Should the epidemiologist be a social scientist or

© International Epidemiological Association 1999
International Journal of Epidemiology 1999;28:S1019–S1021
Printed in Great Britain
Should the epidemiologist be a social scientist
or a molecular biologist?*
Mervyn Susser
Should the epidemiologist be a social scientist or a molecular
biologist? To this question, posed for me in the symposium, the
short and unadorned answer is neither. A less simplistic answer
recognizes, quite properly, that the question assumes the underlying integrity of epidemiology as a discipline. That is, on this
assumption, one can take the question to be aimed at epidemiologists for epidemiologists, while recognizing that epidemiologists
can be of different types.
Also implied is a second more disturbing assumption, namely,
that the integrity of current risk factor epidemiology is threatened. This is the kind of epidemiology that has predominated in
the era after World War II up to the present. It evolved around
a theory of multiple causes applied to exposures as they impinged on chronic disease in aggregations of individuals.1 The
implied threat lies on two sides—one is the biological microworld and the other the social and environmental macroworld.
And that threat is, I believe, true and more than an assumption.
In response to the implications of that threat, I expand my
answer. I will now say that epidemiology, if it is to survive as a
discipline with a claim to common ground, must find room for
and encompass both social science and molecular biology.
Underlying the question I see a third assumption lurking. This
says that the demand for so broad a sweep, across the macroand the micro-worlds, is unattainable. Well, yes. I can agree that
at this time few epidemiologists succeed in working both
on the micro- and the macro-levels as well as the customary
individual level. And yet my answer is the contrary; it can and
must be done. Somehow, the successive levels of organization
that are inherent in the life of human beings—each the more
complex in the ascent from molecules to societies—must be
accounted for in our thinking.2,3
An appeal to history shows that several great epidemiological
discoveries stemmed from multilevel thinking. Some notables
on my list include, in this decade, the human herpes virus 8 in
Kaposi’s sarcoma4,5 and the breast cancer gene;6,7 before that,
the hepatitis B virus8 and then its causal role in liver cancer;9,10
in the 1960s the slow virus (now the prion) and kuru,11,12 and
later Creutzfeld-Jacob, etc.; in the early 1950s, lung cancer
and smoking—the medical student Wynder as well as Doll and
Bradford Hill went on the trail, provoked by the marked rising
trends in mortality;13,14 early in this century Ross, having
Sergievsky Professor of Epidemiology Emeritus, Columbia University in the
City of New York, USA.
* Readers should be warned that this is the fourth among papers successively
solicited for publication on topics similar to my keynote address on Choosing
a Future for Epidemiology at the Pan-American and Iberian Epidemiology
Congress of 1995 (Salvador, Brazil, April 28). Overlap is inevitable as the basic
ideas are reframed and developed. Also, in response to a reviewer’s requirement, reference to my own previous work is heavier than I would wish.
discovered the transmission of the malaria protozoa from anophelene mosquitoes, developed multilevel statistical models for
prevention;15 and in the mid-19th century, the thought and
works of the epidemiological hero John Snow (we canonized
him only well into the 20th century) swing wonderfully across
multiple levels.16 And before Snow, we have the remarkable
Peter Ludwig Panum (more of him later).17
All this shows that multilevel epidemiology is not unattainable. You may demur: are these examples not exceptional, merely
special cases and exceptions? I have a retort to such objections;
what is exceptional but important at the time of discovery or
invention, becomes the commonplace of the future. That is
to say, we can learn to be multilevel and multidimensional
epidemiologists. Moreover, we shall have to learn, if we are to
escape splitting our discipline. Epidemiology is so fundamental
to public health in practice that to break it up will be to impair
our capacity to deal fully, effectively, and thus truly with health
as an attribute of populations.
Note well that I do not say that to work at multiple levels we
must become either social scientists or molecular biologists. I do
say we must comprehend and deploy the basic premises and the
nature of the information these other disciplines yield. A split
between levels, and hence a loss that neither the public health
nor epidemiology can afford, looms before us. To reach into both
those dimensions, from our current individual level practice, will
be to bridge the potential gaps between levels of organization.
