are comparing Zero deaths with a vaccine to Zero deaths WITHOUT a vaccine! Does this make sense? Is it an ethical claim??

Now, we're talking about trials, like for the Moderna vaccine, with something like 20,000 people who got vaccine and 20,000 who got placebos! And maybe (I can't remember the actual numbers) 150 people who received placebos got Covid? out of 20,000? (similar for Pfizer, of course).

I did some calculations yesterday and I found that the infection rate for placebo in the original Moderna trial was something like .006! (that's not .006% -- that's .006)

Whereas, in New York City at this moment, the infection rate is something like 4-8%!

Now, granted, the current 4-8% is based on positive tests of people who elected to go get a Covid test -- and some of them, you have to assume, went in because of symptoms (although a lot of people now get tested for work, or school, or because they think a negative test means they can go to a party). So they're going to have a higher positivity rate than 20,000 trial volunteers who, presumably, may be more health conscious and safety conscious than the general population (?) BUT -- why is it THAT different?

Isn't it sort of insane, that so few people in a pandemic actually got the disease? Are the populations used in the trials really representative? Or -- the trials were done a long time ago, when, in fact, not that many people did get Covid? Is it even fair to extrapolate from the trial population, when, apparently, either the disease itself or the conditions at the time were so different from what we're experiencing now?

Let's assume that the difference between placebo and vaccine is, in fact, legit and statistically significant -- I really do think that's true -- but, given the numbers of people who got sick in total, is it valid to say that "this vaccine prevents deaths!" when nobody who got a placebo died either? I understand that the trials we have are the trials we have, and "we're going to know a lot more in the coming months" < repeat repeat repeat until exhausted> but my question has to do with Current Claims of pharmaceutical companies and how it might affect our understanding and behavior (I know, wear a mask, social distance, etc.) -- but how about the behavior of the pharmaceutical companies?
posted by DMelanogaster to Health & Fitness (18 answers total) 2 users marked this as a favorite
 
You are neglecting the element of time. The trials measured the number of people infected over a short period of time, two or three months.
posted by JackFlash at 10:00 AM on February 21 [2 favorites]


The sample population is likely very not representative, and that is in part because there's a fine line between "pharmaceutical trial" and "medical experimentation" but also a fine line between "the outcome we want" and "the outcome we don't want". Recruiting people who are in poor health or have a high-risk medical condition is not ethically great or good for your data, which means the candidate pool is going to be thin, healthy, not on any ongoing medication (except maybe contraception, because they don't want pregnant people in the trial), not pregnant or recently pregnant or nursing, and OVERWHELMINGLY white.

So, not at all the people who are dying in the overall population. And I don't think these trial populations are at all controlled for environmental risk, so you can't know how many participants were frontline healthcare workers or living in a dorm/crowded living situation vs people who only go out for groceries or not at all.

Yes, there is very funky extrapolation in (pretty much all) pharma math, but it is on purpose and by design and blessed by the governing bodies because killing people/letting people die is bad, so the aim is sort of to kill (or prevent the deaths of) theoretical people based on assumptions that can be made from very controlled testing on real people. Some of the math in this case will be modeled on what we know about coronaviruses in general and what we know about the progression and behavior of COVID-19 in less healthy patients. That is the least terrible way to do this, from a human cost standpoint, but it means the data is weird if you try to take it at face value.

Epidemiologists seem to generally be on board with these vaccines, and they are who I'm looking at for reassurance here. I realize it's mostly a "something's better than nothing" situation where we may end up with regrets of varying kinds somewhere down the road, but presumably they are trained to consider that as well.
posted by Lyn Never at 10:56 AM on February 21 [10 favorites]


when you look at the actual data from the trials, you find that, not only did very few people who received the vaccine get Covid, you also find that very few people who received the placebo got Covid, and NOBODY WHO RECEIVED A PLACEBO DIED EITHER.

I don't think that's correct. In the Johnson & Johnson trial there were 5 covid-related deaths in the placebo group and none in the vaccine group. Technically a 100% reduction, but I agree that's overstating the case. We need more data.

I did some calculations yesterday and I found that the infection rate for placebo in the original Moderna trial was something like .006! (that's not .006% -- that's .006)

According to this report it was 185 cases out of 15,210. That's 1.2%, did you remember to divide the participant group in half (two equal cohorts)? Also, these were symptomatic cases.
posted by justkevin at 11:01 AM on February 21


I didn't divide in half because I thought it said 20,000 in each cohort, and, well I don't remember the number of sick/dead anymore. Also maybe you're looking at Israel's more recent study? I don't even know anymore. But 1.2% sounds like a very reasonable number, unlike mine.
posted by DMelanogaster at 11:11 AM on February 21


This recent letter in The Lancet might also be helpful: What does 95% COVID-19 vaccine efficacy really mean?
posted by wavelette at 11:12 AM on February 21 [6 favorites]


Now, granted, the current 4-8% is based on positive tests of people who elected to go get a Covid test -- and some of them, you have to assume, went in because of symptoms (although a lot of people now get tested for work, or school, or because they think a negative test means they can go to a party). So they're going to have a higher positivity rate than 20,000 trial volunteers who, presumably, may be more health conscious and safety conscious than the general population (?) BUT -- why is it THAT different?

