Continue
We use cookies to enhance your experience on our website. By clicking 'continue' or by continuing to use our website, you are agreeing to our use of cookies. You can change your cookie settings at any time.
Find out more
Sign In
Issues
More Content
Submit
Purchase
Alerts
About
Register
Advanced Search
All American Journal of Epidemiology
Invited Commentary: Do-It-Yourself Modern Epidemiology—At Last! Alfredo Morabia American Journal of Epidemiology, Volume 180, Issue 7, 1 October 2014, Pages 669–672, https://doi.org/10.1093/aje/kwu221 Published: 04 September 2014 Article history Volume 180, Issue 7
Views
1 October 2014
PDF
Cite
Permissions
Share
Abstract
Article Contents
In this issue of the Journal, Keyes and Galea (Am J Epidemiol. 2014;180(7):661–668) propose “7 foundational steps” for
Abstract
introducing epidemiologic methods and concepts to beginners. Keyes and Galea's credo is that the methododological and
CLASSICAL FOUNDATIONAL CONCEPTS
conceptual components that comprise epidemiology, today scattered in textbook chapters, come together as an integrated and coherent methodological corpus in the process of designing studies. Thus, they expound, the process of designing studies should be the core of teaching epidemiology. Two aspects of their 7-steps-to-epidemiology, do-it-yourself user manual stand out
MODERN FOUNDATIONAL CONCEPTS THE DO-IT-YOURSELF APPROACH
steady state evolving across time, and 2) the ambition to teach modern epidemiology in introductory courses, instead of the popular mix of classical and modern epidemiology that is often used today to keep introductory courses simple. Both aspects are of potentially great significance for our discipline.
ACKNOWLEDGMENTS
Keywords: causal frameworks, classical epidemiology, modern epidemiology, teaching
REFERENCES < Previous
as novel: 1) the approach, because of its emphasis on modern epidemiology's causal framework of a dynamic population in a
Topic: steady state, epidemiology, teaching, galea aponeurotica Issue Section: Commentaries
Next >
In an accompanying article (1) and in their book (2), Keyes and Galea propose “7 foundational steps” with which to introduce epidemiologic methods and concepts to beginners. Their objective is to impress upon students the coherence of the whole epidemiologic endeavor, from conceptualization of the target population to the measurements, comparisons, association measures, critical assessment, inference, and consequences. In their view, today the components of epidemiologic theory are taught separately as full, final, labeled products, leaving students at a loss to assemble the pieces of a disjointed epidemiologic machine and make it work. In contrast, they propose to incorporate these components progressively into the process of evaluating a specific hypothesis in a population study, and their do-it-yourself manual provides the procedures with which to “articulate” the individual pieces of theory. I believe their approach is novel—even if Keyes and Galea modestly claim it is not—and could potentially facilitate access to modern epidemiology for a public currently baffled by the often abstract and mathematical content of textbooks. However, one point in their rationale deserves clarification. Keyes and Galea claim that their approach is “foundational,” whereas the prevalent approach is “historical.” They contrast the approach they submit, an “approach grounded not in labels but in foundational concepts” (1, p. 661), with the existing approaches introducing epidemiology's components essentially in the order in which they historically appeared: designs first, then measures of association, then confounding, and so on. The term “foundational” appears 12 times. It is dear to Keyes and Galea but unfortunately never defined. I understand their “foundational concepts” as being equivalent to a causal framework and standing for a set of simplifying assumptions and principles about the studied reality. If you imagine each epidemiologic study as being a piece of a jigsaw puzzle, very roughly, these foundational concepts are like the mental representation needed to start assembling the first pieces of the puzzle when the complete picture is missing. Keyes and Galea's framework is a schematic representation of a population and the transformation it undergoes as the epidemiologic strategy for revealing a causal link unfolds. Sticking to Keyes and Galea's terminology, I argue here that clarifying what is meant by “foundational concepts” helps to characterize the novelty of their teaching strategy and to situate it in history. Of course, it is possible to put the cumulative accrual and refinement of epidemiologic methods and concepts on a timeline and review it in specific histories of cohort studies, case-control studies, confounding, and so on (3), but this has never been the typical way in which epidemiology has been taught. Typically, methods and concepts are organized according to foundational concepts which also evolve across time, but much more slowly, dividing the history of epidemiologic teaching and research into discrete phases, 4 of which (preformal, early, classical, and modern) can be identified in the past and more of which will emerge in the future (4). These phases are so different from each other that it is usually easy to relate a textbook to a specific phase on the basis of its foundational concepts.
