Hi,
I was hoping that somebody with a bit more experience with this sort of analysis might be able to clarify something for me. I am conducting a project looking at patients with cancer who are matched to up to 4 controls each. One of the outcomes I am interested in is comparing the number of general practice consultations each cancer patient had, compared with their matched controls. My hypothesis is that cancer patients will consult more frequently.
My outcome variable is the number of consultations (total_consults) over 3 years, which is a count and is highly overdispersed. Through using -prcounts- I have established that this data fits a negative binomial distribution. My explanatory variable is whether or not the patient had cancer (cancer, coded 0=control 1=case). I also have a variable indicating the case/control matches (groupid).
I think I should be using conditional negative binomial regression, similar to conditional logistic regression for dichotomous outcomes. The command I am using is:
. xtnbreg total_consults cancer, fe i(groupid)
Will this work for this type of analysis? From my understanding, using a fixed effects model will give me an overall estimate based on all of the matched case/control groups, within which I am testing whether the mean consultations of matched controls is equal to the single value of the matched cancer case in that same matched group. Is this interpretation correct?
Sorry for the long question, but I haven't been able to find the answer to this question anywhere else...
Thanks
Nada
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
