Trend analysis using survey data?

[SAS93]

I'm using SAS to look at prevalence of a dichotomous variable (y/n--most interested in "yes" responses) from survey data by year; I want to know if there's any significant increase/decrease over time.

I've seen similar posts like this:

I am now trying to run the trend analysis from the NHANES survey data. In general data, the trend analysis are simple for continuous variable, which can be run like

proc freq data=data;
table CycleYear*age/JT; run;

 and categorical variable, which can be run like

proc freq data=data;
table Diabetes*CycleYear/trend;
run;

This question doesn't have an answer, however. 

So, how do I appropriately look at simple trends of overall prevalence over time using procedures that account for complex survey data?

[ballardw]

Dichotomous outcome usually points to Proc SURVEYLOGISTIC as the tool with complex survey data or Proc Logistic.

One major consideration you have when dealing with multiple survey periods, which I assume you have, is getting an appropriate weight for the variables. If a single year of survey data is weighted to reflect the population total then when you combine 3 years of data the sum of population estimates will be almost 3 times the last year's population total.

You would have to have a variable for year to place on a strata statement since year becomes an element of the stratification and then a separate variable for data year since SAS procedures will complain about using the same variable for strata and analysis.

There are a number of approaches to reweighting the data, either adjust everything to use a middle year period if you have have "nice" interval that has an actual interval, pick one of the years as reference and reweight that year's population totals. Or ignore the population totals, which means you cannot ever report an estimated population size (percentage okay but not n) and don't reweight at all.

The longer the interval the data covers the more the base population is likely to change the influence of years will change if the data weights are not standardized.

So, how many years are involved?

[gcjfernandez]

Trend analysis involving NHANES survey data is a very complex subject because of the nature of the probability survey designs used  every two years.

Please refer these online resources first and come out with an analysis plan:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5961819/

https://19january2017snapshot.epa.gov/sites/production/files/2015-06/documents/statcomparisonsnhis-n...

Thanks

George