Forgive me if this is a very basic question. I am using a large database of healthcare encounters that uses a stratified sampling approach. Each row in the database is a sampled encounter, and has a sample weight and stratum. A small subset of encounters will result in death. There are also covariates including income quartile and insurance. If I summed the weights of those who died by income quartile and insurance, I would get the number of deaths by income/insurance, and I could calculate the confidence interval. I have census data that has the number of individuals by income quartile/insurance, and could use that to calculate mortality rates (which would be an incidence).

I'd like to take an additional step and calculate mortality incidence rate ratios (and 95% confidence intervals of those IRRs) for the covariates (income quartile and insurance). I may be approaching this wrong, but it is not clear to me how to use an offset variable when the counts are the sums of the individual-level data. And if I pre-calculate the sums by covariate pattern and compute a non-survey-weighted Poisson regression using the weight sums as the counts and the census data for the offset information, I am no longer taking the survey design into account.

Would love your thoughts on how to approach this - my sense is it is something simple, but I am not able to find this question previously answered in the forum. Thanks so much!

  • $\begingroup$ Are you using survey procedures in SUDAAN, or SAS by chance? $\endgroup$ – StatsStudent Jan 9 at 18:49
  • $\begingroup$ So sorry - I should have specified. I am using the survey package in R. $\endgroup$ – Ken Jan 9 at 19:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.