I have a data set of UB92 records. I am analyzing the costs and charges for various groups. Some groups are broken down into subgroups. For example, patient with a certain diagnosis are broken down into male and female. I've been using an unequal variance t-test two compare the mean charges of the diagnosis group against that of the non-diagnosis group. However, I have also calculated sex-adjusted mean charges by weighting the means of the male and female subgroups by the proportion of males and females in the standard population. This is the same as calculating sex-adjusted incidence rates. The problem is that now I am not sure how to calculate the two sample t-test to compare the sex-adjusted mean charges of the diagnosis group vs the non-diagnosis group. I don't want to simply use the sex-adjusted variance of the groups. This seems mistaken because the distribution of male and female in the sample groups is also a random variable, which adds extra variation to the adjusted means. Can I get some help? It would be much appreciated.
First, ditch the t-test. The CMS cost data are so skewed that even the unequal variance approach to the t-test is inadequate.
If you want to do adjusted analyses, then you will need to use some of the approaches to cost analyses described by Kevin Anstrom and Anastasio ("Butch") Tsiatis. I think the articles are in JASA or Biometrics.
An older (and less satisfactory) way to get adjusted tests and estimates is "smearing" (N. Duan, JASA 1983); you'll still see that reference in some in the econometrics literature.