A fundamental starting point when considering the teaching of epidemiologic methods is the scope of the subject matter. After reviewing a number of current and historical epidemiology textbooks as well as the published literature, I have been surprised to find the lack of any clear definition or boundaries of what constitutes “epidemiologic methods.” This is likely a consequence of the history and evolution of our field, where individuals with heterogeneous training and experience—including physicians, statisticians, and demographers—have taken part in shaping the areas of interest to the discipline.1,2
What then constitutes those methods that enable us to study “the distribution and determinants of disease frequency”? Morabia2 describes a useful framework for characterizing interrelated areas where epidemiologic concepts and methods have evolved, as follows.
Population Thinking: “a mode of conceptualizing issues for a whole group of people defined in a certain way.”2 Methods associated with population thinking are those typically termed “descriptive epidemiologic methods,” including those related to:
* Defining populations over time. This would include the methods and approaches to define and identify target, source, and study populations. Clinical disease surveillance and cohort studies are designs that require defining and following populations over time.
* Measurement. This includes case definition and ascertainment, exposure assessment, the familiar metrics of mortality, disease occurrence and risk, and also methods for determining the quality of measurements, components of measurement error, and methods to assess and correct for mismeasurement. Practical issues in measurement include questionnaire development and study conduct. The identification and use of biomarkers, intermediate outcomes, and refined measures of exposure and outcome are natural outgrowths of concerns for measurement in epidemiology.
These areas are not solely the domain of epidemiologists but are shared with demography, statistics, sociology, econometrics, and other disciplines that are involved with measurement of biological, psychological, behavioral, or societal-level processes that define and affect population outcomes.
Group Comparison: “contrasting what is observed in the presence of exposure to what would have occurred had the group of interest not been exposed to the postulated cause.”2 Methods associated with group comparisons constitute those typically thought of as “analytical epidemiology.” In combination with use of modern statistical analysis methods, methods that are essential to comparisons in epidemiology include:
* Causal inference. Causal inference concepts and methods help epidemiology to define, identify and overcome threats to appropriate inference (such as confounding and bias). Causal inference has advanced considerably over the past 2 decades and epidemiology has been an important contributor to these advances.
* Study designs. The theoretical and practical underpinnings of observational and interventional studies are among the bedrocks of epidemiologic methods.
As with population thinking, these are not domains solely relevant to epidemiology. Causal inference methods, for example, have grown not only in the field of statistics, but play an important role in other disciplines where nonexperimental methods may dominate (eg, economics, education).
Because of my own background (Ph.D. in statistics), I am periodically asked how epidemiologic and statistical methods are distinguished and how training in epidemiologic methods meaningfully extends training in statistics. I believe the areas outlined above comprise a set of theoretical and practical methods that, when coupled with statistical methods, produce a fundamentally unique epidemiologic approach to evaluating diseases in populations. Classical statistical training typically does not cover the above areas, although many biostatistics and other ‘applied’ programs integrate some of these components. Strong foundations in these methods of population thinking and group comparisons set the stage for specialized research and application in public health as reflected in the many subspecialties within epidemiology. (I have yet to meet a self-identified “infectious disease statistician” whereas I interact daily with infectious disease epidemiologists.)
Consider, for example, what is necessary to evaluate whether a particular exposure is associated with earlier time of death. Understanding the mathematical structure of a variety of nonparametric, semiparametric, and parametric approaches for linking the exposure to mortality, the format of the required data components, techniques for model estimation, the distributional properties of the estimated parameters, and approaches to evaluate model assumptions should all be covered in statistical training for conducting a survival analysis. Training in epidemiologic methods should provide approaches for the definition and enrollment of appropriate populations, determining the appropriate nature and timing of the exposure and confounding variables, models for how the exposure is linked to mortality, the appropriateness of different study designs, the ascertainment of the outcome, the appropriate conduct of the study and data collection methods to minimize bias and error, and the ultimate assessment of causal association.
These definitions and examples may help to facilitate communication between disciplines. The purpose is not to define rigid domains of influence, but to identify common ground where methods can be advanced, and shortcomings where expectations for methods education are not well met by either discipline.
Serving the Needs of Students With Interest in Epidemiology Methods
As of 2007, the Council on Education for Public Health has accredited 39 schools of public health and 69 graduate public health programs.3 As a testament to their importance as cornerstones for public health, epidemiology programs at both large and small institutions serve an increasingly wide range of students. At Johns Hopkins, for example, in addition to students in the Epidemiology Department, we teach undergraduate and professional graduate students in public health sciences and biotechnology enrolled at the undergraduate level and students in the School of Public Health masters and doctoral programs outside epidemiology (eg, International Health, Health Policy and Management, Biostatistics). Furthermore, students and trainees from the School of Medicine are enrolled in joint degree and certificate programs in clinical investigation.
There is obviously substantial diversity in the needs and expectations of these student populations interested in learning epidemiologic methods. Perhaps equally diverse are the views of epidemiology faculty and course instructors in terms of what these different students should know. This presents a challenge for designing a cohesive curriculum. A useful organizing principle is to consider a hierarchy of 4 increasingly sophisticated objectives for describing the goals for teaching epidemiology methods4: (1) basic knowledge about epidemiologic methods to establish terminology and basic interpretation of concepts that arise in epidemiologic studies;(2) use of epidemiologic methods to increase the students’ ability to interpret and critically evaluate study designs and methods of epidemiologic analysis; (3) practice of epidemiologic methods to prepare students to perform epidemiologic research and communicate this research to other epidemiologists and professionals in other areas; and (4) teaching epidemiologic methods to prepare students to develop their own educational curricula and approaches to instruction.