CLASSICAL FOUNDATIONAL CONCEPTS Take the foundational concepts of classical epidemiology, those formed between 1945 and 1965. They connect vital statistics and study designs in a “2-stage” approach (5, p. 51). The first stage consists of developing a representation of what is happening in the population from the analysis of health, morbidity, mortality, and census-based statistics. Classical epidemiologists used to acquire in their training the expertise to manipulate these data, obtain valid contrasts across personal characteristics (mainly age, sex, and socioeconomic status), time, and geographical locations, interpret those contrasts, and generate hypotheses to be tested, in a second stage, in specific group comparisons. For example, before World War II there had been a long discussion about whether the upward trends in lung cancer were real or artifacts of better diagnosis (6). Before the epidemic of lung cancer was recognized after World War II, there was no motivation to investigate a link with tobacco use. Vital statistics are the kingpin of classical epidemiology. Classical textbooks are typically founded on some form of the triad “persons, time, and place” (7–10). Some of the introductory textbooks mentioned by Keyes and Galea still dedicate a specific chapter, usually entitled “descriptive” epidemiology, to this triad, but it has lost its centrality. It is telling that an instruction to “examine population health, morbidity, and mortality data to get an idea of what is going on in this population” is not a “foundational step” in Keyes and Galea's study-building progression. Their approach is not classical.
MODERN FOUNDATIONAL CONCEPTS The nature of the foundational concepts referred to by Keyes and Galea is encapsulated in their statement, “At its very core, epidemiologic science involves watching groups of individuals develop health outcomes over time. … Individuals enter and leave the population across time, visually illustrating the dynamic nature of the population” (1, p. 667). The dynamic population in a steady state, which constitutes the first building block of their approach, is the unmistakable “source population” of the foundational concepts that detached modern epidemiology from classical epidemiology (11, p. 229):
[T]he source population of subjects tends to constitute a (dynamically) static group in each category of age (figure 1), with new people continually entering it at the lower bound (and within the range) of age and others exiting it (within the range and) at the upper bound[.]
Modern epidemiology reduced the 3 parameters of the classical triad to 2, persons and time. Figure 1, adapted from Miettinen's epochal paper (11), depicts the modern foundational concepts. Individuals could move into and out of the population, but these movements were assumed to keep the main characteristics of the population—areas C and D of the figure—constant, that is, in a steady state. “Place” was gone. The loss compared with the classical triad was that the new foundational concepts were purely abstract. The gain was that all of the connections involving persons and time could be expressed mathematically.
Figure 1.
View large
Download slide
Foundational concepts of modern epidemiology, as presented by Miettinen (11) in 1976 (reconstruction). The legend of the figure reads: “Figure 1. Static population (e.g., a particular age group) over the time-span of a case-referent study based on incident cases. The sizes of the different component populations remain static, but there is turnover of membership in each compartment. The arrows indicate occurrences of new cases, i.e., transitions from the candidate pools to the prevalence pools. Note that the incidence-densities are zero in each of the prevalence pools, and that the incident cases are referred to the follow-up experiences ¢¢ ¢¢ ¢¢ ¢¢ in the candidate pools only. Also note that the incidence-density ratio is (a /b )/(C/D), with a /b and C/D estimable from the (incident) cases and referents, respectively, regardless of the levels of incidence or prevalence” (11, p. 228).
