Thank you very much to everyone who answered to my email, you have all been very helpful.
For those who were wondering, I am working on a cohort study based in Plymouth of 300 children who are measured on metabolic and anthropometric variables in relation to the metabolic syndrome (EarlyBird Study). Following the advice given by allstaters, targeted literature searches and meetings with other members of the team we have decided to attempt both analyses on our data.
I would be glad to discuss issues related to the application of Factor and Cluster Analysis with anyone who is interested
My original email was:
>Hello,
>I am currently researching on the applications of multivariate statistics techniques in medical research. I have reasons to beleveive >that both cluster analysis and factor analysis suit our data, but they don't seem to be very popular in medical settings... Any
>particular reason I am not aware of? Any references - papers you can suggest?
>Thanks for any help
>Regards
>Sandra
Please find a summary of the responses below:
From:
Stephen Senn [[log in to unmask]]
Stephen Senn
Professor of Statistics
Department of Statistics
15 University Gardens
<http://www.gla.ac.uk>University of Glasgow
G12 8QQ
Tel: +44 (0)141 330 5141
Fax: +44(0)141 330 4814
email [log in to unmask]
Dear Sandra,
Multivariate analysis is a means of finding the answer when you don't know
the question. It rarely produces anything of particular use as regards
practical decisions although it can produce interesting patterns. Hence its
low rate of us in medical contexts.
One possible practical application, however, is in classification problems.
So that the cluster analysis and discriminant analysis area is a reasonable
one to look at.
Here are some possible papers and books reviewed in Statistics in Medicine
which had multivariate in the title.
Multivariate meta-analysis
Statistics in Medicine
Volume 22, Issue 14, Date: 30 July 2003, Pages: 2309-2333
In-Sun Nam, Kerrie Mengersen, Paul Garthwaite
Abstract | References | Full Text: PDF (186K)
select this item for viewing Multivariate survival analysis with
doubly-censored data: application to the assessment of Accutane treatment
for fibrodysplasia ossificans progressiva
Statistics in Medicine
Volume 21, Issue 17, Date: 15 September 2002, Pages: 2547-2562
Geoffrey Jones, David M. Rocke
Abstract | References | Full Text: PDF (144K)
select this item for viewing Analysis of Multivariate Survival Data. Philip
Hougaard, Springer, New York, 2000. No. of pages: xvii+542. Price: $84.95.
ISBN 0-387-98873-4
Statistics in Medicine
Volume 20, Issue 16, Date: 30 August 2001, Pages: 2533-2534
Rob Henderson
Abstract | Full Text: PDF (29K)
select this item for viewing Multivariate Taxometric Procedures. Niels G.
Waller and Paul E. Meehl, Sage Publications, U.S.A., 1998. No. of pages:
150. ISBN 0-7619-0257-0
Statistics in Medicine
Volume 19, Issue 19, Date: 15 October 2000, Pages: 2713-2714
Tiberiu Postelnicu
Abstract | References | Full Text: PDF (48K)
select this item for viewing Multivariate spatial models for event data
Statistics in Medicine
Volume 19, Issue 17-18, Date: 15 - 30 September 2000, Pages: 2469-2478
Alastair H. Leyland, Ian H. Langford, Jon Rasbash, Harvey Goldstein
Abstract | References | Full Text: PDF (106K)
select this item for viewing Analysis of Incomplete Multivariate Data. J.
