What else can you do? Of course, unless the expected
incident differences are very large, there is little hope of
obtaining a meaningful result. You can't assume a binomial
because of the high correlation of results within a ward --
underdispersion. Even if you could, the sample sizes
required for independent binomial values are large. For
example to detect a one sided difference between an
incidence of 0.25 and 0.025 with a size 0.05 test at a power
of 0.9 requires about 95 total independent binomial values
-- about 47 in each group.
Bernard Higgins wrote:
>
> Hi
>
> I have been approacted by a specialist registrar wishing to conduct a study in which
>a
> treatment is to be given to patients to reduce the incidence of diarrhoea. He plans
>to
> use two wards in one hospital. All new patients attending one ward will be allocated
>to
> the test treatment and all new patients in the other ward will form a control
>group. The
> outcome of interest is incidence of diarrhoea during one month of follow up. He then
> plans to allow for a "washout" period, of up to one month, after which the two wards
> crossover to the other treatment. His question was; how many patients do I need? (He
> didn't think about how he might analyse his results!)
>
> My first reaction was that this is a cluster randomised trial with one cluster per
>arm
> (ignoring the "crossover" for the moment) and in the absence of any background
> information regarding likely incidence in the control group, I was not in a position
>to
> offer much advice.
>
> I do not feel that a crude linear contrast of incidence rates involving the two
>hospital
> wards over the two treatment periods is really appropriate. What do othes feel? Can
> anyone suggest a sensible way forward?
>
> regards
>
> Bernie
> -----------------------------------------
> Bernie Higgins, email: [EMAIL PROTECTED]
>
> Department of Mathematics, University of Portsmouth
> Room 18, Floor 1, Buckingham Building
> Lion Terrrace, Portsmouth PO1 3HE
>
> Tel 023 92 846361 Monday & Tueday
> 023 92 286236 Wednesday - Friday
>
>
>
--
Bob Wheeler --- (Reply to: [EMAIL PROTECTED])
ECHIP, Inc.
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
