Statistical tests for a pre-post study

#1
Hello everyone. I am doing the analysis of a pre-post intervention study. The participants were divided into 4 groups and we want to demonstrate the differences in the three measurement moments....1. pre-intervention, 2. post-intervention, and 3. 1-year post-intervention control. Initially, we thought of doing only a Cochrane Q-test and others tell me that a mixed model is better. The response variable is dichotomous, presence or absence of a parasite. Now I am really confused as to what would be the best way to analyze my data. Thank you in advance for your help.
 

Karabiner

TS Contributor
#2
Do you want to analyse the total group between the 3 time points, or do you want to
compare the 4 groups over time? If yes, what are these groups and how were subjects
allocated to them?

And how large is your sample size?

With kind regards

Karabiner
 
#3
We want to compare the differences between the 4 groups throughout the measurements since some received one type of treatment, others another type, and one was the control. The sample is 807, 200 per arm approximately, although in the 2nd and 3rd measurements some participants were lost.

Thank you very much for your interest and help.

Ivana
 

Karabiner

TS Contributor
#7
The sample is 807, 200 per arm approximately,
I cannot see any reason for performing statistcal tests of inference with a total sample size of 3.2 million.
The standard error of a proportion would be something in the region of 0.0005 (less than 1/1000) for each
group. Or, is the parasite extremely rare (or extremely frequent, respectively), so that regardsless of
the enormous sample sizes, one outcome group consistes of only some dozen cases or so?

By the way, did you identify each subject over time, so that you can say something like "our subject No.
145,790 initially had a parasite, then at t1 this subject had no parasite, and at t2 he still had none"?

With kind regards

Karabiner
 
#8
Sorry...I misspelled the numbers, it is actually 807 participants who were divided approximately 200 something per branch.... and we do have identified individuals between the 3 measurements so we know who was infected before and who was infected after the intervention.

Thank you for your comments.
 

Karabiner

TS Contributor
#9
Sorry...I misspelled the numbers, it is actually 807 participants
Oops, I did mis-read that.

If you want to compare the groups over time, then Cochran's Q is useless, since it
compares between time points for one group only. Moreover, if this is an observational
study without randomization, you perhaps want to include additional variables to
adjust for group differences.

You could indeed consider a mixed model. Or you could perform logistic regression analyses,
for example predicting outcome at t2 by group and outcome at baseline, or predicting
outcome at t3 by group and outcome at t2.

With kind regards

Karabiner
 
#10
Tha
Oops, I did mis-read that.

If you want to compare the groups over time, then Cochran's Q is useless, since it
compares between time points for one group only. Moreover, if this is an observational
study without randomization, you perhaps want to include additional variables to
adjust for group differences.

You could indeed consider a mixed model. Or you could perform logistic regression analyses,
for example predicting outcome at t2 by group and outcome at baseline, or predicting
outcome at t3 by group and outcome at t2.

With kind regards

Karabiner
Thanks a lot for your comments, it was really helpfull.......Then we will improve our analysis taking account all the details.