Brent:
There are a number of ways to interpret your questions. Assuming that you
are asking about alternative ways to analyze the same 59 pts.
retrospectively, there is only one way to calculate the odds ratio in this
situation (okay, two ways if you count the inverse). It doesn't matter
whether you first partition into those with and without Ursodiol, or those
who had colonic dysplasia and those who didn't. To construct the 2 x 2 table
to calculate the odds ratio you have to do both. You are talking about
looking at this 2 x 2 table straight up or rotated a quarter turn. You still
get the same odds ratio either way. Using standard methods you also have the
same power and get the same confidence interval, which is based on the
standard error, calculated as the square root of the sum of the inverse of
the four cells of the 2 x 2 table. The relative risk of having colonic
dysplasia with or without use of Ursodiol is also the same, no matter how
you initially divide up the results. However, the absolute difference
calculation gives a different result depending on whether you are asking
about the difference in colonic dysplasia rate in those with and without
Ursodiol, or the difference in Ursodiol use rate among those with or without
colonic dysplasia. These different proportional differences will have
different powers and different confidence intervals. The former is a
difference of 32% - 72% or a decrease of 41 percentage points (95% CI: 13%
to 60%); the latter is a difference of 50% - 85% or a decrease of 35
percentage points (95% CI: 11% to 55%).
If on the other hand you are asking about the best experimental design to
use if you could do the study over prospectively, that is a more complicated
question. We would need to know more about the action of Ursodiol, the
natural history of colonic dysplasia, and the diagnostic efficacy of
colonoscopy surveillance and biopsy, as well as the time and economic
resources available. In most medical situations, the most information is
gained by multivariate analysis.
Hope this is helpful.
David L. Doggett, Ph.D.
Senior Medical Research Analyst
Health Technology Assessment and Information Services
ECRI, a non-profit health services research organization
5200 Butler Pike
Plymouth Meeting, Pennsylvania 19462, U.S.A.
Phone: (610) 825-6000 x5509
FAX: (610) 834-1275
e-mail: [log in to unmask]
-----Original Message-----
From: Beasley , Brent [mailto:[log in to unmask]]
Sent: Monday, February 19, 2001 5:52 PM
To: [log in to unmask]
Subject: Cross sectional studies?
RE: Annals of Intern Med Jan 16th, 2001, vol 134(2):88-95.
Hello Colleagues,
The study is cross-sectional. 59 pts with Ulc Colitis who get surveillance
colonoscopies. Analysis was done based upon dividing this group into those
with vs. without use of Ursodiol (41/59). Use of Ursodiol was assessed 1)
by chart records and 2) by direct patient contact if not found in records.
Association of interest was Colonic dysplasia assessed by biopsy (26/59).
Ursodiol was strongly associated with decreased prevalence of colonic
dyplasia (32% vs. 72%, OR=0.18).
Questions:
It seems this study could have been done in a case-control format, dividing
groups on the basis of dysplasia.
1. What would have been the BEST methodology?
2. Which methodology would be most conservative/least biased?
3.Does it matter?
Brent W. Beasley, M.D.
St. Luke's Hospital, Dept. of Medical Education
4401 Wornall Rd
Kansas City, MO 64111
(816) 932-3409
Fax: (816) 932-5179
Pager: (816) 440-3968
[log in to unmask]
"When men have come to the edge of a precipice,
it is the lover of life who has the spirit to leap backwards,
and only the pessimist who continues to believe in progress."
--G. K. Chesterton
|