You are way underrating the bias of the covid test sample. The vast majority of people do not just get random covid tests. They are inconvenient, can be hard to schedule, and somewhat unpleasant. Most people tested either are exhibiting symptoms or else know that they have had close exposure. These two factors make the likelihood of being positive thousand of times more likely than just selecting a random person off the street. So, yes, it's that different a sample population.
posted by JackFlash at 11:15 AM on February 21 [11 favorites]


I don't think these trial populations are at all controlled for environmental risk, so you can't know how many participants were frontline healthcare workers or living in a dorm/crowded living situation vs people who only go out for groceries or not at all.

Anecdata here but I heard from someone I believe in one of the MeFi COVID threads that for the trials they were actually looking for people who were in fact getting out and about and might have some environmental risk factors. This MeFite was basically rejected from a trial because they were more the "Only go out for groceries" type.

In Vermont an awful lot of the testing is being done at or around schools (college students who are at in-person classes get tested quite regularly) as a condition of people being on campus, but other than that, yes, most people get tested if there is a reason to, not just randomly.
posted by jessamyn at 11:24 AM on February 21 [5 favorites]


I"m aware of the Lancet article and that issue, which I don't think is what I'm asking about. I know what 95% efficacy means. I'm asking a different question.
posted by DMelanogaster at 11:27 AM on February 21


Sorry, justkevin, of course not the Israel study. I screwed up. Thank you for this information, it's very helpful. Even saying that subjects were "at high risk" for Covid is VERY interesting.
posted by DMelanogaster at 11:31 AM on February 21


The UK did a very large randomized study of antibodies in June and found a 6% prevalence in the population. That means about 6% of people in the UK had ever had COVID by the end of June. So the percentage of people to catch COVID during the stage 3 trials (which last a single-digit number of weeks) is necessarily going to be way lower than that.

As to deaths: if you don't get severe COVID, you're not going to die of it either. So if you can entirely prevent severe COVID then you can be confident you're preventing deaths too.
posted by BungaDunga at 11:33 AM on February 21 [1 favorite]


Pfizer gives a breakdown of the participant diversity in their trials - they were approx. 60% white. Moderna were approx. 80% white, apparently.

You are way underrating the bias of the covid test sample

Just for illustration, I checked for Germany (as I could find the numbers fairly easily). Approx 1.8% of the population tested positive over the last 3 months. (Which is about the same period as the Moderna vaccine trials seemed to be.) The test positivity rate has been between 6% and 12% over that period.
posted by scorbet at 11:33 AM on February 21 [1 favorite]


So they are comparing Zero deaths with a vaccine to Zero deaths WITHOUT a vaccine! Does this make sense? Is it an ethical claim??

Yes. The reason is statistics. Even though there have been 500,000 deaths, it is still a rare occurrence statistically. So it is extremely hard to measure death rates over a short period of time. Since day 1 of the pandemic, a full year, only 1.5 per thousand have died of covid in the U.S. So if you try to count deaths in a group of 20,000, selected not to be the most at risk, over just a couple of months, it would not be surprising to not have any deaths recorded. You might have to run the test experiment for one or two years to get statistically significant death data.

So what do you do instead? You count the number of symptomatic infections which is a much easier number to achieve in a couple of months. And since you know that a certain percentage of infections leads to death, you know that if you can prevent infections, you can prevent deaths, even if no deaths showed up in your sample period.

The same would apply to a vaccine like measles. Deaths from measles is rare and would be hard to measure in a small population of test subjects. But you can measure the number of symptomatic cases of measles and you know if you can prevent measles, you can prevent deaths. Since you know pretty accurately the number of measles case that result in death, you know pretty accurately how many deaths you can prevent if you can prevent measles infections. Even if you never counted a single measles death during your vaccine trial period, which is limited in both time and sample size.
posted by JackFlash at 11:39 AM on February 21 [17 favorites]


I am not an epidemiologist and I am not your epidemiologist. However, I do follow a number of respected epidemiologists on Twitter.

