| |
Epi Wit & Wisdom Resources
Two Part Review of the
2nd Edition of Modern Epidemiology by Rothman and Greenland, Part I
By Robert
McKeown, PhD
[Editor’s
note: Because of the anticipated interest in the second edition of
Modern Epidemiology, the Epi Monitor is publishing a lengthier than
usual review. The first installment appears this month and a second
part which compares the new textbook to other existing texts will
appear in the March or April issue. The review was written by Robert
McKeown who has been teaching introductory and advanced-level
epidemiology courses for both epidemiology majors and non-majors at
the University of South Carolina for eight years. He has had a keen
interest in epidemiology textbooks and has searched extensively for
“books that work” in teaching epidemiology at all levels and to
diverse audiences. Dr. McKeown is a “second career” epidemiologist and
holds a Ph.D in both theology and epidemiology. Besides his
appointment in the Department of Epidemiology and Biostatistics at the
University of South Carolina, he is Research Director of the
University’s Center for Bioethics.]
One would be hard pressed to
think of another book in epidemiology that has been as widely and
eagerly awaited as this second edition of Modern Epidemiology.
Kenneth Rothman has now been joined by Sander
Greenland to give us another indispensable text which is sure
to challenge, enlighten and perhaps, provoke us.
This is no minor revision. It is
much enlarged in size (more than twice the length of the previous
version), scope and contributors. The first three parts are Rothman
and Greenland’s expansion and update of the first edition, with a
chapter on “Field Methods in Epidemiology” by Patricia Hartge
and Jack Cahill. The latter is a practical, how to
guide to conducting epidemiologic studies, from designing data forms
and subject selection to data collection and management. The last part
of the book is a collection of chapters on special topics by various
contributors, all prominent in their respective areas.
Reading this text is like a
roller coaster ride, though one in which the climbs are more thrilling
than the descents. There are some sections which are striking for
their novel approach and clarity. Other sections, however, left me
perplexed or disappointed, wondering why the authors felt compelled to
approach an issue in this way--or to this extent (witness the long
section on the various uses and abuses of the p-value). On further
reading, however, the more appropriate metaphor is a drive in the
mountains: with each new switchback, one is circling around for
another look at a topic from a different perspective. Sometimes this
places material in more appropriate contexts and enhances
understanding by building on previous treatments. Other times it is
simply frustrating. For example, in chapter four causal effect
measures are described as counterfactual. However, it is not until
chapter nine that we get a clearer definition and illustration of what
that means. Gone is the excellent short chapter on standardization of
rates, replaced by a few paragraphs–-seeming like an afterthought at
the end of an early chapter–-and then “context sensitive” discussions
of standardization interspersed throughout the text. In the end, this
approach may be better, but one misses the excellent introduction to
the concept, not to mention the discussion of direct versus indirect
methods, terms no longer used in this current issue.
The book is strong where those
acquainted with the first edition would expect it to be: in the
explanation of methodological and analysis issues. The heart of this
text, for me, has always been the discussion of study design and
conduct and the introduction to statistical analysis, particularly the
treatment of bias, confounding, and effect modification. The second
edition does not disappoint in those matters. If anything, the
expanded treatment is, for the most part, even better. The treatment
of case-control studies is not only an essential guide to
understanding, designing, implementing, and analyzing these studies,
but makes an important contribution to their vindication in the
scientific community. As with any revision, one finds that favorite
passages have been rewritten and must be given up like a comfortable
old shoe. One hopes the reformulation will, in time, also come to fit
and perhaps provide even better support.
The litmus test of a student
text book for me is the treatment of rates and risks and the
explanation of confounding and effect modification. The discussion of
the former is, as one would expect, rigorous, though not always lucid,
and less advanced students will still be asking what these things are
and how they’re different. The initial discussion of confounding, in
the context of a discussion of the distinction between measures of
effect and association, on the other hand, is likely to leave even
seasoned students accustomed to the classic definitions, wondering.
