Interpretation of GEE output from Proc Genmod with repeated continuous outcome

[Kunko]

Hi there,

I am new to using GEE concepts to analyse longitudinal data. Would you please help me with interpreting the GEE output table with 2-time points of BMI, I am intending to evaluate the effect of income on the average change in BMI over time?  Also, can I use 95% CI in the table?

Thank you so much in advance.

Tol 

Analysis Of GEE Parameter Estimates
Empirical Standard Error Estimates
Parameter		Estimate	Standard Error	95% Confidence Limits		Z	Pr > |Z|
Intercept		21.9915	0.3007	21.4021	22.5808	73.13	<.0001
Time		2.5613	0.2118	2.1462	2.9764	12.09	<.0001
Middle_income	1	-0.9585	0.4173	-1.7764	-0.1405	-2.30	0.0216
High_income	2	-1.8587	0.4210	-2.6838	-1.0336	-4.42	<.0001
Low_income	0	0.0000	0.0000	0.0000	0.0000	.	.
Time*Middle_income	1	0.5388	0.2992	-0.0476	1.1252	1.80	0.0717
Time*High_income	2	0.6464	0.2835	0.0908	1.2020	2.28	0.0226
Time*Low_income	0	0.0000	0.0000	0.0000	0.0000	.	.

[StatDave]

To get single tests of each model effect (main effects of time and income and their interaction), add the TYPE3 option in the MODEL statement. If the interaction is significant, then that means that the effect of income depends on the time (and vice versa). if not, then you could consider removing the interaction term and then you'll have a test of the overall effect of income. 

[Kunko]

Thank you so much for your detailed explanation. I have already added Type
3 in Model statement and the interaction term is significant. So, I wanted
to have the average estimates of BMI for each categories of income with
their respective 95% CI. Thanks you again for your unreserved support.

Kind regards,
Tol

[StatDave]

Since you say that your TIME variable is binary, add it in the CLASS statement, then add a SLICE statement to obtain the estimated mean at each income at each time. For details on the SLICE statement, see the documentation for that statement.
slice income*time / sliceby=time means ilink cl diff;

[Kunko]

Dear StatDave,

Thank you again for your further information.

What I need is NOT estimated mean for each income at each time, but mean change in BMI over time for each income level, with their respective 95% CI.

For instance from the original SAS output table above, I can see (from my little understanding from what I read) mean BMI change at time 2 for those in low income level is 2.5, with 95% CI of 2.15, 2.98). The corresponding values for middle and high income levels are 2.5 (value above for low level group) + 0.54 = 3.09 and 2.5 + 0.65 = 3.15. This is how we can do manually and I am NOT sure we follow the same principle of adding 95% CI for the latter mean BMI values of middle and high income levels.

Sorry if I am wrong in interpreting the results. I am just learning how to calculate estimates from GEE and interpret the results to apply to my data set. 

Kind Regards,

Tol

[StatDave]

I think what you want is this equivalent formulation of the model that gives you the effect of time in each income.

class id income;
model wheeze = income time(income) / noint ;