Paired samples - Odds ratio and confidence interval

[sl8]

Hello everyone.
I have to analyze the effect of a certain treatment among the same sample of people, before and after that certain treatment.
Except for applying the McNemar Test, I'm also interested to apply an odds ratio.
Is there a specific odds ratio and/or confidence interval for paired samples?

What would you recommend?

(I don't mean matched samples)

 

[hlsmith]

Did everyone get the treatment?

 

[fed2]

the odds ratio associated with mcnemars test is described on page 183 of lachin
https://media.oiipdf.com/pdf/861e62eb-0556-480c-ad91-5cc78a3e98fd.pdf

turns out to be the ratio of 'f' to 'g', i believe this applies to repeated measures in time, same as matched samples. you will want to check me out on that.

good luck on your amazing analysis

 

[sl8]

Did everyone get the treatment?

Yes. All participants were treated.

 

[hlsmith]

So what is the outcome, in that they can get it twice? Does there have to be sufficient time between measurements to ensure it resolves and on the second measurement how do you ensure you didn't miss the outcome since it can resolve?

Sounds like a state space outcome, which can require controlling for time.

Just jumping into odds could misrepresent true effects if the above factors exist! That and a change in percent would be more informative.

 

[gw1]

Hey, the first link below might be of interest for you generally, for pretest posttest modelling:

https://journals.physiology.org/doi/full/10.1152/advan.00018.2015

 

[hlsmith]

I would also propose that without knowledge of the outcome, we don't know if the probability of the outcome changes. I'll use an accessible example. Say I had a sample of people. I collect their prior COVID exposure/infection status Next, I will have them all get vaccinated. Now I will see how many get infected. Ignoring exposure differences between them, time is super important in the example, such as time since initial exposure, time to vaccine, and time to second exposure. Also, if the person was previously exposed, that may impact how the post-vaccine infection presents. So without knowledge of the exposure and outcome, we can make general recommendations but whether they can identify the estimate under investigation is highly questionable.

 

[fed2]

So what is the outcome, in that they can get it twice? Does there have to be sufficient time between measurements to ensure it resolves and on the second measurement how do you ensure you didn't miss the outcome since it can resolve?

yes i think it would have to be an actual 'cross-over' study for mcnemars test to apply, ie each subject has untreated and treated period, each with the potential for a disease/outcome. Thats pretty rare in epidemiology i think, but might be the case for some short acting compounds, like an eye-drop or something.

sounds like that is not the case according to original post, hard tot tell.