Baseline as confounding factor in logistic regression

#1
Hello guys

I want to know if swallowing complaints (continuous) can predict dynapenia (binary) in an follow-up cohort.

Classically, I should exclude participants with dynapenia at baseline and estimate the coefficients for follow-up, correct?

The problem is that dynapenia is a modifiable condition, which means that the presence of dynapenia at baseline will not necessarily result in dynapenia at follow-up.

Given this, would it be statistically acceptable to consider dynapenia at follow-up as dependent variable, and dynapenia at baseline as confounding factor (covariate)?

Thanks in advance.
 

Karabiner

TS Contributor
#2
It primarily depends on what you want to find out, not so much on statistical considerations.
Wouldn't any physician include current dynapenia into his prediction of future dynapenia?
Or, are you specifically interested in prediction of future dynapenia, in people currently
without it?

If you want to make a reliable prediction, then maybe you could use three predictors,
current swallowing complaints, current dynapenia, and their interaction; the predictive
value of swallowing complaints will possibly depend on whether there is already
dynapenia, or not.

Just my 2pence

Karabiner
 
Last edited:

hlsmith

Less is more. Stay pure. Stay poor.
#3
Also, are you measuring everyone for dynapenia at the same time and time interval? This seems more of a state space model. People are moving through states, where logistic reg is just gonna look at cross-sectional association which can misrepresent information.
 
#4
It primarily depends on what you want to find out, not so much on statistical considerations.
Wouldn't any physician include current dynapenia into his prediction of future dynapenia?
Or, are you specifically interested in prediction of future dynapenia, in people currently
without it?

If you want to make a reliable prediction, then maybe you could use three predictors,
current swallowing complaints, current dynapenia, and their interaction; the predictive
value of swallowing complaints will possibly depend on whether there is already
dynapenia, or not.

Just my 2pence

Karabiner
I liked your approach. I hadn't thought of that and will try to run the model according to your suggestion.

Dynapenia at baseline influences the prognosis of functionality in older adults. For example, an older person with low levels of muscle strength will probably experience physical difficulties in the future.

I say "probably" because physical performance is highly susceptible to change in response to exercise, mobility in living spaces, caloric intake, etc.

With that, I want to know if swallowing dysfunction at baseline can predict dynapenia in the future (after 8 years), correcting the estimates due to the fact that some people have dynapenia at baseline.

So I proposed to use dynapenia at follow-up as the dependent variable, and dynapenia at baseline as the covariate. Does it make sense?
 
#5
Also, are you measuring everyone for dynapenia at the same time and time interval? This seems more of a state space model. People are moving through states, where logistic reg is just gonna look at cross-sectional association which can misrepresent information.
Good point. That is another doubt I had. The time is fixed, which means I don't know the exact time when the participants presented dynapenia, I just know that there was a change after 8 years. In this case, the estimates really look like a cross-sectional study. Is this a limitation or a statistical/theoretical inconsistency?