Whether we can stave off a split remains a question. The
micro-level exerts the same forceful attractions as did the germ
theory at the end of the 19th century.18 The germ theory eclipsed
the considerable power and prestige of the miasmatic epidemiology that preceded it. That particular history thus goes to some
degree against my case for bridging levels.
The complete recovery of the discipline from that eclipse was
slow. It had to wait upon the advent and full development of
our current multiple cause theory of disease and its application,
in populations, as risk factor epidemiology at the individual
level. Now molecular biology offers the same kind of illusion
as did the germ theory. It is the illusion that the unarguable
definitiveness and specificity of this extreme biological microlevel can explain everything—which is to say, everything at all
the successive and increasingly complex levels of organization
above. That was never true then and is not true now, so long as
our concern is with the dynamic disease process as it occurs
within and across populations.
The macro-level of analysis is still in the making. Its tools
are less advanced, and they will always be less precise. Yet
global communication and information, and the computers and
networks on which they depend, offer formidable arms to
macro-level analysts. These arms, once they are effectively
S1019
S1020
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
applied and suitable data bases have been built, will bring qualitative change to our epidemiological potential, just as molecular
biology has done at the micro-level. The economists and their
econometric equations already manipulate the markets of the
world. We should surely do as well or better in health matters.
Great attractions as well as potential schism thus exist on both
sides of the individual level ‘black’ box of current epidemiology.
This level has preoccupied the discipline for a half century.
Now, offshore, one can already hear the siren songs. What my
argument portends is the exhaustion of the multiple risk factor
paradigm and the necessity to replace it.
To envisage what this situation entails, and what may follow
from it, requires an understanding and acceptance of the theoretical construct of paradigms, (as initiated by Ludwik Fleck
and developed by Thomas Kuhn).19,20 No science can merely
describe nature as it exists. The limitless observations that can
be made must be selected and organized to advance understanding, prediction and control. In other words, we are required to
learn how purposefully to restrict our perceptions within a
coherent structure of ideas.21
That structure—the framework of the leading concepts—
constitutes the dominant paradigm of a given era. Such a paradigm sets unspoken limits on the legitimacy of questions, methods
and ideas. Truly anomalous findings are often overlooked,
underplayed or reinterpreted to fit. One consequence is an
inevitable tension between tradition (the established paradigm)
and innovation (the aspiring challenger). In Kuhn’s view (to
which I do not wholly subscribe), a paradigm that merits a title
as dominant requires revolution to displace it. It can only be
deposed by new theories and discoveries that constitute a
scientific revolution. In epidemiology, the archetypal cause is
the displacement of the miasma theory, dominant in the 19th
century era of undifferentiated sanitary statistics, by the germ
theory. By the end of the century, miasma, like phlogiston in
chemistry, had been pitched out, never to be heard from gain in
polite scientific circles.
In my view, however, dominant paradigms can also be
replaced by simple attrition—less dramatically, but no less completely,22 as when in the post World War II ‘chronic disease’ era,
the specific-cause germ theory no longer served for the elucidation of the causes of so-called degenerative, but still mysterious,
chronic diseases. The slow evolution of multiple cause theory
heralded the dominance of the current risk factor paradigm.
Kuhn might have replied that this theory was merely an extension of what he called ‘normal science’ and hence no revolution
was needed to displace the old paradigm. I would not agree. The
germ theory itself can be taken as a case in point. Before its
climactic acceptance, a protracted struggle extended from Jacob
Henle in 1840 to Robert Koch in 1882.23
Paradigms have consequences for the public health as great as
for epidemiological research.22 The sanitation that supposedly
dispelled miasma did a good deal for the health of populations
by limiting waterborne infectious disease. But it was a total
failure in coping with the person-to-person transmission of
many infectious diseases. For that, we needed the germ theory,
vaccines, chemotherapy and antibiotics.
Infection is back as a factor in chronic disease, but that
realization was not owed to germ theory. That work was guided
by the risk factor paradigm of multiple causes studied at individual level. What then ails risk factor epidemiology? Why move
on? After all, it has scored great triumphs—with heart disease,
cancer, and much else. It has given us the principles of a
new multivariate methodology, based on design and analytical
development. We have learned thereby how to pursue multiple
risks. We pay due attention to bias and to confounding, and
allow for the range of error in our estimates.