Never fear, the old example of pizza and beer still appears later on
in a much expanded discussion of confounding. As in the first
edition, the fuller explanations of confounding versus effect
modification, both treated in more than one section, are models of
clarity and insight, though some illustrative material from the first
edition has, alas, been lost. For those interested in using this as a
student text (as I have decided to do), one must be prepared for a
text intensive volume, with very few graphics, little white space, and
no problems or study questions to guide the student. This edition does
add a brief outline at the beginning of each chapter.
The addition of Greenland as an
author has certainly shaped both content and presentation. One of the
most prolific of the theoretical/methodological epidemiologists (there
are some 70 of his sole or first author publications in the references
at the end of the book), he shares the methodological orientation of
the original author and the concern for causal inference (witness the
section on that topic in his collection of seminal papers, The
Evolution of Epidemiologic Ideas, 1987). The result is a marked
expansion of chapters dealing with data analysis. Some of this
material appeared in the first edition, but it has been extensively
reworked and much new material added. Indeed, part III might serve
well as a text in epidemiologic data analysis in its own right. In
this part of the book some sections are marked with an asterisk,
indicating, “Sections with more advanced content that may be skipped
on first reading.”
In keeping with this
methodological orientation, there is scant attention to “pre-modern”
epidemiology, particularly the history of public health and the
intimate relation of epidemiology to the public health movement. But
where other texts might dwell on Snow and Farr, et al. and give
cursory attention to causal inference, Rothman and Greenland give
attention to Bayesian inference and Karl Popper and
the theoretical basis for making causal claims, even returning to
causal inference in the chapter on “Concepts of Interaction.” (Snow
does get his treatment, but in the overview of study designs.) Still
present is the helpful discussion of sufficient causality and causal
complements (using the familiar causal “pies”) and the explanation of
the impact of varying prevalence of causal complements across
populations on measures of effect. One might quibble over what is
included, and the treatment of Popper and current debate on causal
inference may not be sufficiently critical to satisfy the
philosophically sophisticated (though Rothman and Greenland are not
ones to mince words when they have an opinion), this extensive
treatment is a welcome addition, which complements the current
available literature. The presentation should challenge
epidemiologists once again to take more seriously the philosophical
underpinnings of what we do.
Sometimes the authors may appear
to be making overly fine distinctions (even to this reader who once
minored in medieval philosophy!), and introduce new terms instead of
staking a claim to a clarified definition of older ones. (Perhaps it
is time for an expert panel on nomenclature to define our terms
consistently.) One does wonder, however, why other distinctions are
glossed over, notably the distinction between disease and illness, a
subject of considerable discussion in the philosophical literature.
Interestingly, there is no mention of health, and little discussion of
public health in the authors’ chapters, though the excellent
discussion of “public health interactions” as departures from
additivity remains, and public health issues do arise in the special
topics chapters of part IV. Surely one of the many characteristics of
modern epidemiology has been its broadening scope, moving beyond
traditional concepts of infectious and chronic disease, not only to
other adverse outcomes, but also to measures of health. There is scant
attention to the role of psycho-social, cultural and economic factors
as determinants of health and disease in populations. Surely the
assessment of these dimensions is of concern to epidemiologists. This
is related to the current debate over “upstream” versus “downstream”
epidemiology, which was conspicuous by its absence in a text known for
taking on disputed topics. To be sure, the text represents a
compilation of the most significant developments in methods and
analysis which we would ignore at our peril. However, there still
remains the need to reconcile epidemiology’s desire as a discipline to
achieve precision and clarity in the use of terms and definitions with
what has been its genius historically to reach beyond its borders and
work across disciplines.
Given the title of the book and
Rothman’s well-known concern for the ethics of placebo-controlled
randomized trials (which is mentioned in that chapter), the absence of
a section on ethical issues in modern epidemiology is a puzzling
omission. Also puzzling, though perhaps less surprising, is the
absence of a chapter on some of the most important extensions of
epidemiologic methods in program evaluation and health services
research, matters which receive some scant attention in the special
topics chapters, e.g. on surveillance and screening.
