STATISTICS IN MEDICINE

Statist. Med. 2001; 20:2209–2217

BOOK REVIEWS
Editor: Niels Keiding
1. J. Mark Elwood; Critical Appraisal of Epidemiological Studies and Clinical Trials. 2nd edn.
2. Christopher Jennison and Bruce W. Turnbull; Group Sequential Methods with Applications to
Clinical Trials.
3. A. C. Cullen and H. C. Frey; Probabilistic Techniques in Exposure Assessment. A Handbook for
Dealing with Variability and Uncertainty in Models and Inputs.
4. Joseph L. Gastwirth (ed.); Statistical Science in the Courtroom.
5. P. M. Fayers and D. Machin; Quality of Life. Assessment, Analysis and Interpretation.
6. David Neumann and Carole Kimmel; Human Variability in Response to Chemical Exposures.

1. CRITICAL APPRAISAL OF EPIDEMIOLOGICAL STUDIES
AND CLINICAL TRIALS. 2nd edn. J. Mark Elwood, Ox-
ford University Press, Oxford, 1998. No of pages:
448. Price: $29:50. ISBN 0-19-262745-7
This book oBers the basic tools that are needed for
critical appraisal of studies providing the grounds
for evidence-based clinical medicine and public
health. Although there are a number of good books
available that have the same aim and scope, this
particular one has several features which make
it very useful for anybody with no training, or
with only rudimentary training, in that kind of epi-
demiological and clinical research. I suggest that
it might be a suitable textbook in courses for ei-
ther medical students during the later part of their
curriculum or for graduate students entering clin-
ical or epidemiological research programmes. Of
course, it may be useful in any continuing medi-
cal education programme for doctors who have not
been engaged in this kind of research before.
The Crst feature is that the author persistently
builds up the argument that the essential core of
the research is its ability to produce evidence for
or against causal relationships. Thereby, it helps
the reader to explicitly recognize that it makes no
sense to recommend or take any action aiming at
improving health of the patients or of the people
unless there is convincing evidence that the action,
either as a cause in itself or as an interference with
the causes of ill health, does increase the likelihood
of achieving improvement. I think the way the ar-
guments are put together should make it easy for
the reader to understand that it is hard to conceive
of any other ways to provide the evidence than by
the kind of research presented in the book.
The second feature is that the book makes no
distinction between, on one hand, clinical research
aiming at assessing therapeutic actions and, on
the other hand, epidemiological research aiming
at assessing either preventive actions or just pos-
sibly modiCable causes of diseases. This is a wise
strategy. There is a prevailing tendency among
clinicians to consider public-health oriented
epidemiology as weak, being based only on
‘correlations’, as compared to both observational
or experimental clinical research. However, often
they do not realize the similarities in the basic
tools and conditions of research or the obvious
diBerences in opportunities that, despite the simi-
larities, make it meaningless to compare the types
of research in terms of strength. On the contrary,
among public health workers there is a prevailing
temptation to rely upon less sharp evidence than
can possibly be obtained by rigorous application
of the tools presented in the book.
A third feature is the structure and the writing
of the book, which indicates great experience in
teaching in this Celd. The chapters are well deCned
and in a logical sequence. Every important point is
clearly explained and illustrated by easily readable
exhibits (tables or Cgures), and well chosen exam-
ples from either clinical medicine or public health.

Copyright ? 2001 John Wiley & Sons, Ltd.

2210 BOOK REVIEWS
There is at the end of each chapter a series of appro-
priate self-test questions with the answers placed
at the end of the book. There is so little mathe-
matical statistics that the average medical student
should not be scared. On the other hand, the sta-
tistical methods and analysis as presented are set
up in such a way that the students should have no
diIculty in following and repeating the calcula-
tions and even understanding the basic principles
of the methods, of which those based upon or de-
rived from the Mantel–Haenszel test are chosen as
the central ones. The statistical formulae, worked
examples and tables are put together in appendices.
The last six chapters of the book present a thor-
ough and detailed critical appraisal – applying all
the tools just put forward and excellently summa-
rized in Chapter 9 entitled ‘The diagnosis of cau-
sation’ – of six papers on important topics and
published in highly esteemed journals.
I have found very little to criticize. However,
a few points follow, which should not be seen
as a reservation of my initial recommendation of
the book. There is a tendency that every writer
of an epidemiological textbook selects and even
creates their own terminology, which continues to
confuse the communication in the Celd. The idea
that strength of an association, measured for ex-
ample in relative risk, may reJect nothing but the
relative frequency of the component causal fac-
tors in the population is neglected. I miss some re-
Jections about how confounding factors, which by
deCnition are part of the causal web of the outcome,
may actually become associated with the putative
causal factor, the eBect of which they confound.
In view of the role of time-to-event or counting
process analysis is modern epidemiology and clin-
ical research, there is rather little on these methods
2. GROUP SEQUENTIAL METHODS WITH APPLICATIONS
TO CLINICAL TRIALS. Christopher Jennison and
Bruce W. Turnbull, CRC=Chapman & Hall, U.K.,
2000. No. of pages: xviii + 390. Price: $ 39:00.
ISBN 0-849-30316-8
The use of group sequential methods in random-
ized trials is the largest change in the design and
analysis of clinical trials in the last 15 years.
Even though the major properties of the sequential
probability ratio test were worked out by Abra-
ham Wald and others more than 50 years ago,
the particular demands of clinical trials, espe-
cially multi-centre trials, made it impractical to use
Copyright ? 2001 John Wiley & Sons, Ltd.
here, most of it being put into an appendix. It is a
bit confusing, although of course not wrong, that
stratiCed sampling or frequency matching is con-
sidered equivalent to serial restricted sampling (p.
126), and I am not sure that the recommendation
to disregard the stratiCcation in analysis of such
data is appropriate (p. 146). In presenting the as-
sumptions of the simple linear and additive multi-
variate model for a continuous dependent variable,
it is wrongly stated that the usual methods of cal-
culation make the assumption that this variable as
such (rather than its residuals) has a normal dis-
tribution (p. 151). Maybe the explanation of the
conCdence limits – ‘we can be 95 per cent conC-
dent that these limits will include the true value’
(p. 174) – can be debated dependent on what is
meant by ‘conCdent’?
Many of my colleagues, also among the younger
generation, will strongly object to the chosen po-
sition of ‘basic physiological principles’ alongside
anecdotal experience in the bottom of the hierachy
of evidence relevant to human health studies (p.
229)! The expected opposition may be more will-
ing to accept the arguments if this position had
been justiCed rather than just taken. On the other
hand, seeing this diminutive role of what they be-
lieve is the ideal basis for clinical practice could
possibly act as an eye opener.
THORKILD I. A. SHRENSEN
Professor of Clinical Epidemiology
Danish Epidemiology Science Centre
Institute of Preventive Medicine
Copenhagen University Hospital
DK 1399 Copenhagen K
Denmark
classical methods for interim monitoring. Peter Ar-
mitage began adapting sequential methods to ex-
periments in human populations in the 1950s, but
the papers by Pocock [1] and O’Brien and Fleming
[2] mark the turning point when statisticians began
to think of sequential designs as more the rule than
the exception. The literature on group sequential
designs for clinical trials has grown rapidly since
1980. Jennison and Turnbull have been among the
major contributors to this literature in the last 10
years, and now they have summarized the state of
the art and science of group sequential designs in
this well written and comprehensive book.
Group sequential designs specify early stop-
ping boundaries for experiments in which
Statist. Med. 2001; 20:2209–2217