L. Schafer, Chapman & Hall, London, 1997. No. of pages: xiv+430. Price:
£39.95. ISBN 0-412-04061-1
Statistics in Medicine
Volume 19, Issue 7, Date: 15 April 2000, Pages: 1006-1008
Leonhard Knorr-Held
Abstract | References
select this item for viewing Multivariate Dependencies. Models, Analysis
and Interpretation. D. R. Cox and N. Wermuth, Chapman & Hall, London, 1996,
No. of pages: 272. Price: £39. ISBN 0-412-75410-X
Statistics in Medicine
Volume 19, Issue 2, Date: 30 January 2000, Pages: 277-278
Morten Frydenberg
Abstract | References
select this item for viewing Multivariate linear mixed models for multiple
outcomes
Statistics in Medicine
Volume 18, Issue 17-18, Date: 15 - 30 September 1999, Pages: 2479-2492
Mary Sammel, Xihong Lin, Louise Ryan
Abstract | References
select this item for viewing Multivariate outlier detection applied to
multiply imputed laboratory data
Statistics in Medicine
Volume 18, Issue 14, Date: 30 July 1999, Pages: 1879-1895
Kay I. Penny, Ian T. Jolliffe
Abstract | References
select this item for viewing Multivariate competing risks
Statistics in Medicine
Volume 18, Issue 9, Date: 15 May 1999, Pages: 1023-1030
Jan Wohlfahrt, Per Kragh Andersen, Mads Melbye
Abstract | References
select this item for viewing Multivariate non-parametric methods for
Mann-Whitney statistics to analyse cross-over studies with two treatment
sequences
Statistics in Medicine
Volume 18, Issue 8, Date: 30 April 1999, Pages: 989-1017
Jin-Whan Jung, Gary G. Koch
Abstract | References
select this item for viewing Computer-Aided Multivariate Analysis. Third
edition. A. A. Afifi & V. Clark, Chapman & Hall, U.K. 1996 (reprinted
1998). No. of pages: xxi+455. Price: £45. ISBN 0-4127-3060-X
Statistics in Medicine
Volume 18, Issue 5, Date: 15 March 1999, Pages: 624-625
Ene-Margit Tiit
Abstract | References
select this item for viewing Book review: Multivariate Analysis Techniques
in Social Science Research. Jacques Tacq, Sage Publications, London, 1997.
No. of pages: 411. Price: $29.95 (paperback). ISBN 0-7619-5273-X. Price:
$79.95 (hardcover), ISBN 0-7619-5272-1
Statistics in Medicine
Volume 17, Issue 20, Date: 30 October 1998, Pages: 2418-2419
Jakob Bue Bjorner
Abstract | References
select this item for viewing Book review: Multivariate Models and
Dependence Concepts. Harry Joe, Chapman & Hall, 1997. No. of pages:
xiii+399. Price: £39. ISBN 412073 315
Statistics in Medicine
Volume 17, Issue 18, Date: 30 September 1998, Pages: 2154-2155
Steen Arne Andersson
Abstract | References
select this item for viewing BOOK REVIEW: Recent Advances in Descriptive
Multivariate Analysis. Wojtek J. Krzanowski, Oxford University Press, 1995.
No. of pages: 384. Price: £35. ISBN: 0-19-852285-1
Statistics in Medicine
Volume 16, Issue 15, Date: 15 August 1997, Pages: 1787-1788
Annibale Biggeri
Regards
Stephen
..............................................................................................................
From:
Allan Reese [[log in to unmask]]
Hard to know where to start. One point might be that you seem to
imply cluster and factor are somehow alternatives. They address
different problems. Factor analysis works on the correlation matrix
and makes assumptions that the sample are homogeneous. I have often
commented on the need to check that assumption, and cluster analysis
is a useful tool to look for un-anticipated segmentation. I think if
you do a keyword search you will find plenty of applications.
Allan
.....................................................................................................
From:
Carole A Cull [[log in to unmask]]
Dr Carole A. Cull
University Research Lecturer
Diabetes Trials Unit
Oxford Centre for Diabetes, Endocrinology and Metabolism
Churchill Hospital
Headington
Oxford
OX3 7LJ
tel +44 (0) 1865 857251
fax +44 (0) 1865 857254
Dear Sandra
I don't have an answer, but part of it may be that most of the PCP and
Factor analysis methods have been developed in psychometry or social
sciences and carry a sort of stigma from that - psychometrics is
sometimes seen as 'less precise' than 'real medical statistics'. Silly,
but I've heard that said. Don't quote me!
