Reply
Wed 23 May, 2018 07:30 am
It was suggested to me to analyze my data using a CACE model. I have a randomized cluster trial. Patients were either assigned to intervention hospitals and offered telehealth services or to control hospitals where telehealth was not offered. Most of the patients in the intervention facilities did not use telehealth. A small number of patients in the control facilities did use telehealth. So, we have non-compliance in both groups. I'm struggling to find a software package that includes all the elements I need. I need to include a random effect for facility (hospital). The outcome is binary and the compliance indicator is binary. The patient either used telehealth or they didn't. I can't find anything in SAS (this is the software I know best and prefer). I don't have access to MPlus. I've tried caceCRTBoot in the R eefAnalytics package but it doesn't appear to like my binary compliance indicator. I coded it 0% or 100% to look like a percentage. Won't run. I posted the error message on StackOverflow but no one has responded. I looked at CACEcluster from the R experiment package. It requires a matched-pair design and ours is not a matched pair design. I looked at the mediation package in R and a CACE option in Stata but both use an instrumental variable approach and we don't meet the assumptions of IV for all our outcomes. Does anyone have a SAS macro that would do this? Or know a way to make this work in R or Stata? I would very much appreciate any advice you are willing to give. Thank you.