[EMAIL PROTECTED] (Scheltema, Karen) wrote in message news:<[EMAIL PROTECTED]>...
> I am struggling with how to analyze the following design.
>
> Morphine is known to cause urinary retention, which can result in increased
> discomfort, LOS, $$, etc. I have an experimental protocol covering three
> types of surgeries in which patients are randomly assigned to treatment as
> usual or an additional drug to counteract the side effects of morphine.
> There are two outcome variables, urinary retention (yes/no) and pain level (0-> 10).
> Urinary retention is measured every 8 hours and pain level is measured > every 4
> hours. The catch is that not everyone is on morphine for the same
> length of time. In addition 24 hours of additional data is collected after
> the morphine is discontinued because the side effects of morphine hang around > that
> long. How do I account for some people being on morphine for 24 hours (a > total of
> 6 measures of urinary retention), and others for 48 hours (a total
> of 9 measures of urinary retention)?
>
> Karen Scheltema, M.A, M.S.
> Senior Statistician
> HealthEast
> Research and Education Department, Midway Campus
> 1700 University Ave W
> St. Paul, MN 55104
> Ph: (651) 232-5212 fax: (651) 641-0683
> mailto:[EMAIL PROTECTED]
>
> .
Hi Karen,
First, please preview you posting - the text didn't wrap and was
hard to read (at least using Google groups). That said, off the top of
my head...
Try looking at SAS System for Mixed Models, by Littell, et.al.
Even if you don't use SAS, the examples should help you, at least with
the single outcome Pain Level. If you don't have that, I think
documentation for PROC MIXED has some relevant examples; e.g.
longitudinal data, different #measures/time points per individual.
You can try treating time on drug as a covariate, e.g.,
Pain=Intercept + Time on Drug + Covariates
and then account for the correlated outcomes by specifying the error
variance matrix
To look at Retention alone, look at GEE or Generalized Mixed Models.
To look at both, you may want to think about a hierarchical
(multi-level model; e.g.
Retention=Intercept + Pain
Pain=Intercept + Time on Drug + Covariates
I'm not very familiar with Gen. Mixed Models, but I'm Guessing it can
handle hierarchical models with both Discrete and Continuous outcomes.
Or try a Bayesian approach.
Of course, the above is just a start, and not necessarily *the* only
(or correct) way to go.
HTH and good luck
- Iyue
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