With some good will, admittedly, Miettinen's 1976 paper (11) can be shown to forerun Keyes and Galea's 7 steps. Where classical epidemiology, referred to as “classical rationale” by Miettinen, categorized cohort and case-control studies separately and conceived of controls as people who were free of the disease at the end of the risk period, the modern perspective was that whether a study qualified as a cohort study or case-control study depended on the ways in which the dynamic population was sampled. Here is what Miettinen wrote in 1976 (11, p. 227):
The classical rationale does not, however, bear on the ordinary type of case-referent study in chronic-disease epidemiology —the type of study in which ascertainment occurs before the individual risk-periods are over, and in which incident rather than prevalent cases are enrolled. (…) For a given exposure and illness, the objectives of a case-referent study are basically no different from those of a follow-up (“cohort”) study.
These ideas are echoed in Keyes and Galea's paper (1, p. 665):
We demonstrate 3 ways to sample a dynamic population. … If at any point we take a sample of exposed and unexposed people without disease, we could follow them forward and conduct a cohort study. (…) [I]f at any point we take a sample of diseased and nondiseased persons and then assess exposure status, we conduct a case-control study.
The following excerpt from Miettinen (11, p. 227) resonates like a residual vibration for Keyes and Galea's steps 6 and 7:
The internal validity of the study involves the following components: a) validity of selection: the probability of ascertainment is uninfluenced by the exposure history or status itself; b) validity of observation: lack of misclassification between cases and non-cases (referents, comparends, “controls”) and between exposed and nonexposed; and c) validity of comparison: the use of a reference entity (usually diagnostic category) unrelated to the exposure, and the control of confounding.
The coherent integration of occurrence measures, study design, bias, confounding, and interaction is the hallmark of modern epidemiology, in comparison with previous phases of our discipline. Modern Epidemiology, the textbook that gave the new phase of epidemiology its name (12), expressed these foundational concepts so cogently that textbooks ever after internalized its gross architecture, as astutely illustrated in Keyes and Galea's heat map. Unless Keyes and Galea provide additional information, their novel approach to epidemiology teaching appears to belong to modern epidemiology and not to more recent developments. The modern foundational concepts worked best for noncommunicable diseases. They stumbled when the human immunodeficiency virus epidemic revived infectious disease epidemiology. Causal models conceived within the potential outcome framework (13–15) (too often but uselessly comingled with the abstruse philosophy of counterfactuals (16)) proved more adequate to address the specific methodological issues associated with time-varying treatment or exposure and covariates (e.g., see Cole et al. (17)). In the potential outcome framework, the causal association is conceptualized at the individual unit level before being estimated in populations. Such a step is absent in Keyes and Galea's approach. Another argument against the antagonism between “foundational” and “historical” teaching strategies is that Keyes and Galea could also use their approach to revisit historical epidemiologic studies, identify their strengths and weaknesses, and witness the progressive emergence of the modern epidemiologic corpus (18).
THE DO-IT-YOURSELF APPROACH Thus, in this perspective, the confrontation does not take place between a “foundational” approach and a “historical” approach, as Keyes and Galea seem to view it, but between different ways of distilling the same foundational concepts. The former is more theoretical and the latter is more pragmatic, striking me as a 7-steps-to-epidemiology, do-it-yourself approach in which 2 aspects stand out as novel: 1) the approach, because of its emphasis on modern epidemiology's causal framework of a dynamic population in a steady state evolving across time, and 2) the ambition to teach modern epidemiology in introductory courses, instead of the popular mix of classical (e.g., the odds ratio approximates the risk ratio when the outcome is rare, and so on) and modern epidemiology often used today to keep introductory courses simple. Both aspects are of potentially great significance for our discipline. Graphics are an important dimension of Keyes and Galea's didactic strategy and perpetuate a tradition assuming that public health practitioners learn better using graphics than mathematics. According to Eyler, Arthur Newsholme imparted this view 110 years ago (19, p. 38):
To make the process [of constructing life tables] accessible to Medical Officers of Health (MOH), few of whom knew calculus, Newsholme championed the use of a graphical method of interpolating the numbers living or dying at each year of life (…) Newsholme argued that his graphical method put the construction of a life table for his jurisdiction within the reach of every MOH and he attempted to demonstrate that the results obtained by this method were as accurate as those obtained from the same data using more analytical means.