I have myself begun to develop an analysis of a large and complex
dataset using PCP to identify groupings of related variables in relation
to outcome - but it's mostly just at the idea stage. No publications
yet. However, there is one bit of medical research which originally
relied upon PCP - have you heard of Syndrome X, also known as the
metabolic syndrome, or the insulin resistance syndrome? The original
paper was by Gerry Reaven back in the late 1980s - I think this is the
one - if not it'll give you the starting point for a chase
Reaven GM. Role of insulin resistance in human disease. Diabetes
1988;37:1595-1607
in which he uses PCP to identify a group of metabolic variables which
predispose to heart disease. Sadly since then most of the work in this
area has relied on less elegant, and possibly not as reliable, ways of
looking at the problem.
HTH
Carole
.......................................................................................................
From:
Matz David [[log in to unmask]]
Sandra
I suggest you address this to the Medical statistics section of the RSS who
will have a wealth of expertise
contact is [log in to unmask]
the only comment I have is that in general multivariate techniques are
likely to be less used than more basic methods simply because the underlying
assumptions and mathematics may be harder to understand than simpler methods
[initial data analysis, basic regression etc etc] and are consequently
taught in more advanced textbooks/courses.
Equally some specific methods have been developed specifically as a result
of medical statistics [re sequential trials etc] - indeed there is [or used
to be] a medical statistics MSc run at Souhamption University.
regards
David
[writing in a purely personal capacity]
..................................................................................................................
From:
Denis Shields [[log in to unmask]]
There is a lot of gene expression analysis ("microarrays")
applied to tumour cells taken from patients
that has used both cluster and principal component
techniques.
In fact there is a vast literature on this
mostly in the last 3 years.
Denis.
....................................................................................................
From:
SAM Ayis [[log in to unmask]]
Hi sandra
I suggest two references for factor analysis
1- C. Chatfield and A. J. Colins " introduction to Multivariate Analysis"
Chapter 5.
2-Jae-On Kim & W. Muellerl "Factor Analysis, Statistical Methods and
practical Issues" Series: Quantitative applications in social sciences, A
Sage University paper.
good luck
Salma
........................................................................................................
From:
Andrew Zelin [[log in to unmask]]
Andrew Zelin
Head of Sampling and Statistics
MORI
UK
Sandra,
Your note interested me as I run a lot of Factor and
Cluster analysis, which prove popular in Market
Research to simplify batteries of attitudinal
questions and group respondents into like-minded
typologies, respectively. There is no reason why
these techniques cannot be applied to any sphere,
including Medical Research. However, amongst the
spectrum of statistical methodologies, the way we use
FA and ClA constitutes one of the "softer", more
descriptive techniques, as a useful way to simplify
and understand the data, drawing out the main themes /
trends.
As I understand Medical Satistics (and this might be a
great generalisation!), it may be more about setting
"rules" to define and classify individuals or
situations - ie Does this treatment work or not? Does
this person have a disease or not? With what
likelihood? - or determining measures (with levels of
certainty / sigificance - What is the relative risk
of X on Y?
going back to first principles, I would tend to think
of the objective of the study or assessment first and
work backwards from that. Which type of analysis
would enable you to answer the overall research
question(s)? Although they way in which MR uses FA
and CLA might not always be transferable in a totally
un-modified way to medical stats, there would
certainly be situations where it can be used -
anything Psychological / Attitudinal (eg attitudes of
Doctors to treatments) would be of relevance. It
might also possibly be of use in attempting to look at
patterns of a set of less-well understood disease
patterns or syndromes, in an attempy to classify and
hence better understand them. Some typologies of a
disease process might respond better to some forms of
treatment than others.
I hope this helps and happy to discuss further.
Andrew
|
