i think that this has been one of the most interesting discussions ever
held on e-b-h.
it deliciously combines science, clinical practice, and grub street (the
world of struggling authors and those who publish them).
in response to the questions posed to me following my earlier note (that,
as editor of an e-b journal, i reject "indeterminate" negative studies):
1. i think there is a strong case for demanding that every piece of
research involving humans MUST be published in some forum, for reasons of
both science (to permit later incorporation into systematic reviews) and
ethics (it strikes me as wrong to subject folks to, at the least, invasion
of their privacy without some common good coming from it).
2. and lots of folks brighter than me hold the same view and are doing
something about it:
a. the move toward compulsory registration of all RCTs (could/should be
extended to other architectures).
b. the amnesty on RCTs now being offered by some journals.
c. the brilliant decision by drug firms (schering first, and now
glaxo-wellcome) to give all their trial data (published and unpublished)
to the cochrane collaboration as soon as the Ctee on the Safety of Drugs
has ruled on it.
3. but from my perspective, the communication of indeterminate results is
most appropriate when conducted in (to borrow from brian haynes's
terminology) "scientist to scientist" vehicles [data banks and the
sub-specialty journals], and they are of insufficient relevance to
clinicians (whose reading times are 30 mins a week or less) to go into
journals dedicated to "scientist-to clinician" [the general medical
journals] or "clinician-to-clinician" [the e-b journals] communication.
4. a nifty pioneer in this field, tom [not the other nifty pioneer, iain]
chalmers urged investigators to go bayesian (that is, use the total of all
prior study results) in setting sample sizes and calculate how many MORE
patients they'd have to study in order for the aggregate to achieve an
importantly narrow confidence interval. (this is part of a larger issue
that [fairly, i think] tweaks our noses for being ready to incorporate
"priors" into diagnostic decisions (pre-test probabilities) but not into
therapeutic decisions (part of the hedge here is that our diagnostic
decisions apply to just one patient, but our therapeutic decisions may
apply to several; but, as steve simon suggested, we might want to accept a
wider confidence interval when deciding therapy for just one patient, as
we often do in our "N-of-1" RCTs).
5. paul kamill's analogy to "indeterminate" lab results (likelihood ratio
= 1; ROC on the 45 degree diagonal; sens=spec=0.5) was nifty and helpful.
but when i follow it to my logical conclusion, getting an indeterminate
test result tells me i've learned nothing by it and i'm no wiser than
before. for that reason, i'd want the indeterminate result safely tucked
away somewhere (for inclusion in a systematic review), but not in my
journal!
sorry for this rambling.
cheers
dls
............................................................................
Prof David L. Sackett
Director, NHS R&D Centre for Evidence-Based Medicine
Consultant in Medicine Editor, Evidence-Based Medicine
Nuffield Department of Medicine, University of Oxford
Level 5, John Radcliffe Hospital, Oxford OX3 9DU, England
Phone: +44-(0)1865-221320 Fax: +44-(0)1865 222901
Email: [log in to unmask] WWW: http://cebm.jr2.ox.ac.uk
............................................................................
On Fri, 15 Jan 1999, Peter Griffiths wrote:
> Jenny Keating Wrote:
>
> Can you recommend a convincing discussion for those who refuse to enter a
> debate on alpha level?
>
>
>
> Jenny
>
> I'd recommend Satisitical inference: A commentary for the social and behavioural
> sciences by M Oakes (1986) Used to be published by John Wiley but I have a
> feeling it may now be with another publisher. It contains an excellent outline
> of the various schools of statisitical inference and a critique of some of the
> received wisdoms of the frequentist approach, including the arbritary selection
> of levels of alpha. All sounds very technical and it is at times but it is
> essentially a very readable book and the first few chapters should (in my humble
> opinion) be required reading for anyone who ever quotes (or reads) a p value!
>
> Peter
>
> --
> Peter Griffiths
> Lecturer, Research in Nursing Studies Section
> Florence Nightingale Division of Nursing and Midwifery
> King's College London
> Waterloo Rd
> London, UK
> SE1 8WA
>
> +44 171 872 3012 (DDI)
> +44 171 872 3219 (Fax)
> [log in to unmask]
>
>
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|