Part IV, Special Topics,
constitutes about one third of the text. Again the approach is
methodological rather outcome oriented. So, for instance, there is no
chapter on cancer or CVD epidemiology--neither the descriptive
epidemiology of these outcomes nor the particular methodological
issues pertinent to their study though there are chapters on
infectious disease epidemiology and reproductive epidemiology. That
being the case, it is no surprise that there are no chapters dealing
with the outcomes beyond the traditional infectious and chronic
diseases: e.g. mental disorders, violence, quality of life,
pharmacoepidemiology. What is there, however, is a real treat. A
chapter on surveillance by James W. Buehler is an
excellent introduction to a topic often overlooked or treated
superficially in other texts. Hal Morgenstern’s
chapter on ecologic studies also gives greater attention and rigor to
a study design now being revived, perhaps in light of new techniques
and sources of information, and perhaps in recognition of the
important role of larger “ecological” or “environmental” factors. One
does wonder why this chapter should appear in special topics rather
than in the section with other study designs. A chapter on “Analysis
of Vital Statistics Data” (S. H. Moolgavkar, J. A. H. Lee, R. G.
Stevens) is a useful guide with sections on cohort effect and helpful
examples. Alan Morrison’s chapter on “Screening”
provides both the introductory measures and more advanced
applications. Noel Weiss provides a (too brief)
introduction to “Clinical Epidemiology” which repeats some of the
material in other chapters. Elizabeth Halloran provides the chapter on
“Concepts of Infectious Disease Epidemiology,” which, like the other
chapters, is oriented (quite helpfully) toward methodological issues,
not descriptive epidemiology. The “Environmental Epidemiology” chapter
by Irva Hertz-Piciotto also provides clear
introductory material concerning exposure assessment (including
biomarkers) and a range of analysis approaches. Clarice
Weinberg and Allen Wilcox furnish the
chapter on “Reproductive Epidemiology,” again focusing on
methodological issues rather than more specific descriptive
epidemiology, and giving important guidance to the student or
researcher planning a study of reproductive outcomes. Muin
Khoury contributed an introduction to “Genetic Epidemiology”,
for which those new to the field will be grateful. Walter
Willett’s chapter on “Nutritional Epidemiology” is an
excellent introduction and overview of methods and issues, though it
lacks an acknowledgment of the difficulties of applying current
research methods and instruments in cross cultural settings,
especially in minority populations. Finally, Greenland re-enters with
the last chapter in the book on “Meta-analysis,” (Is this placement a
little Greek joke by the authors?) to provide a clarifying guide to
those of us inclined to think this approach to combining and
summarizing previous studies has more to do with smoke and mirrors
than with rigorous method.
This is a book I will keep handy
and refer to often--as I did the first edition (which I still plan to
keep). And like the first edition, it will no doubt become the object
of something like a love-hate relationship for many of us who teach
and do research in epidemiology. It is a notable accomplishment by two
already accomplished epidemiologists. I look forward to the creative
dialog and debate it is bound to evoke in our discipline. Thanks to
Ken Rothman and Sander Greenland.
Published February
1998
Rothman and Greenland’s
Modern Epidemiology Compared With Other Epidemiology Textbooks, Part
II
By Robert
McKeown, PhD
Part I of this review of Rothman
and Greenland’s (R&G), Modern Epidemiology (2nd Edition, Lippincott-Raven,
1998) in the Epi Monitor dealt with the textbook on its own, with
little attention to how it compares to other textbooks for specific
course applications. This overview offers a personal perspective on
that topic. There is certainly no attempt to be exhaustive, and I
apologize to those whose texts I have overlooked. Their exclusion
should certainly not be taken as any indication of their relative
worth. Rather I mention those texts with which I have greater
familiarity.
The first question one might
pose would be the course level for which R&G would be appropriate.
