Weighted GEE for continuous outcomes

[Tol]

Dear All,

I am trying to examine the effect of the fat intake at 30 years on the BMI change over 9 years.

My data had some dropouts which I wanted to account for using Proc GEE with Missmodel statement to calculate weights that are used for reducing biased estimates related to dropouts. I can see we need to create two variables (Prevy and Ctime) to prepare the analysis (SAS Help Center: Example 47.3 Weighted GEE for Longitudinal Data That Have Missing Values). But the explanation for creating Prevy is for the binary outcome (Amenorrhea status) in the previous three-month interval given this study was conducted every three months over a year at 4-time points (See the code below). My question is how can I create Prevy for a continuous outcome. Thank you in advance for your support.

Kind Regards,

Tolassa 

data Amenorrhea;
   set Amenorrhea;
   by ID;
   Prevy=lag(Y);
   if first.id then Prevy=0;
   Time=Time-1;
   Ctime=Time;
run;

[SteveDenham]

The code you present should work whether Y is continuous or categorical, I believe.

SteveDenham

[Tol]

Dear SteveDenham,

>From SAS Help Centre link above it reads "Prevy: the patient’s amenorrhea
status in the previous three-month interval. For the baseline visit, this
is set to an arbitrary nonmissing value (0 here). In this release of PROC
GEE, this arbitrary value must be nonmissing and valid for the response
variable—for example, it should be 0 or 1 for a binary response—but it does
not otherwise affect the results."

As mine is a continuous outcome variable (BMI in kg/square meter), I am
confused with which arbitrary value should I use. Below is a range of my
BMI at two-time points. I can see either 0 or 1 is NOT a valid value in my
case. Can I use any value in the range of BMI as an arbitrary value to
calculate Prevy? Thank you again and appreciate your unreserved support.
Variable N Mean Std Dev Minimum Maximum
MBMI1
MBMI2
2604
1712
24.1115344
27.1624473
4.9327651
6.1337117
14.6923783
15.6507128
50.2721204
67.7393165

[SAS_Rob · Accepted solution]

It does not matter what value you use as long as it is a valid value for the distribution of the response.  In other words, you could still use 0 as long as 0 is in the support set for the DIST= option even though you won't actually observe a 0 value.