For those interested in the 'Use of Bayesian Statistics in Consultancy' exchange, this completes the summary. My original email appears last, below.
---------------------------------------------------------------------------------------
Martin,
There is a lot truth in what you say. I myself and others in my Unit spend a
certain amount of time in a consultancy role. I think movement is needed on
both sides of the divide. I am a little unusual in that I have trained (and
practiced) both as a medi and a statistician. Some statisticians in the past
(we all know examples) tended to patronise or bullshit medics - often
because they had little skill. Medical statistics teaching for medics in the
past has been of an appalling standard -what an impression they were
gven!ad. On the other hand some medics are arrogant ignoramus's with little
regard for what thet perceive as mere "technicalities". Moreover, there has
always been an overiding pressure for young clinicans to publish large
amounts of rubbish - never mid the quality feel the weight! Statisticians
are sometimes seen as just the facilitators for this. If overloaded they
may take the easy option.
In my view this is changing.
1. In my own field of Public Health there is real interest in improving
statistical understanding and we are creating more suitable posts for
statisticians and other professional scientists in Public Health. This was
not the case when I took it up 20 years ago. Environmental concerns of teh
Sellafield type occupy a lot of time since 1985 because the Public are
baffled by the statistical concepts and fiollwoing BSE they do not believe
officialdom.
2. One of the drivers for change in the clincial arena has been the
increased status of "Evidence" and the EBM initiatives of Sackett et al and
in the cancer field the creation of multidiscliplinary teams, agreed
guidelines annd minimum standards during the 1990s
3. Another driver is the focus on clinical audit following the Bristol
enquiry which emerged in 1995 - there is a need now for systematic
structured comparative audit across the UK - not just the navel
comtemplating audits in a single hospital which occured in the past. What
with Shipman as well - well he just put the tin lid on it.
4. In terms of my clinician collaborators and colleagues, I do think people
are phased by Bayesian methods, in fact thet are very impressed by sort of
mult-level analysis presented in the statistical appendices to the Bristol
report. They positively demand modern methods and the latest software.
Clinicians are not stupid but previously statistics did not seem so relevant
- a study design and a P value was all required.
5. I have seen the techniques and software for Bayesian methods emerge
during the 1980s and 1990s - they get easier and more user friendly all the
time. When I started we had to write our own Fortran code.
6. A recent article in Stats Med. took a forward look at medical statistics
and emphasised the need statisticians to get a better grasp of medical
issues. To work successfully with clinicians you need to understand their
issues and speak their language.
John Steward
________________________________________
Dr J A Steward MBBCh BA MSc PhD FFPHM
Director
Welsh Cancer Intelligence & Surveillance Unit
14 Cathedral Road
Cardiff CF1 9LJ
_______________________________________
tel:- 029 20 373500
fax:- 029 20 373511
_______________________________________
---------------------------------------------------------------------------------------
Dear Martin
I would imagine that the contents of your email struck a chord with most
medical statisticians, or at least I hope that this is the case.
Like you, I'm involved in daily consultations with medics and,
occasionally, with nurses and PAMS as well.
I would like to add a specific item to your list, and that concerns the
involvement of statisticians in the whole clinical governance thing. I
realise that this may not be the most exciting arena for many
statisticians, but given the relative importance of research and governance
to most trusts, surely it's time we got properly involved with it, and at
the right level? Fortunately, our Trust has taken heed of my frequent
nagging and I am now involved with the Director of Performance and Quality
in trying to make some sense of the whole audit thing. It looks like it's
going to be a very long job, but at least it's a start.
I think your idea of a separate medstats email list is a great idea. My
only concern is whether any of us would have the time to run it?
I suspect you will have received many responses and look forward to being
involved (perhaps indirectly) with a list, should one materialise.
Tony Hildreth
Trust Medical Statistician
City Hospitals Sunderland
---------------------------------------------------------------------------------------
---------------------------------------------------------------------------------------
Dear Allstatters,
I am aware that this debate closed some time ago, and hope that you will not find this email to be too self-seeking. What follows was not included in the SUMMARY posted, yet I feel it addresses the question in an even-handed way, unlike most correspondence I have seen on the subject.
"In my short experience consulting with personnel ranging from Nurses to
Professors, I find that most are either in awe of 'simple' classical
statistics (unwilling to accept fundamental assumptions such as i.i.d.), or
often positively against such statistics.
I must say, I empathise with this view point (for example, we are all
related, to an extent with each other, be it by blood, location,
time........, and so a random sample is not likely to be i.i.d.) Successful
consultancy is achieved by a compromise between the medic and the
statistician, if both are honest over their fundamental assumptions. In this
sense, classical statistics DOES make use of prior knowledge, and this prior
knowledge comes into play mostly AFTER the study is done. Most people are
more comfortable with this as 'the scientific method', but it does bias
towards self-fulfilling prophecy. Here, however, just how much of an effect
prior knowledge has had remains rather intangible.
Bayesian statistics would, I feel, be a bridge too far for my customers.
While it allows one to express, quantitatively, the extent of the
involvement of prior knowledge, it is not seen to meet the ideal of the
scientific method, and, moreover, compounds the believed lack of adherence
to the fundamental statistical assumptions (e.g. i.i.d.) being made.
Classical statistics is what most know, and most often published. As
research must be published, and is most often undertaken principally with
that objective in mind, I feel that my customers might allow me to adopt
Bayesian Statistics for a more academic, statistical paper that I wished to
write, but are mostly after simple straightforward statistics that they feel
(going to the lowest common denominator) that their colleagues will
understand or wish to read.
I do not feel that even just Classical Statistics will move forward further
into the Medical Statistics in Practice arena until common ground is found
between Statistics and Medicine in terms of their fundamental assumptions.
And this needs/must be driven by the journals themselves. I am interested in
setting up a separate MEDICAL statistics email list to address these
concerns (which are more unique to Medical Statistics). What do you think ?"
I would only add the following philosophical wander:
Science is but art (at the level of the individual) that exists with boundaries agreed upon at the collective level, 'invented' by those chosen by the 'collective' as experts; statistics allows one to test out how individuals might match up to these collective models. Classical statistics has the problem that the method of chosing a model's boundaries is somewhat remote from every day reality; Bayesian statistics has the problem of the boundaries being set depending on the strength of experts existing boundaries (could Einstein have overcome Newton in this way ?).
Best Wishes,
Martin Holt
Medical Statistician
Southern Derbyshire Acute Hospitals Trust
|