The limitations of the paradigm reside, first, in its steadfast
commitment to a single level of organization, the study of individuals, almost always disconnected from each other, even though
assembled in large numbers in populations. This commitment
discourages exploration of context and causal antecedents of
individual risk factors. The ‘ecologic’ macro-level studies thereby entailed have been anathema. Much effort has gone into
displaying the weaknesses of inference at this level, very little to
finding its strengths.
Second, linkages interposed between exposure and outcome
tend to be dispensed with. In so far as they are sought at all, the
motive tends to be not the explanation of sequences in
the causal process, but the testing for and protection against
bias and confounding. Hence the micro-level too is generally
neglected in risk factor studies.
The most rigorous design available—one which I among many
others have set up as an ideal to strive toward—is the randomized controlled trial. Remarkable as it is in its rigour, this is
attained precisely by simplified conditions that exclude context
and mediation as a matter of economy. The economy and the
simplification as much as the rigour become virtually a matter
of principle. As a result this design, as usually practised, typifies
the extremes of both the first limitation of excluding context
and the second limitation of ignoring micro-level mediators.
Great prospects and opportunities have arisen for epidemiology with the extraordinary scientific developments of recent
years but they only emphasize the incapacities of the risk factor
approach in exploiting them. Together, opportunities and incapacities reinforce the need for epidemiology to go beyond the
current one-level paradigm. In moving in a new direction, we
need not turn our backs on all we have learned in the current
climate; contra Kuhn, multiple cause theory and its accompanying methods remain adaptable and usable in the new
situation. This view is in line with my modification of the strict
Kuhnian revolutionary model; it allows for attrition and gradual
change. We have time, and a new paradigm could be ours to
fashion. What then might be the premises of a new paradigm?
Ezra Susser and I have volunteered an outline of a model.1 We
agree that the fundamental prerequisite is a multidimensional
model of causality, expanded in both space and time. In the end,
as we learn to understand and apply such a model, we shall
have gone beyond linear sequences to relations and interactions
within and across systems at different levels.
The space dimension will need to encompass successive levels
of organization, rising from micro-levels within individuals
to societies beyond them. The time dimension, central in the
search for causes, has had more implicit than explicit attention.
Time course and history apply at all levels, whether molecular,
or individual, or familial or societal. Here the signs are propitious. Lately, fresh interest has been generated in the antecedents of individual development beginning with conception. At
the same time, a new look is being taken at the unfolding of the
whole of the life course. And new interest has appeared in the
impact on disease of the historical evolution of societies.
EPIDEMIOLOGIST—SOCIAL SCIENTIST OR MOLECULAR BIOLOGIST?
All this has implications for how we think about health
problems and study them, for instance in dealing with concepts
or risk, or ‘third’ variables, or disease. First, consider risk. In
risk factor epidemiology, this is almost invariably a risk ratio of
exposure to outcome. The value obtained, an average of the
assembled individuals for a given age or moment in time, is
assumed to be stable. This is convenient shorthand. What might
we learn if we could take into account the possible flux in risk
during the time elapsed? Might we reshape our notions of
causality and remedial action if, studiously, we found ways to
include in our models the context as it changes from level to
level. And then, in addition, if we take account of the contextual antecedents of exposure, and the mediators at all levels
between exposure and outcome which constitute confounding
by so-called ‘third variables’? We know, after all, that in a hierarchy of levels of organization, causes contributing to a given
outcome will differ because of emergent properties unique to
each level, the interactions between levels, and the topography
of the overarching environment.22
Perhaps these ideas seem needlessly to complicate research as
well as thought. But to turn to the experience of the microlevels of molecular biology, new discoveries frequently seem to
depend on just such shifts and interactions and emergent
properties between levels. If so, they are bound to be multiplied
by the proliferation of variables and the free ranging lives of
human beings—with which epidemiologists must deal. We shall
surely have to face these difficulties head on.