It’s an intriguing question which I continue to ask. On the one hand,
it clearly tackles subject matter which we would expect to address in
a good introductory course on epidemiologic concepts and methods. On
the other hand, there is clearly much more here than one would expect
to mention, even in passing, in an introductory course, and those
matters which might be covered in a first course are presented in
greater depth than may be appropriate for beginning students. Yet, I
think we do the students and the discipline a disservice when we
introduce concepts in an overly general or “make-do” fashion,
expecting to go back later and correct the oversimplification.
I think R&G could be seen as a
book an epidemiology major might purchase early on, perhaps even in
the first course, to be used through several courses. Very judicious
selection of chapters and sections could certainly supplement an
excellent introductory text such as Leon Gordis’
Epidemiology (Saunders, 1996). The latter is one my students find
accessible and readable.
It is reliable, rather
comprehensive for such a slim volume, and full of helpful graphics and
tables, as well as thoughtful study questions at the end of each
chapter, all sparse or missing in R&G. The Gordis text also has the
advantage of including chapters, albeit brief ones, on research
ethics, health services evaluation, and application of epidemiology to
policy. In addition to the usual material on reliability and validity
of screening tests, there is a chapter on evaluation of screening
programs including a lucid discussion of biases, such as referral and
lead time bias. As noted in part I of this review, there is scant
attention to health services evaluation in R&G, and it is confined to
specific sections of the special topics chapters. An interesting
exercise, though perhaps not one for an introductory class, might be
to combine a text like Gordis’ with portions of R&G, supplemented by
John Last’s A Dictionary of Epidemiology (3rd
Edition, Oxford, 1995), in order to stimulate discussion of the
ferment in the field concerning basic concepts, definitions and
approaches.
Another first semester text with
which readers are likely to be familiar is the widely used Hennekens
and Buring, Epidemiology in Medicine, (Little, Brown & Co., 1987).
Though there are some sections which reflect the public health
applications of epidemiology, the title of the book is an indication
of its orientation. Like R&G, there is a section on analysis, but it
is rudimentary, and certainly differs in approach. Hennekens and
Buring also include appendices with statistical tables and charts for
choosing the appropriate statistical test. (Let the reader make
his/her own judgment about the value of that.) Like R&G, Hennekens and
Buring is text heavy, but it is clear and readable and does contain
more tables and figures, as well as study questions. It has served
well for over a decade but shows its age, and certainly lacks the
scope and depth of R&G. Further, it lacks the orientation to public
health and application which can be found in Gordis and other texts.
The third edition of Foundations
of Epidemiology by David Lilienfeld and Paul
Stolley (Oxford, 1994) was plagued by errors, typographical
and otherwise, in early printings, now corrected. Some of the
material, in spite of extensive updating, is dated. The authors have
added a section on clinical (rather than public health) applications,
an odd choice given the history of this text. It does have the
advantage of providing helpful problems for students at the end of
chapters, but it is not as methodologically sophisticated as R&G.
Lilienfeld and Stolley also include an appendix on “Selected
Statistical Procedures” which seems too little for the beginning
student and not enough for the more advanced student. However, while
R&G’s ignore the history and development of epidemiology, Lilienfeld
and Stolley provide a helpful overview for the uninitiated.
In contrast to R&G, several
texts on the market are aimed at introducing non-epidemiologists to
our field. Friis and Sellers’ Epidemiology for Public Health Practice
(Aspen, 1996) is a notable example of a collaboration between an
epidemiologist (Sellers) and a non–epidemiologist (Friis). The book is
filled with examples and graphics and applications geared toward
public health trainees and practitioners in other disciplines, and
each chapter contains objectives, an outline, a summary conclusion,
and study questions. While the audience and approach may result in a
certain lack of depth and rigor, the book could be useful for
introductory courses for non-epidemiology students, something R&G is
clearly not designed for. (I still prefer Gordis for this group,
though Brownson and Petitti, mentioned below, is attractive for health
administration students.)
