Toby is having problems with sending to PRIMSTAT and has asked me to send
this to the list on his behalf. This must be a record for the number of
e-mails to PRIMSTAT in one day!
Sandra
----- Original Message -----
From: "A T Prevost" <[log in to unmask]>
To: "Sandra Eldridge (E-mail)" <[log in to unmask]>
Sent: Thursday, June 17, 2004 4:20 PM
Subject: FW: randomisation/minimisation
> but could you forward this as a reply to the list; my email with primstat
> has an identity problem...
>
> -----Original Message-----
> From: A T Prevost
> Sent: 17 June 2004 09:50
> To: 'Medical Statisticians interested in primary care'
> Subject: RE: randomisation/minimisation
>
>
> Dear Robert,
>
> yes I would say you are right to be concerned, because the balance would
not
> be effective with such small strata. With so few patients per clinic and
the
> three arms in the trial, I don't think minimisation would be useful
either;
> and may make the randomisation too predictable which is more of an issue
if
> clinics are both able to keep track of assignments and also recruit
> patients.
>
> Perhaps it would be possible to stratify on something other than clinic,
for
> example a characteristic connected with clinics that is predictive of
> outcome. Another option may be to move to a higher geographical level e.g.
> PCT. Presumably with a small number per clinic, you will have a large
number
> of clinics and so clinic effects should balance out between arms quite
> reliably without stratification by clinic.
>
> best wishes
> Toby
>
> Dr AT Prevost
> Institute of Public Health
> University Forvie Site
> Cambridge CB2 2SR
>
>
> -----Original Message-----
> From: Blizard, Robert [mailto:[log in to unmask]]
> Sent: 17 June 2004 09:18
> To: [log in to unmask]
> Subject: randomisation/minimisation
>
>
> I have been asked to help with randomisation in a General Practice trial.
It
> is straight forward but has one complication. If anybody has experience
of
> this, I would be grateful for your suggestions.
>
> The trial has three arms (A,B,C). The subjects need to be stratified by
> disease severity(hi,lo) and GP clinic. Each arm requires equal numbers
and
> the ratio of patients in each severity level is unknown. Under normal
> circumstances, I would naturally prepare two block randomised lists, one
> for each severity level, in each clinic. The potential difficulty arises
> because the expected number of subjects in each clinic is 5.
>
> In perhaps the majority of clinic/severity combinations, there will be
> unfilled blocks, often insuffucient subjects to fill a single
randomistion
> block. Am I right to be concerned about this ?
>
|