Need help choosing the right analysis [unsolved]

#1
Hi!
I am using three surveys, let's call them A, B, and C. From one survey to the next, there is a 5-year interval. From these surveys, I calculate the weight change and I then check if they get a certain diagnosis afterward within 10 years. I want to see if weight increase or decrease is associated with getting the disease.

I divided the data into two study samples:
Group AB: Those who participated in both A and B and did not have the diagnosis from before.
Group: BC: Those who participated in both B and C and did not have the diagnosis from before.

Some individuals participated in all three surveys and therefore are in both groups, so those will be counted in the dataset two times. But even though they are counted two times, the weight change observations will not be the same. Maybe the person went up in weight between A and B and down in weight between B and C.

My question is if I have dependent data and therefore cannot conduct a cox regression?
And if so, what other regression analysis can I use instead?

Thanks a lot in advance.

I have little statistical knowledge and all help is welcome.
 

Miner

TS Contributor
#2
If a subject has an underlying physical/psychological condition (e.g., cancer, eating disorder, genetic pre-disposition etc.), they would be dependent. It would be very difficult to rule these out without making major assumptions.
 
#3
If a subject has an underlying physical/psychological condition (e.g., cancer, eating disorder, genetic pre-disposition etc.), they would be dependent. It would be very difficult to rule these out without making major assumptions.
Thanks for the reply Miner. So if I understand you right, the same individual can not be in the same dataset two times even though there are different observations for them.

An alternative that I already tried is to first do a cox regression for the AB group and then do a cox regression for the BC group. In this way, I don't do an analysis where any individuals are repeated two times. However, the problem here is that I get so few cases. That is why I would like to combine the two study groups. Is there any other regression analysis I can conduct that takes into account that some observations are dependent?

Thanks again.