Other texts in the same
category, and sharing some of the characteristics of Friis and
Sellers, are Timmreck, An Introduction to Epidemiology (Jones and
Bartlett, 1994) and Page, Cole, and Timmreck; Basic Epidemiological
Methods and Biostatistics: A Practical Guidebook (Jones and Bartlett,
1995), the latter being more appropriate perhaps for an undergraduate
course. Texts more oriented to the preventive medicine resident are
Knapp and Miller, Clinical Epidemiology and Biostatistics (Williams
and Wilkins, 1992); Jekel, Elmore and Katz; Epidemiology,
Biostatistics, and Preventive Medicine (Saunders, 1996); and the
companion Katz, Epidemiology, Biostatistics, and Preventive Medicine
Review (Saunders, 1997), all designed for independent study. None of
these texts would be considered in the same category as the R&G text
which is clearly directed toward the serious student of epidemiology
and the active epidemiologist or research practitioner.
For the instructor who wishes a
more content oriented introduction, perhaps following up on a basic
methods course, one could use the special topics chapters and selected
other topics from R&G for methodological issues in conjunction with a
text such as Brownson, Remington and Davis, Chronic Disease
Epidemiology and Control (APHA, 1993). The latter is to be out in a
new edition later this year and will add to its chapters on specific
outcomes and lifestyle risk factors (absent from R&G), chapters on
diet and nutrition and physical activity, the latter also missing from
R&G.
Another approach might be a
slightly more advanced course which covers study design and
implementation and some analysis. My choice in this area is the new
edition of Kelsey, Whittemore, Evans, and Thompson, Methods in
Observational Epidemiology (2nd Edition, Oxford, 1996). It is a highly
readable and practical guide, with numerous examples and exercises.
Again, combinations of this text with selected chapters of R&G would
provide the students in a second level methods course with a solid
foundation. Kelsey et al. is perhaps stronger in clearly laying out
the practical issues in study implementation, including measurement
issues, while R&G is stronger in analytic and theoretical matters.
One of the more surprising
omissions from R&G is a treatment of the applications of modern
epidemiology in outcomes research and health services evaluation. In
that regard we are fortunate to have a wonderful new text from
Ross Brownson and Diana Petitti, Applied
Epidemiology: Theory to Practice (Oxford, 1998). They, with other
contributors, provide us with basic methodological chapters on key
concepts, outbreak and cluster investigation, surveillance, risk
assessment, community intervention trials and screening. But there are
also chapters on outcomes research, economic evaluation, measuring
quality of health care, communicating epidemiologic information, and
the interface of epidemiology and health policy. This text could be
supplemented by certain chapters from the R&G text, especially the
special topics, but it goes far beyond R&G in providing tools for
epidemiologists who will be working in the changing health care
environment. It also holds promise as a text for an introduction to
epidemiology with applications for non-epidemiologists, say public
health professionals or hospital administrators.
One final area of use for the
R&G text, to which I alluded in the previous review, is a course in
epidemiologic analysis. I suggested the third section of the text
could serve as the foundation for such a course (as I plan to use it).
I have used Steve Selvin’s Statistical Analysis of
Epidemiologic Data (2nd Edition, Oxford, 1996) for this course (in the
absence of R&G). Though it suffers from some unexpected errors, it
provides a helpful range of topics. The orientation, however, is
decidedly statistical rather than epidemiological, though some
explanations may be more accessible than--or at least offer an
alternative to--R&G. My colleagues have been pleased with the response
to Kleinbaum, Kupper, Muller, and Nizam, Applied Regression Analysis
and Multivariable Methods (3rd Edition, Duxbury Press, 1998), though
it is clearly more limited in the range of analysis it covers.
As I reach (and overreach) the
space allotted, I think of half a dozen major works that have gone
unmentioned: Schlesselmann, for example, or Kleinbaum, Kupper, and
Morgenstern, or Curt Meinert’s book on clinical
trials on the epidemiology side, and Fleiss or Breslow and Day on the
analysis side. But they are too familiar to require comment from me.
So, to answer the question with which I began this review: For what
courses would this book be appropriate? I can’t imagine a course in
which this would be the only text I would use, nor can I think of an
epidemiology course to which at least some part of this book would not
make a helpful and important contribution. Bottom line: this is not a
book we or our students should be without.
Published March 1998
|
|
