In binary logistic regression, how do you interpret a categorical-by-categorical interaction term?

[Ahinoa]

Hello all,

I am working with multiply imputed data and I have run a logistic regression model with 6 predictors (3 dichotomous and 3 categorical) and their interaction terms, controlling for a number relevant covariates in SAS.

The predictor variables are as follow:
CVD (0 = No; 1 =Yes)
HCA (0 = Low; 1= High)
EDU (1=High School Dropout; 2=Graduated High School; 3=Some College; 4=Graduated College)
POV (1=Low; 2=Medium; 3=High)
LAN (1=Spanish; 2=English)
STA (0=Old, 1=New, 2=None)

My outcome variable is diabetes medication use and is coded as 0 = no use and 1 = current use

I found that for the following:
1. One of the levels of the interaction of EDU*STA is marginally significant.
2. Only one of the levels of the interaction of POV*STA is marginally significant.

My questions are:

1. How do you decipher a categorical-by-categorical interaction when only one level of the interaction is marginally significant?
2. Would it be appropriate to calculate the adjusted odds, as a way to interpret the interaction?
3. If 95% CI do not include 0 but the CI overlap, does it means that the results are not statistically significant?

Thank you

[StatDave]

Rather than look at the parameter estimates table, look at the table labeled either as "Type III Tests" or "Joint Tests". This table gives you a single test for each effect in the model, so there will be a single assessment of the significance of each of your interactions. If an interaction is significant, then that means you need to make comparisons among the levels of one predictor at a fixed level of the interaction predictor. There are several statements that simplify this, but the easiest to use are the LSMEANS or SLICE statements. See this note which illustrates all of them. 

[Ahinoa]

Thank you for the quick reply!!! 

I do understand the rationale for using the SLICE statement, but since I am fitting a multivariable logistic model using multiple imputed data (10 datasets) the issue I am running into is that I cannot figure out a way to use PROC MIANALYZE to pool the estimates. Is there a way to work around this issue?