Thanks in advance!
I have a pre-post analysis Of cases and controls. The outcome variables of interest are emergency room visits and hospitalization visits - both rare occurrences with median = 0 in both periods. I want to evaluate if the change in pre-post is different between cohorts.
Each cohort has 48 months of enrollment pre-intervention and 36 months post-intervention. There are 500 cases and 17,000 controls. Cases and controls are demographically not different and are paired with themselves pre and post periods.
Several issues arise: 1. This is non-normal count data. So I summed the number of visits per period and ran a Wilcoxon sum rank test to assess the pre-post difference between groups; a wilcoxon sign test to assess within group differences. I did not control for the unbalanced pre and post months. Is this okay?
I’m getting heat for having unbalanced periods. So I created a per-member-per-month rate for each subject, pre-post. The data is still nonparametric but scaled over the number of months. What test is best suited for this - sum and sign rank?
Folks want to see the percent difference in utilization. So I calculated a percent of user for each period. This is rolls up to th group level. So there are two observations per cohort for users:pre-users and post-user. I tried sum and sign tests, but there aren’t enough records. I do capture the denominator here and the number of users and percent. What is the best approach to assess if the change in the percent of users before and after the time period is different between cohorts?
Is the McNemar test appropriate?
Suggestions and ideas welcomed!