Search

Over the past century, the field of epidemiology has evolved and adapted to changing public health needs. Challenges include newly emerging public health concerns across broad and diverse content areas, new methods, and vast data sources. We recognize the need to engage and educate the next generation of epidemiologists and prepare them to tackle these issues of the 21st century. In this commentary, we suggest a skeleton framework upon which departments of epidemiology should build their curriculum. We propose domains that include applied epidemiology, biological and social determinants of health, communication, creativity and ability to collaborate and lead, statistical methods, and study design. We believe all students should gain skills across these domains to tackle the challenges posed to us. The aim is to train smart thinkers, not technicians, to embrace challenges and move the expanding field of epidemiology forward.

In an attempt to reproduce the results of an inconclusive 1927 report by the British Medical Research Council on the hereditary versus social origins of rheumatic fever, Read, Ciocco, and Taussig, from Johns Hopkins University, with the support of Frost, conducted a case-control study in 1935 and 1936. Their study, which appeared in the American Journal of Hygiene in 1938, was outstanding for its clear and tidy rationale for separating hereditary from environmental causes. The authors compared the prevalence of rheumatic fever among the relatives of 33 children admitted for "incident" rheumatic fever and 33 control children admitted in a tuberculosis clinic for reasons other than rheumatic fever. Both rheumatic fever (cases) and tuberculosis (controls) were diseases of the poor. All family members of both cases and controls, including uncles, aunts, and grandparents, were eligible for interview and physical examination. The results were compatible with the presence of an "inherited predisposition" to rheumatic fever because the disease was more prevalent among the uncles, aunts, and grandparents of case patients than among those of control patients. Methodologically, the paper by Read, Ciocco, and Taussig is an important but almost completely forgotten milestone in the evolution of case-control studies and of genetic epidemiology.

The number of students and disciplines requiring basic instruction in epidemiologic methods is growing. As a field, we now have a lexicon of epidemiologic terminology and particular methods that have developed and become canonical through the historical development of the field. Yet, many of our basic concepts remain elusive to some students, particularly those not pursuing a career in epidemiology. Further, disagreement and redundancy across basic terms limit their utility in teaching epidemiology. Many approaches to teaching epidemiology generally start with labeling key concepts and then move on to explain them. We submit that an approach grounded not in labels but in foundational concepts may offer a useful adjunct to introductory epidemiology education. We propose 7 foundational steps in conducting an epidemiologic study and provide examples of how these steps can be operationalized, using simple graphics that articulate how populations are defined, samples are selected, and individuals are followed to count cases. A reorganization of introductory epidemiology around core first principles may be an effective way forward for educating the next generation of public health scientists.

For almost a century, epidemiologists have stratified age-specific disease rates by year of birth to better understand the distribution of a disease in a population and its evolution across time. In the present article, I review the contributions of John Brownlee, Kristian Feyer Andvord, and Wade Hampton Frost and, to accentuate the similarities of their approaches, redraw their original graphs of age-specific death rates of tuberculosis organized either by year of death or year of birth. In addition, this article reports on an apparently universally forgotten publication in the American Journal of Hygiene published in 1929, which both upsets the conventional history of the earliest reports of disease rates stratified by birth cohorts and challenges the theory that Frost discovered cohort analysis independently and gave it its name.

In 1946, the physician and statistician Joseph Berkson (1899–1982) pointed out that two diseases that are independent in the general population may become ‘spuriously associated’ in hospital-based case-control studies.1 This spurious association was later referred to, often in lively debates,2–14 as Berkson’s fallacy, Berkson’s paradox or Berkson’s bias. Some authors restricted the interpretation of Berkson’s fallacy to disease-disease associations,2,5,7,8 whereas others thought that the fallacy would also apply to exposure-disease associations in hospital-based case-control studies.10–15 In this article we use directed acyclic graphs (DAGs) to describe the structure of Berkson’s fallacy, first for disease-disease associations and then for exposure-disease associations. This permits us to understand the contentious debates and strongly differing opinions about Berkson’s fallacy, and has practical implications for study design and interpretation (see Box 1). Box 1. Practical implications of Berkson’s fallacy The fallacy that became eponymous for Berkson caused controversy from its initial formulation onwards. Some held that it biased all case-control studies in hospitals; others maintained that it was only pertinent for associations between prevalent diseases, and did not exist for exposure-disease associations. The DAG analyses in this paper show that Berkson’s fallacy can exist when studying exposure-disease associations, be it only in an ‘indirect’ form and in exceptional circumstances: when in a hospital-based case-control study persons with a prevalent diagnosis are enrolled who were hospitalized for another disease that is associated with this exposure. The DAG of the problem that Berkson originally described has the same structure of all biases due to conditioning on a collider, but cannot be endowed with a causal interpretation since it was formulated and worked out as a problem of the association of prevalent diseases. When using incident cases in hospital-based case-control studies, Berkson’s fallacy becomes highly unlikely for exposure-disease associations, unless there are many people who have developed two different new diseases more or less at the same time and are hospitalized for the other disease, i.e. the disease that is not the subject of the case-control study. When incident cases in a case-control study consist only of people who have been hospitalized for that disease, Berkson’s fallacy is not possible. It is likely that Berkson’s fallacy has had very limited, if any, impact on the findings of epidemiological studies.