The program at Johns Hopkins has evolved over the past several years, taking several approaches to instruction. First there is an array of introductory survey courses that integrate basic epidemiologic measures, study designs, and concepts of confounding and bias. These courses are tailored to student populations that need a basic knowledge of epidemiologic methods. Courses are given in-person at the School of Public Health and at the undergraduate campus and satellite facilities, as well as online.
Second, there is a series of additional methods courses that serve as an “applied track” for students at the School of Public Health. These courses are aimed at students who have completed the basic survey course and need additional training to use epidemiology and critically evaluate epidemiologic studies and literature. This includes more advanced consideration of epidemiologic measurement (eg, partitioning person-time in cohort studies), study designs (eg, nested case-control studies), causal inference, threats to validity (eg, confounding, selection and information bias, and methods for addressing these problems), as well as interpreting and communicating epidemiologic information. Most of these students come from the School MPH program as well as masters and doctoral students from departments outside of epidemiology. For this student population, it is vital to integrate these methods with practical topics important to public health problems, such as identifying sources of data and use of epidemiologic thinking to assess the burden of health problems. Separate courses have been developed for medical students and trainees, with expanded coverage of measurement, design/analysis, and diagnosis/prediction issues.
Lastly, there is a separate “research” sequence to serve students who intend to perform epidemiologic research. This sequence is required by epidemiology doctoral students, as well as students in other departments and the MPH program who are interested in research-level epidemiologic methods. Doctoral students are also required to take a series of second-year seminars to present and discuss papers on topics relevant to epidemiologic principles and practice. Additional advanced methods classes are offered in the department and through the Department of Biostatistics.
An important part of this sequence development has been coordination with the Department of Biostatistics, where epidemiology students are required to take courses as part of the methods sequence. Key aspects of overlap between the courses have been identified, including aspects of multivariate regression models and survival analysis. Typically, the courses are sequenced so that the analytic approaches are introduced in biostatistics and then closely followed with discussion in our epidemiology course. By using similar terminology, the same notation and even similar slides these coordinated courses help students to integrate the material.
Teaching. Experience in teaching is a vital part of doctoral training. An understanding of the fundamentals of epidemiologic methods is reinforced by instructing others in these principles. Moreover, because epidemiologic research interfaces directly with public policy, most graduates will be faced with the need to communicate to larger groups who may have little understanding of epidemiology. For these reasons, epidemiology graduate students are expected to participate as a laboratory instructor in at least one class for which the primary focus is epidemiologic methods and practice.
Laboratory Exercises and Computing
Most epidemiologists would agree that practical exercises help to complement and extend material presented in didactic teaching. There may be less agreement on the content of these exercises, and especially the extent to which computers and data analysis should be included.
Chapters by Abramson,4 Florney,5 and Noah6 in the book Teaching Epidemiology7 provide suggestions for exercises and computing. In the research methods courses at Hopkins, the sequence provides a progression of approaches to enhance students’ abilities and sophistication. Our first- and second-term courses provide tables and figures from published papers and require students to make hand calculations of epidemiologic measures (eg, person-time, measures of disease occurrence, life-table, Kaplan-Meier). This is supplemented with the provision of simple datasets and Stata code for computerized analyses. Our use of Stata reflects the choice of a flexible, relatively inexpensive, easy-to-use analytical package. Laboratory exercises are presented with annotated output and questions directed towards interpretation. The third-term course provides more complex datasets derived from ongoing research studies. Only in the fourth-term course, oriented towards doctoral students, is some independent data analysis and programming expected. As part of this course, students participate in an independent practicum that consists of either an analysis of data sets provided in the course or a methodologic investigation.
The Need for Teaching Resources
There are few resources that discuss methods for teaching both general epidemiologic concepts and epidemiologic methods. A leading resource is the second edition of Teaching Epidemiology edited by Olsen, Saracci, and Trichopoulus, published in 2001.7 This book has several chapters on epidemiologic methods, although they are oriented primarily towards introductory students. In 2002, the American College of Epidemiology and Association of Schools of Public Health also jointly sponsored a workshop on Doctoral Education in Epidemiology.8 There was also an invited session on teaching epidemiology organized at the 2007 Society for Epidemiologic Research annual meeting, at which an earlier version of this paper was presented.
Overall, this is a relatively small amount of material and more attention is needed to improve resources available for epidemiology methods education—particularly beyond the introductory course. There is little information to aid in curriculum design or in innovative ways to develop lecture and laboratory components. Such assistance would be an outstanding service by the major epidemiologic societies. They might consider taking a cue from the American Statistical Association that supports 2 sections devoted to education.
Professional societies could raise both the profile and interchange of educational methods by (1) devoting at least one spotlight session to education at each annual meeting; (2) accepting contributed abstracts on methods for teaching in addition to research; (3) soliciting members for examples of curricula, course syllabi and materials, and providing a central online repository for members; (4) securing funds for teaching awards, with particular recognition for graduate students and junior faculty members who may have the most to gain professionally; and (5) supporting more written work about education—such as sponsoring new books, commissioning member newsletters, or soliciting journal articles. The future health of epidemiology would strongly benefit from such investment.
I appreciate the efforts of and discussions with faculty and staff contributing to the Johns Hopkins Epidemiology research methods sequence, including Jonathan Samet, Elizabeth Platz, Lisa Jacobson, Eliseo Guellar, Shruti Mehta, Stephen Cole, M. Daniele Fallin, Thomas Glass, Steven Goodman, and Andrea Vilanti.
ABOUT THE AUTHOR
STEPHEN J. GANGE is Professor of Epidemiology at the Johns Hopkins Bloomberg School of Public Health. Trained in statistics, his research interests include cohort study methods, evaluation of therapies in observational studies, and modeling of disease biomarkers. Dr. Gange has been active in education initiatives, including developing an online introductory class for biotechnology graduate students and classes in both the “applied” and “research” track methods sequences.