There was a review in The Lancet ~ 6-7 years ago that mentioned this
(afraid I can't be any more precise than this). Their conclusion was
that subgroup analyses are rarely valid, especially if they lie in the
opposite direction to the whole-group results (which is when they are
interesting....)
HTH,
Matt
Brian Alper MD wrote:
> I would suggest a pre-planned stratified randomization with analyses within
> the stratified groups could provide level 1 evidence. This could be called
> a "subgroup analysis" but would be the equivalent of independent randomized
> trials being reported as a single trial.
>
> All of the usual considerations for bias would have to be considered at the
> level of the "subgroup" -- adequate sample size, adequate follow-up rates,
> intention-to-treat analysis, etc.
>
> In addition consider whether the effects of other trial activity (including
> statistical evaluations) could in any way bias the chances of finding and
> reporting significant differences in the "subgroup".
>
>
> If the randomization was not stratified by the subgrouping, then the
> clustering of confounding factors along with the "subgroup factor" could
> nullify the initial reason for randomization. Adjusting for recognized
> confounding factors cannot exclude bias from unrecognized confounding
> factors.
>
> --------------------------------------
> Brian S. Alper, MD, MSPH
> Editor-in-Chief, DynaMed (www.DynamicMedical.com)
> Medical Director, EBSCO Publishing
> 10 Estes St.
> Ipswich, MA 01938
> office (978) 356-6500 extension 749
> cell (978) 804-8719
> fax (978) 356-6565
> home (978) 356-3266
> "It only takes a pebble to start an avalanche."
> -----Original Message-----
> From: Evidence based health (EBH)
> [mailto:[log in to unmask]] On Behalf Of Olive Goddard
> Sent: Wednesday, May 09, 2007 4:14 AM
> To: [log in to unmask]
> Subject: Re: Subgroup analyses - are they ever best evidence
>
> Dear Colleagues,
>
> Would anyone be prepared to respond to this query from Gayle Robins.
>
> All good wishes,
>
> Olive
>
>
>
>>>> "Robins, Gayle" <[log in to unmask]> 09/05/2007 06:01
>>>>
> Hello Mrs Goddard
>
> When I evaluate a clinical trial, I use the Oxford Centre for Evidence
> Based Medicine's recommendation as a guideline for whether the
> information provided by the trial could be considered as best
> evidence.
> I note that individual randomised controlled trials with narrow
> confidence intervals are considered as level 1b on your levels of
> evidence chart. Please can you advise me where, if at all, subgroup
> analyses of these level 1b randomised clinical trials would fall on
> the
> best-evidence hierarchy.
>
> I realise that there are many different types of subgroup analysis:
> those that are defined apriori versus retrospectively or on an ad-hoc
> basis; those that address the same outcome of interest that the
> randomised controlled trial was designed to assess versus other
> outcomes; and those that are part of a plethora of subgroup analyses
> of
> the same trial and so require correction for multiplicity, to name a
> few.
>
> Are any subgroup analyses of level 1b randomised controlled trials
> ever
> considered best evidence?
>
> Thanks you for your time taken to read this email. I would appreciate
> any advice that you can give me, or people that I could contact, in
> this
> regard.
>
> Gayle Robins
> Team Leader
> Clinical Trials Insight
>
> Adis International
> Wolters Kluwer Health
> Ph: 09 4770700
> Email: [log in to unmask]
--
http://acl.icnet.uk/~mw
http://adhominem.blogsome.com/
+44 (0)7834 899570
|