The epidemiologic approach to causal inference (i.e., Hill's viewpoints) consists of evaluating potential causes from the following 2, noncumulative angles: 1) established results from comparative, observational, or experimental epidemiologic studies; and 2) reviews of nonepidemiologic evidence. It does not involve statements of statistical significance. The philosophical roots of Hill's viewpoints are unknown. Superficially, they seem to descend from the ideas of Hume and Mill. Hill's viewpoints, however, use a different kind of evidence and have different purposes than do Hume's rules or Mill's system of logic. In a nutshell, Hume ignores comparative evidence central to Hill's viewpoints. Mill's logic disqualifies as invalid nonexperimental evidence, which forms the bulk of epidemiologic findings reviewed from Hill's viewpoints. The approaches by Hume and Mill cannot corroborate successful implementations of Hill's viewpoints. Besides Hume and Mill, the epidemiologic literature is clueless about a plausible, pre-1965 philosophical origin of Hill's viewpoints. Thus, Hill's viewpoints may be philosophically novel, sui generis, still waiting to be validated and justified.

In the summer of 1924, an outbreak of scarlet fever occurred in Flint, Michigan. Unable to trace it to the usual causes, particularly fresh milk, the Michigan Department of Health used a novel approach to disentangle the enigma: The 116 cases of scarlet fever were compared with 117 “controls” selected from neighbors of the quarantined cases and from patients at the City Health Center who had been treated for ailments unrelated to scarlet fever. The extraordinary culprit was ice cream, which had a frequent/occasional/none consumption prevalence of 60%, 34%, and 6% among the cases and 24%, 51%, and 25% among the controls, respectively. The 1925 report reads, “Detailed epidemiological investigation, by means of case histories and control histories on well persons, confirmed early suspicions and established the fact that the epidemic was spread by ice cream” (Am J Hyg. 1925;5(5):669–681). This forgotten epidemiologic study is the oldest study using the case-control design to have been resurrected thus far. The case-control study design may have been conceived simultaneously, but independently and for different purposes, in England (Janet Lane-Claypon’s 1926 report on the determinants of breast cancer) and the United States. case-control studies; disease transmission; epidemiologic methods; history of medicine; ice cream; scarlet fever

Wade Hampton Frost, who was a Professor of Epidemiology at Johns Hopkins University from 1919 to 1938, spurred the development of epidemiologic methods. His 6 publications in the American Journal of Hygiene, which later became the American Journal of Epidemiology, comprise a 1928 Cutter lecture on a theory of epidemics, a survey-based study of tonsillectomy and immunity to Corynebacterium diphtheriae (1931), 2 papers from a longitudinal study of the incidence of minor respiratory diseases (1933 and 1935), an attack rate ratio analysis of the decline of diphtheria in Baltimore (1936), and a 1936 lecture on the age, time, and cohort analysis of tuberculosis mortality. These 6 American Journal of Hygiene /American Journal of Epidemiology papers attest that Frost's personal evolution mirrored that of the emerging “early” epidemiology: The scope of epidemiology extended beyond the study of epidemics of acute infectious diseases, and rigorous comparative study designs and their associated quantitative methods came to light.

