comparing change from baseline (two treatment groups, binomial data)

[gopher]

I am comparing the effectiveness of a vaccine in reducing the prevalence of the targeted protein. I have baseline data (before the vaccine was even available), and now have data from vaccinated and non-vaccinated individuals.

I have used chi-square/odds-ratios to show that the prevalence of the target protein is significantly reduced in both the vaccinated and non-vaccinated individuals, compared with baseline. I've also done a chi-square between the vaccinated and non-vaccinated individuals to show no significant change.

My question is - The prevalence change for the vaccinated group is -20% (from 23% down to 3%), and the prevalence change for the non-vaccinated group is -18% (from 23% to 5%). What statistical method can I use to test if this change is not/significantly different for the two groups?

Thanks,
G

 

[Karabiner]

Are the baseline data from the very same persons which later
constituted the vaccinated and the non-vaccinated group?
Was vaccination versus non-vaccination determined at random?
If yes, then the chi-square between the vaccinated and
non-vaccinated individuals normally should be sufficient
to state that no effect of vaccination could be demonstrated,
compared to non-vaccination.

By the way, in order to show that a prevalence change
from 23% to 3% is significantly different from 23% to 5%
the perdentages would have to be based on a very, very
large sample.

With kind regards

K.

 

[gopher]

@ Karabiner, good question! No they are not same persons, however they both represent the population (compared against census data. In regards to the randomness, as this is an epidemiology study we have no say over who was vaccinated/not-vaccinated. Yes, I would need a very large sample - I actually want to prove that there is no difference in the change, which helps to establish herd immunity.

I've been advised from elsewhere to "use a Poisson or Logistic Regression, comparing the outcome data (5/n1 vs 3/n2) adjusted for the baseline data (23/n1 vs 23/n2)". However, I have to admit I have no idea what this entails. I know what the Poisson/Logistic Regressions are, but am most confused by the data input :-S.

 

[Karabiner]

No they are not same persons, however they both represent the population (compared against census data.

I normally would be a bit surprised if people with and without vaccination
did not differ with regard to other - potentially relevant - characteristics.
For example, I would not expect that baseline rates were the same. But
you probably have of much more information regarding the study populations etc.

As far as I can see you can do not much more than to state that both groups
did not significantly differ with regard to presence of the target protein.

With kind regards

K.