In scheming their textbook, Keyes and Galea were inspired by “the economics field” for its ability “to reduce itself to core canonical concepts, which are cast as ‘laws’ ” (1, p. 664). Interestingly, the classical epidemiologist Lilienfeld, in an early reaction to the emergence of modern epidemiology, which he viewed skeptically as a type of “mathematization of biological or of natural phenomena,” felt similarly (20, p. 144). Lilienfeld first cited what the economist Robert Solo had written about econometric models (21, p. 48):
Ultimately, the mode of expression that will most conform to the scientific ideal will not be the image-free symbols of mathematics but rather the imagery of normal communication and intercourse with its reference base in the specifics of experience, for only if the general statement is so framed can it be continuously bridged into direct observations and contrasted with ongoing experiences.
Lilienfeld then added, “This judgment, it seems to me, applies equally well to epidemiology” (20, p. 144). Forty years later, Keyes and Galea are proposing to infuse the “imagery of normal communication” into modern epidemiology. At last!
ACKNOWLEDGMENTS Author affiliation: Barry Commoner Center for Health and the Environment, Queens College, City University of New York, New York (Alfredo Morabia); and Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, New York (Alfredo Morabia). A.M. was supported by grant 1G13LM010884 from the National Library of Medicine. I thank Zoey Laskaris for comments on an earlier version of this article. Conflict of interest: none declared.
REFERENCES 1 Keyes KM, Galea S. Current practices in teaching introductory epidemiology: how we got here, where to go, Am J Epidemiol , 2014, vol. 180 7(pg. 661-668) Google Scholar
CrossRef
PubMed
2 Keyes KM, Galea S. , Epidemiology Matters: A New Introduction to Methodological Foundations , 2014New York, NYOxford University Press 3 Morabia A. Ahrens W, Pigeot I. History of epidemiologic methods and concepts, Handbook of Epidemiology , 2014Berlin, GermanySpringer(pg. 43-74) 4 Morabia A. , A History of Epidemiological Methods and Concepts , 20041st edBasel, SwitzerlandBirkhäuser 5 Lilienfeld AM. Epidemiological methods and inferences in studies of non-infectious diseases, Public Health Rep , 1957, vol. 72 1(pg. 51-60) Google Scholar
CrossRef
PubMed
6 Cancer of the lung, Br Med J , 1942, vol. 1 4247(pg. 672-673) CrossRef
PubMed
7 Morris JN. , Uses of Epidemiology , 19571st edEdinburgh, United KingdomE. & S. Livingstone Ltd. 8 MacMahon B, Pugh TF, Ipsen J. , Epidemiologic Methods , 1960Boston, MALittle, Brown & Company 9 Lilienfeld AM. , Foundations of Epidemiology , 1976New York, NYOxford University Press 10 Doll R, Peto R. The causes of cancer: quantitative estimates of avoidable risks of cancer in the United States today, J Natl Cancer Inst , 1981, vol. 66 6(pg. 1191-1308) Google Scholar
PubMed
11 Miettinen O. Estimability and estimation in case-referent studies, Am J Epidemiol , 1976, vol. 103 2(pg. 226235) Google Scholar
PubMed
12 Rothman KJ. , Modern Epidemiology , 19861st edBoston, MALittle, Brown & Company 13 Greenland S, Robins JM. Identifiability, exchangeability, and epidemiological confounding, Int J Epidemiol , 1986, vol. 15 3(pg. 413-419) Google Scholar
CrossRef
PubMed
14 Rubin DB. Estimating causal effects of treatments in randomized and nonrandomized studies, J Educ Psychol , 1974, vol. 66 5(pg. 688-701) Google Scholar
