Prevalence Rates with Cross-sectional survey data

[KMack]

I am  analyzing the relationship between a binary outcome and mixture of categorical/continuous exposure variables.  The data comes from a cross-sectional survey with complex sampling methodology. 

I would like to generate prevalence ratios (not odds ratios) while working within the framework of one of the complex survey methodologies.  It seems to me genmod cannot work with complex survey methogogy but CAN produce prevalence ratios while surveylogistic works with complex survey methodology but cannot directly produce prevalence ratios.  Is this correct?  Thanks

[KMack]

I wanted to follow-up on my question:

I've been reading a bit more and found that I can use a cox model to estimate prevalence rates from cross sectional data.  Since there is a survey procedure (surveyphreg) available, this seems like the best choice for me to both account for the sampling strategy and to get the appropriate effect estimate (prevalence rate) .  However, I'm not sure how to construct my model.  I seem to be getting inappropriately small confidence limits and I'm *sure* it has to with incorrect coding.

This article suggests that a robust estimator is needed in the Cox model, but aren't all variance estimators in surveyphreg robust? 

DOAJ -- Directory of Open Access Journals

My (simplified) code is:

proc surveyphreg;

class exposure;

model outcome=exposure;

strata stratax;

weight weight;

run;

I've never used cox for x-sectional data before and I'm not sure how to write my model statement (it seems to me it would be different from a prospective study with follow-up time, yes?)

[choffmire]

Hello,

I was wondering if you have recieved any feedback from others regarding this question - I am in a very similar situation. My main question also related to weather using the robust variance estimators in a surveyproc is reasonable or required as surveyphreg already uses taylor series, balanced repeated replication, or delete-1 jack-knife methods.

Barros suggests using the robust sandwhich estimators in his paper to more accurately esitmate the CI for the prevlanence ratio - which it seems you can request directly in the regular PHREG proc as the COVS option on the PROC PHREG statement.

I do not think you can even request this in SURVEYPHREG, nor would it be appropriate.

If you have any follow-up insight to share since posting this question, I would love to hear it. Thanks!