Outcomes too require multilevel definition to suit multilevel
analysis. In individuals alone, outcomes can range from molecular, biochemical, physiological and immunological biomarkers,
through organic impairment and specific disease, to illness or
dysfunction at the level of the psyche, and the social roles of
sickness and handicap. Beyond the individual, family, community and societal health are not outside the epidemiological
and public health remit.
As yet, I can rise to no more than a sketch or meagre outline
of what multidimensional epidemiology is or can do. On the
other hand I may have done enough to frighten off the serious
epidemiologist because of the inherent difficulty of the task
before us and the lack of tools to accomplish it. In truth the
situation is not so bad. One can offer some small comforts; they
follow from observation of the course of epidemiology over the
past half century, complemented by the history of modern
science. These convince me that those epidemiologists who
would choose to preserve our public health tradition will be
obliged to make a conscious choice to broaden and deepen
our approach in the future. With the choice made, direct
experience of the risk factor era confirms other historical
evidence of scientific and technological advance—the demands
of the endeavour in themselves generate the needed techniques
and methods.
When the risk factor era began with the search for multiple
causes, it took a while before the theory was comprehended
fully, and the designs and methods had to be developed piecemeal and newly minted.23 The same prospect is before us now.
One noted above that several epidemiologists, like Moliere’s
M. Jourdain who for 40 years had been speaking prose without
knowing it, have in fact thought and practised in multiple
dimensions. The missing elements are not only the methods,
but the concepts and the theory to override the straitjacket of
S1021
the risk factor paradigm. With that in hand, epidemiology can
move into a new era, developing its tools of design and analysis
along the way.
Epilogue
Peter Ludwig Panum was one of the first among our great predecessors to think and work on multiple levels of organization.
Panum was 26 years old, and had graduated from Kiel in
medicine only a year before he was selected by the government
of Denmark to investigate the outbreak of measles in the Faroe
Islands in 1846. In fulfilling his assignment more than 150 years
ago he carried out a truly remarkable field investigation; in my
view worthy of the best standard operation the Centres for
Disease Control in the US would do today.
Still more remarkable was its multilevel nature. He studied
the whole way of life of the 7782 people scattered across the
several islands (about 6000 islanders were killed by the disease)
in the conviction that he would find clues to the nature of
the then deadly epidemic. And indeed, from what he found
he deduced firstly the transmissible nature of the disease and its
incubation period, which he arrived at by noting the isolation of
the islands and by tracing the successive timing of contact and
disease. Secondly he noticed the greater mortality of attacks
with increasing age only after age 20 (except for infants under
one year old) and up to age 60, which he recognized from the
distribution of mortality. Thirdly he understood the individual
immunity conferred by infection, which he inferred from the
history of the previous epidemic more than 60 years before
in 1781, and the absence of repeat infection in the surviving
population of 1846.
Jacob Henle published his influential paper on infection as a
cause of disease, challenging the dominant miasma theory,
in 1840 shortly before Panum performed his feat. But it was
another multidimensional thinker, the great Rudolf Virchow,
who on meeting Panum arranged for the publication and dissemination of his observations in the first volume of the new
journal Virchows Archives of Pathology in 1847.
It is worth citing two brief paragraphs of Panum’s report* to
allay sceptics and illustrate my point:
He writes: ‘When a physician is called to work in a place
where climatic and dietetic conditions are different from those
to which he has been accustomed, his first problem is to study
the hygienic potentialities which affect the state of health of the
inhabitants. It is, in fact, these hygienic conditions which contribute to the development and frequency of some diseases and
the improbability and rarity of others, and which, more or less,
modify the symptoms of every disease, and it is indeed on these
conditions that the geography of disease, the special study of
which subject will soon, perhaps, elevate it to the status of
an independent science, is based.’17 Of course, that science is
epidemiology.
He continues… ‘I shall then, try to set forth here the hygienic
forces proceeding from the conditions on the islands, and as far
as the observations I have been able to make permit me to do
so, I shall attempt to show the influence which each of these
forces in particular exerts on the general state of health of the
* We owe thanks to Wade Hampton Frost, who commissioned the translation
by Mrs Hatcher for the US Public Health Service.
S1022
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
inhabitants and on the frequency, development and method of
propagation of different diseases. Together with the mortality
rates of the country, I shall also seek to illustrate this further by
statistical data collected during my sojourn on the Islands. In
another section I shall then present some observations in regard
to the measles, inasmuch as they may be of general interest to
the medical public.’
