Re: How to compare these two rates??
- From: valter.sundh@xxxxxxxxxxxxxxx
- Date: 16 Jan 2006 08:05:00 -0800
If it is the total number of medicines in each group you have measured,
then a natural test statistic would in principle be the average number
of medicines per person in the two groups. But in practice this would
not work well, as the number of medicines per person has a very skewed
distribution: the large majority have none, some have 1 or 2 and a
small number may have many (10-15) medicines.
I would present an analysis of this kind as a contiguency table,
something like this:
no drugs 1 drug 2-3 drugs 4-9 drugs 10 or
more
treatment f1 f2 f3 f4
f5
control e1 e2 e3 e4
e5
and test the difference with the Cochran-Armitage test for trend.
If confidence intervals are required, I would present them for the
proportion
of persons with one or more drugs in each group, or maybe for the
proportion of persons with more than one drug, depending on what seems
most suitable for the data I have.
An other possibility is to calculate a logistic regression model from
this continguency table and report the odds ratio (with CI) that will
describe how the odds of belonging to the treatment group increases (or
decreases) for each higher level of the medicine scale.
Valter Sundh
.
- References:
- How to compare these two rates??
- From: pp
- How to compare these two rates??
- Prev by Date: Re: How to compare these two rates??
- Next by Date: Re: Survival Curve - How to Add Number at Risk ?
- Previous by thread: Re: How to compare these two rates??
- Index(es):
Relevant Pages
|