Logistic Regression Maximum Likelihood p-value doesn't match Odds Ratio

[David17]

I am running a proc surveylogistic regression analysis on a binary outcome.  I am getting a significant p-value for the analysis of maximum likelihood estimate (p=0.0347) for Black/Hisp/Asian vs. White, but the corresponding odds ratio is not significant:  OR 0.928 (95% CI 0.720-1.196).  I wanted to know if someone could help explain what might be going on and how to interpret these results.  Can this be considered significant or "borderline"? My understanding is that they should match.

[StatDave]

Be sure you are using the PARAM=GLM or PARAM=REF option in the CLASS statement. Note that the computation of the odds ratio depends on the model and the parameterization used for categorical (CLASS) predictors. For detailed explanation, see this note. Also see this note.