Panum’s subsequent contributions were mainly in physiology. They do not deny a vision of research that probes multiple
levels of organization and makes exemplary use of the time
dimension.
8 Blumberg BS. Adaptation to infectious disease. Australia antigen and
hepatitis. Am J Phys Anthropol 1970;32:305–08.
9 Prince AM, Szmuness W, Michon J et al. A case-control study of the
association between primary liver cancer and hepatitis B infection in
Senegal. Int J Cancer 1975;16:376–83.
10 Beasley RP, Hwang LY, Cin CC et al. Hepatocellular carcinoma and
hepatitis B virus. A prospective study of 22 707 men in Taiwan. Lancet
1981;ii:1129–33.
11 Gajdusek DC, Zigas V. Degenerative disease of the central nervous
system in New Guinea. N Engl J Med 1957;257:974–78.
12 Lindenbaum S. Sorcery: Disease and Danger in the New Guinea Highlands.
Palo Alto, California: Stanford University Press, 1973.
13 Wynder EL, Graham EA. Tobacco smoking as a possible etiologic
Acknowledgement
Ezra Susser and Sharon Schwartz allowed me to draw on a
number of ideas we developed together in preparing a joint
paper for Annual Reviews of Public Health to be published in
June/July 1999.
References
1 Susser M, Susser E. Choosing a future for epidemiology. Parts 1,2.
Am J Public Health 1996;86:668–73; 674–78.
2 Susser M. Causal Thinking in the Health Services: Concepts and Strategies in
Epidemiology. New York, London: Oxford University Press, 1973.
3 Susser M. The logic of ecologic. Parts 1,2. Am J Public Health 1994;84:
825–29; 830–35.
4 Beral V. Risk of Kaposi’s sarcoma in persons with AIDS: or sexually
transmitted infection? Lancet 1990;335:123–28.
5 Chang Y, Cesarman E, Pessin MS et al. Identification of herpesvirus-
factor in bronchogenic carcinoma. JAMA 1950;143:336–38.
14 Doll R, Hill AB. Smoking and carcinoma of the lung. Br Med J 1950;
ii:1071–81.
15 Ross R. The Prevention of Malaria (2nd edn). London: John Murray, 1911.
16 Snow J. On the Mode of Communication of Cholera (2nd edn, much enlarged).
London: J. Churchill; 1 & 55. Reprinted as Snow on Cholera. New
York, NY: Commonwealth Fund, 1936.
17 Panum PL. Observations made during the epidemic of measles on the
Faroe Islands in the year 1846. Bibliothek for Laeger, Copenhagen,
3 R 1847;1:270–344. Transl. Hatcher AS. In: Medical Classics 1939;3:
802–66.
18 Van den Broucke JP 1989. Is the cause of cancer a miasma theory for
the end of the twentieth century? Int J Epidemiol 1988;17:708–09.
19 Fleck L. Genesis and Development of a Scientific Fact (1935), Transl.
Bradley F, Trenn TJ. In: Trenn TJ, Merton RK (eds). Chicago 1.ii.:
Chicago University, 1979.
20 Kuhn TS. The Structure of Scientific Revolutions (2nd edn). Chicago: Chicago
University Press, 1970.
like DNA sequences in AIDS associated Kaposi’s sarcoma. Science
1994;225:1865–69.
21 Schwartz S, Susser E, Susser M. A future for epidemiology. Ann Rev
6 Ottman R, Pike ML, King M-C et al. Practical guide for estimating risk
22 Susser M. Does risk factor epidemiology put epidemiology at risk?
for familial breast cancer. Lancet 1983;ii:556–58.
7 Hall JM, Lee MK, Newman B et al. Linkage of early onset familial
breast cancer to chromosome 17Q21. Science 1990;250:1684–89.
Public Health 1999;20:15–33.
Peering into the future. J Epidemiol Community Health 1998;52:608–11.
23 Rosen G. A History of Public Health (Expanded edn). Baltimore and
London: The Johns Hopkins University Press, 1993, pp.273–363.