Linear regression of one sample pre- /post- data

[zta3]

Hi group, 

I have a clinical dataset with about 80 patients. They had a baseline score (pre-intervention score) and a post intervention score. I used paired t test to compare the mean difference. 

Now I need to get parameter estimate (coefficient) of the post score  in a linear model and possibly add multiple covariates in later.  To my knowledge,  I can't really use ANCOVA, because it requires post-score as the dependent variable, and pre-score as the predictor. I also thought of using the difference of pre-/post- score as the outcome, but i'm not sure what the predictor would be in that model. 

I'm planning on using proc glm. So the code will be something like:

"proc glm data=have;

model difference= ? /clparam;"

So in short, I want to test the effect of intervention in score.

Thanks for any tips and helps! 

[PaigeMiller]

Now I need to get parameter estimate (coefficient) of the post score  in a linear model

What linear model are you thinking about? What is the independent (or predictor) variable? What is the response variable?

We can't really tell you what the model is, it's your problem and your data, you have to tell us what the model is.

[zta3]

The main predictor would be the use of the intervention. So instead of using a paired t test, I'd like to see the effect in the linear model if it's possible.  

So I'm thinking

proc glm data=have;

model diff=/solution; 

run;

Is it just basically a one sample ttest? 

[Rick_SAS]

Correct. If all you want to know is whether there is a significant difference in the means between the pre- and post-intervention scores, then you can use a one-sample t test on the paired differences.  The t test assumes that the population of difference is normally distributed. If you don't believe that assumption, you can use the nonparametric Wilcoxon signed rank test.  Both of these can be run in SAS by using PROC UNIVARIATE on the (post-pre) differences. See the first section of the article

https://blogs.sas.com/content/iml/2023/07/19/wilcoxon-signed-rank.html

[zta3]

Thank you so much!

[Rick_SAS · Accepted solution]

It sounds like your response variable is the (post-pre) difference. If Diff is that difference, then you can include covariates (age, sex, ....) in GLM like this:

proc glm data=Have;
class Sex;
model Diff = Age Sex / solution;
quit;

[Ksharp]

I think you could try GEE model.

https://support.sas.com/kb/46/997.html
https://support.sas.com/kb/23/127.html