My understanding is that it is misleading to say that the J&J vaccine is less effective than the Moderna and Pfizer. But wait, isn't 86% efficacy less than 95% efficacy? No, it's not, because the results were not part of the same trial. In particular, the J&J trial included the South African variant, and the Moderna & Pfizer trials did not.

Bottom line, take whichever of the three vaccines you have access to. They are all highly effective.

(I know that wasn't exactly your question, but I did want to point out this misunderstanding of the J&J vaccine's relative efficacy.)
posted by Winnie the Proust at 11:41 AM on February 21 [5 favorites]


I found this article by Zeynep Tufekci (who's coverage of the pandemic has been generally great) helpful. I'm not sure it answers your specific question but it put in to context for me the difference between what trials measure and what's important from a public heath standpoint.
posted by tomp at 12:41 PM on February 21 [2 favorites]


I think the numbers you're starting from are off.

The Pfizer study of 43,661 participants demonstrated 95% efficacy for the vaccine. It found that .74% of the placebo group became infected, and .04% of the vaccine group became infected.

To then determine vaccine efficacy, you find the difference between the two groups: .7%, or in other words, infection risk was reduced by .7 percentage points. It doesn't seem like much until you put it into a vaccine efficacy formula:

Risk among unvaccinated group − (minus) Risk among vaccinated group
/ (divided by) Risk among unvaccinated group

so: .7/.74= .94594, or about 95%. This is the proportionate reduction in disease among the vaccinated group, or a 95% reduction from the number of cases you would expect if they have not been vaccinated. It doesn't say anything about deaths, as you pointed out- but that's not part of how efficacy is calculated. We can assume deaths are being prevented because cases are being prevented. I'm really not seeing how this is unethical.
posted by oneirodynia at 3:18 PM on February 21 [2 favorites]


oneirodynia makes an important point.

From the CDC: "Vaccine efficacy and vaccine effectiveness measure the proportionate reduction in cases among vaccinated persons."

Efficacy measures cases not deaths. And cases means symptomatic infections, not asymptomatic. The CDC has very specific criteria that they use to define a case for each disease. Here are the criteria for covid. A case requires specific observed symptoms and/or laboratory tests. If a person is infected but not symptomatic and not tested, they are not a case.

So efficacy is a measure of the reduction of symptomatic cases of infection. Reduction of deaths is an important side benefit of efficacy, but not the primary measurement.
posted by JackFlash at 4:48 PM on February 21


The sample population is likely very not representative, and that is in part because there's a fine line between "pharmaceutical trial" and "medical experimentation" but also a fine line between "the outcome we want" and "the outcome we don't want". Recruiting people who are in poor health or have a high-risk medical condition is not ethically great or good for your data, which means the candidate pool is going to be thin, healthy, not on any ongoing medication (except maybe contraception, because they don't want pregnant people in the trial), not pregnant or recently pregnant or nursing, and OVERWHELMINGLY white.


I'm sorry, what? They did publish all kinds of things about the sample population, you know, so you can actually look into this. Here's some info from the Moderna study:

The Phase 3 COVE study was designed in collaboration with the FDA and NIH to evaluate Americans at risk of severe COVID-19 disease and completed enrollment of 30,000 participants ages 18 and older in the U.S. on October 22, including those at high risk of the severe complications of COVID-19 disease. The COVE study includes more than 7,000 Americans over the age of 65. It also includes more than 5,000 Americans who are under the age of 65 but have high-risk chronic diseases that put them at increased risk of severe COVID-19, such as diabetes, severe obesity and cardiac disease. These medically high-risk groups represent 42% of the total participants in the Phase 3 COVE study. The study also included communities that have historically been under-represented in clinical research and have been disproportionately impacted by COVID-19. The study includes more than 11,000 participants from communities of color, representing 37% of the study population, which is similar to the diversity of the U.S. at large. This includes more than 6,000 participants who identify as Hispanic or LatinX, and more than 3,000 participants who identify as Black or African American.

Anyway, I think the question has been answered, but clearly not by the answer marked as "best" which has a number of demonstrably false claims in it.
posted by ch1x0r at 5:00 PM on February 21 [14 favorites]


I'm not sure from your question if you want to get a better handle on how good the vaccines are, or more interested in the statistical problem of what we can and can't say from the trials. If the former, you might look at the Israeli experience with a mostly Pfizer-vaccinated population. I have not studied it myself, but the real world results have been pretty much in line with the trial results. And they have been able to extend those results to conclude that the vaccine reduces the rate of transmitting the virus to others by 95%, a question not studied in the trials at all.

There is a possibility that the J&J vaccine will be recommended to go to a two dose model, at least in some places.
posted by SemiSalt at 5:58 AM on February 22


« Older Project management ad infinitum   |   I need historical fiction about... Newer »

You are not logged in, either login or create an account to post comments