Additionally, if you have ever done research you know that you have to disclude a handful of cases from the data for miscellaneous reasons
I have for years and years with human subjects, and it has never been considered acceptable anywhere I have published to not include itemized breakdowns of any non inclusions including specific reasons (and genders for that matter) The FDA is just more lax than journals, apparently. Take that as you may..
Like I said though, I ignored that issue for my estimate. I'm merely keeping track of conservative estimates made, not justifying any sort of math adjustment.
A) Repetition of studies
I have never seen a repeated clinical trial for any vaccine. Maybe it has happened here and there, but the vast majority have one clinical trial.
Also, as mentioned, I already generously multiplied the sensitivity by 10,
already accounting for multiple repeated studies in case they exist (which again I haven't actually seen). Specifically in anticipation of stuff like this.
Large scale studies on vaccines are done not once, but multiple times
Okay, can you find me 5 different clinical trials on the MMR vaccine Priorix? (Keeping in mind that even if you do, I've already accounted for that hypothetical, but I'm still interested).
It seems you classified risk on a linear scale rather than by risk of spread - which is all but eliminated at 80% immunity.
No, I was using the terms risk and benefit from the perspective of vaccination. Lower risk of spread of disease is classified as a benefit. Please see the graph in section #5 of the OP. "Risk" here is risk of vaccine damage. it's simply the other way around from the words you're using. Notice that benefit slopes, and that it has a logarithmic shape which is indeed near its maximum benefit ("all but eliminated" depends on other nations not just immunization rate here) in my graph at around 80% immunity (please be sensitive to it being a hand drawn MS paint graph, I'm no wizard)
Immunological responses to a vaccine would be STRONGER in babies with healthy immune systems, plus the data itself would not be confounded by unrelated illnesses. Screening out genetic disorders and babies who aren't developing normally/are sick often is a great way to standardize results. Otherwise we'd be complaining about all of the cases they had to exclude to get any meaningful data, rather than just displaying a spread of problems all caused by external factors.
http://en.wikipedia.org/wiki/External_validityExplaining how stuff would work really well in a made up world is not the point of a clinical trial.
In reality, drugs can have interactions that are
greater than the sum of their parts and thus require clinical trials BOTH in isolation like you explain AND in any common cocktails (which for vaccines these are almost always given to kids in bigger bunches). As just one example. In the cocktail version, what you would do is then count up complications, and subtract off the expected complications from baserate + each of those other individual cocktails = the cocktail overall synergy effect, which you can both compare to other vaccines' and as an adjustment to overall risk compared to not vaccinating.
Again, keep in mind that I gave them a mulligan on this and didn't actually take it into account, am simply noting.
@Neonivek:
That is why you get a lot of babies and have a control group.
A lot of babies: Yes, they run thousands, but not a large enough "a lot" to get to the numbers they need to compare to the small disease benefits at current vaccination levels for some things (like measles)
Control group: Most clinical trials ALSO do the completely irresponsible thing mentioned earlier where they compare
to another vaccine and not anything else, not to other ACTUAL control babies who for instance haven't been vaccinated with anything in the last month (never vaccinated at all would be scientifically preferable, but I'm being realistic). Since nearly every clinical trial does this, most vaccines are only testing their brand-specific dangers, not the actual danger of the whole vaccine that a child actually gets.
A reasonable actual experiment here would have 4 conditions, assuming ~1 month is a standard complications window, which it is:
Control 1) Babies with no vaccine in the last month
Control 2) Babies with whatever
other vaccines in the last month an MMR receiving baby would typically have on average.
Experiment 1) Control 1 + MMR, i.e. Babies with only MMR in the last month
Experiment 2) Control 2 + MMR in the last month, i.e. full average cocktail as seen in normal usage.
vs reality most of the time:
"Control" 1) Another vaccine only in the last 30 days for the same disease
Experiment 1) This vaccine only in the last 30 days.
This is like using McDonald's as your control group for a study on the health effects of Burger King.@Putnam
A clinical trial of 2000 showing no negative reactions means that the overall risk is less than 1/2000 anyway...
It is lower than that, due to uncertanties. As I mentioned previously in my example of a 5,000 subject trial, the most they can guarantee at alpha = 0.05 would be 1/1,100 rate or lower, for instance (so 5x less rate guarantee than the subject number). It changes with population size. You have to use a binomial probability function. (A normal curve is sometimes used to approximate for large numbers)
Edit: to establish a rate of 1/10,000 or lower guaranteed at alpha 0.05, you would need a total population of
45,000 research subjects. A typical clinical trial has about 1,000-2,000, large ones go up to 5,000 or a little more.
To establish a rate of 1/60,000,000 at alpha 0.05, you'd need
~300,000,000 subjects
Edit #2: I use binomials because I'm still mainly focusing on death, and you're dead or not. You can also do e.g. differences in severity of something, but that typically requires even more subjects to establish a given rate.