Model for Correlated data

[analyst_work]

Hi, 

I am trying to decide which model would best be suited for my study. I have two continuous variables A (independent) and B (response). We are measuring these for 20 hospitals for each month across the year 2020 - so each hospital has 12 values for each month for each variable. I am currently using proc genmod with the repeated statement to account for clustering within each hospital. Wondering what might be a good way to account for time in this case? Sharing my code below:

proc genmod data = dataset;

class hospid;

model B = A ;

repeated subject = hospid;

run;

[StatDave · Accepted solution]

You can specify your Month variable in the WITHIN= option in the REPEATED statement and then, if desired, use a correlation structure such as TYPE=AR that allow the correlation strength to diminish with time.

[analyst_work]

Is there a way to replicate this using proc glimmix instead?

[jiltao]

proc glimmix data = dataset;
class hospid month;
model B = A ;
random month / residual subject = hospid type=ar(1);
run;

[analyst_work]

Thanks! One follow-up question - my variable A is a continuous variable, and will not appear in the class statement. How do I use lsmeans (or something similar) to get the rate estimates in this case? Thanks!

[jiltao]

What exactly do you want to estimate for A? 

Depends on what you want to estimate, you might use the S option in the MODEL statement to get the slope estimate for A, or use the ESTIMATE statement. For example,

proc glimmix data = dataset;
class hospid month;
model B = A ;
random month / residual subject = hospid type=ar(1);

estimate 'response when A=5' int 1 A 5;
run;