CrossRef
15 Hernan M, Robins JM. , Causal Inference Boca Raton, FLChapman & Hall, Inc In press. http://www.hsph.harvard.edu/faculty/miguel-hernan/causal-inference-book/ Accessed April 8, 2014 16 Lewis D. , Counterfactuals , 1973Oxford, United KingdomBlackwell Publishing Ltd. 17 Cole SR, Hernán MA, Robins JM, et al. Effect of highly active antiretroviral therapy on time to acquired immunodeficiency syndrome or death using marginal structural models, Am J Epidemiol , 2003, vol. 158 7(pg. 687-694) Google Scholar
CrossRef
PubMed
18 Morabia A. , Enigmas of Health and Disease: How Epidemiology Helps Unravel Scientific Mysteries , 2014New York, NYColumbia University Press 19 Eyler JM. , Sir Arthur Newsholme and State Medicine, 1885–1935 , 1997Cambridge, United KingdomCambridge University Press 20 Lilienfeld AM. Epidemiology of infectious and non-infectious disease: some comparisons, Am J Epidemiol , 1973, vol. 97 3(pg. 135-147) Google Scholar
PubMed
21 Solo RA. New maths and old sterilities, Saturday Review (pg. 47-48) January 22, 1972
© The Author 2014. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail:
[email protected].
View Metrics
Email alerts New issue alert Advance article alerts Article activity alert
Receive exclusive offers and updates from Oxford Academic
More on this topic Invited Commentary: Consequential(ist) Epidemiology: Let's Seize the Day Galea Responds to “Consequential(ist) Epidemiology: Finally” Counterpoint: The Treatment Decision Design Counterpoint: Epidemiology to Guide Decision-Making: Moving Away From Practice-Free Research
Related articles in Google Scholar
Related articles in PubMed Investigation of Scarlet Fever Outbreak in a Kindergarten. Epidemiology, Microbiological and Clinical Features, Treatment, and Outcomes of Infective Endocarditis in Crete, Greece. Epidemiologic and clinical characteristics of nontransfusion-dependent thalassemia in the United States.
Citing articles via Google Scholar CrossRef
Latest Most Read Most Cited Real World Evidence for Regulatory Decisions: Concomitant Administration of Zoster Vaccine Live and Pneumococcal Polysaccharide Vaccine Multilevel Regression and Poststratification: A Modelling Approach to Estimating Population Quantities From Highly Selected Survey Samples Parental Unemployment and Offspring Psychotropic Medication Purchases: A Longitudinal Fixed-Effects Analysis of 138,644 Adolescents Contribution of Smoking to Tuberculosis Incidence and Mortality in High Tuberculosis Burden Countries Chronic Medical Conditions and Perinatal Mental Illness: A Systematic Review and Meta-Analysis
About American Journal of Epidemiology Editorial Board
LinkedIn Purchase Recommend to your Library
Author Guidelines Facebook
Advertising and Corporate Services
Twitter
Journals Career Network
American Journal of Epidemiology
Online ISSN 1476-6256 Print ISSN 0002-9262 Copyright © 2018 Johns Hopkins Bloomberg School of Public Health
About Us Contact Us Careers Help Access & Purchase Rights & Permissions Open Access Connect Join Our Mailing List OUPblog Twitter Facebook YouTube Tumblr Resources Authors Librarians Societies Sponsors & Advertisers Press & Media Agents Explore Shop OUP Academic Oxford Dictionaries Oxford Index Epigeum OUP Worldwide University of Oxford Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide
Copyright © 2018 Oxford University Press
Cookie Policy
Privacy Policy
Legal Notice
Site Map
Accessibility
Get Adobe Reader