How to compare patient-time data between groups

#1
I have a retrospective clinical dataset containing one column with general follow-up time and particular columns for each unwanted outcome containing number of unwanted outcomes during follow-up period (hospitalization, transplantation, death). Since follow-up time is not outcome specific (except maybe death), I guess I need to compare outcomes between baseline subgroups using patient-time / patient-years approach. But I have a problem how to properly compare information on incidence rates between subgroups.

Does anyone have experience what statistical test can I use to compare incidence rates between subgroups?
What comes in mind is calculating 95% confidence interval for incidence rate in each subgroup and see if intervals cross, but I don't know how to do multivariate analysis based on this approach?

Thank you
 
Last edited:

Karabiner

TS Contributor
#2
What about using Kaplan-Meier survival analysis? E.g. if you want to analyse transplantation, then you have observation periods limited by the event of interest (transplantation) and observation periods limited by other events (censored cases). You could then compare survival estimates between groups.

With kind regards

Karabiner
 
#3
Unfortunatelty this is unknown. I only have information about general followup and events some time during this period - therefore cox regression is not an option i guess
 

hlsmith

Not a robit
#4
So you have multiple outcomes, but don't always have time to event/censoring time data for these events? If so, you may be looking at using Poisson regression.

Also, it sounds like proportional hazards regression is what you should be using and possibly controlling for competing events (e.g., death).
 
#5
Thank You guys for your suggestions. As I found out in the meantime I should first transform my data to be able to analyze it with Poisson regression in a way that every patient-year with specific properties makes one row of data, I have to deal with optimal way to do this though. Does anyone know fast way to do this?
Cox (proportional hazards) regression does not seem like an option since I do not know follow up time till specific/composite event of interest, only total follow time.