George W. Comstock (1915–2007), MD, MPH, DrPH, was lecturer and then professor of epidemiology at the Johns Hopkins University School of Hygiene and Public Health from 1956 to 2007 and served as editor-in-chief of the American Journal of Epidemiology from 1979 to 1988. This interview of George W. Comstock took place in Hagerstown, Maryland, in the spring of 1990. The selection of questions and answers published here represent approximately 10% of the whole interview, which had been reviewed and hand-corrected by Dr. Comstock. He first describes how epidemiology was taught at Hopkins in the 1950s and 1960s. He then distinguishes “epidemiology per se” from a “practical epidemiology” that works closely with local health departments, and he finally expresses his wish that in the future, epidemiology would become more widely involved in policy and accepted by policy makers. Photo of George, Margaret, and Gordon Comstock taken during World War II, most likely in 1944, while Dr Comstock was serving as Medical Officer in an Escort Destroyer Division. Keywords: history, interview, policy, teaching In this first publication of the new section “Epidemiology in History,” I pay a tribute to George W. Comstock, with whom I had the privilege to work (1, 2). George passed away in 2007. He was 92 years old. His exit was suffered as a great loss by the scientific community at large, with obituaries appearing in The Washington Post (3), The New York Times (4), Lancet (5), American Review of Respiratory and Critical Care Medicine (6), and Revista de Saude Publica (Brazil) (7). The October 1, 2008, issue of the American Journal of Epidemiology was entirely dedicated to Comstock, with personal remembrances (8–11) and discussion of his contributions to the Washington County Training Center (12) and the fields of tuberculosis research (13), noninvasive imaging and hypertension (14), respiratory disease epidemiology (15), and the epidemiologic use of biological samples (16). The articles by Comstock that were republished in that memorial issue are still great readings and examples of careful studies at a time when constraints on article length were not what they are today and articles provided a thorough review and discussion of the literature. It is fortunate that in 2003, Dale Sandler had a videotaped conversation with Comstock for “Voices” in the journal Epidemiology (17). A snippet of the video is viewable online (18). Comstock was already 88 years old then, but the video nicely captures his unforgettable voice and the gentle way in which he made authoritative statements. There also exists a short movie about George Comstock (19). I happen to have interviewed George in 1990. The cover letter of his hand-edited copy of the 55-page interview is reproduced in this article. The interview was transcribed verbatim and published without alteration except for the inclusion of a few edits handwritten by Comstock on the paper copy. However, I selected questions that were different from those published by Dale Sandler in her conversation with Dr. Comstock (17). The first set deals with his reminiscence of the way epidemiology was taught when he was a student. Comstock mentions 2 exercises that have been used over very long periods of time in the Department of Epidemiology. The first was the charity luncheon problem, which, according to Comstock, “has been used with practically no break in Epi 1 since 1955.” The exercise is “Principles of epidemiology: problem 1: An outbreak of acute sore throats following a charity luncheon” from the Department of Epidemiology, The Johns Hopkins University Bloomberg School of Public Health, 2006. The exercise is based on a paper by Farber and Korff (20). The second exercise mentioned by Dr. Comstock is “Epidemiology 2A: exercise 4: methods of longitudinal study” from the Department of Epidemiology, The Johns Hopkins University School of Hygiene and Public Health, 1986–1987. It analyzes the longevity of members the Radiological Society of North America and the American Academy of Ophthalmology and Otolaryngology and is based on 4 papers by Seltzer, Sartwell, and Matanoski (21–24). Cover letter of Dr. Comstock's hand-edited copy of the 55-page interview. In a second set of questions, Comstock contrasts what he calls “epidemiology per se,” which has no direct implications for the community, with “practical epidemiology” that works closely with local health departments. In the last set of questions, Comstock expresses his wish that in the future, epidemiology would become more widely involved in policy and accepted by policy makers. Comstock spoke plainly and was not shy about defending ideas, no matter iconoclastic they sounded. For example, when asked whether the concept of interaction was taught at Hopkins at the time of his doctorate of public health in the 1950s, he answered, “I'm sure we didn't worry about that as such. Of course, you probably know I still don't.” This 23-year-old conversation with Comstock is a historical document that reflects the thinking of a classic epidemiologist who felt his role was not only to identify causes of diseases but also to apply this knowledge to improving people's health. The generation of epidemiologists to whom Comstock belongs was driven more or less by this same mission. Looking at the world through their lenses for a moment is terribly important to perceive the historical forces that shaped the epidemiology we practice today.

Reviews are provided for the articles "The Impact of Obesity on U.S. Mortality Levels: The Importance of Age and Cohort Factors in Population Estimates" by R.K. Masters, E.N. Reither, D.A. Powers, Y.C. Yang, A.E. Burger, and B.G. Link which appears in a 2013 issue of this journal and "Obesity and U.S. Mortality Risk Over the Adult Life Course" by Masters, Powers,and Link which appears in a 2013 issue of the "American Journal of Epidemiology."

Aims To determine the prevalence of atrial fibrillation (AF) in a population-based sample of adults. Methods and results Between January 2005 and December 2007 individuals aged ≥50 years, residents of the city of Geneva, who had participated in a previous random survey were invited for follow-up examination. AF was assessed on a single resting 6-lead ECG. Reported prevalences were standardized for the age distribution of Canton Geneva. Overall participation was 72.8%. Twenty-nine cases of AF (22 men) were diagnosed among 3285 subjects (1696 men). The crude prevalence of AF (95% CI) was 0.88% (0.86, 0.90) overall, but higher in men [1.30% (1.26, 1.34)] than in women [0.44% (0.41, 0.47)]. The age-standardized AF prevalence was slightly higher [overall: 0.94% (0.91, 0.97), men: 1.23% (1.19, 1.27), women: 0.54% (0.47, 0.61)]. AF prevalence increased with age in both sexes. A ‘history of suspected arterial embolism’ (brain or legs) was higher in the AF cases (10.3 vs. 3.3%; P = 0.03). Conclusion This population-based survey of a general Swiss population indicates that the prevalence of AF remains below 1%. These results are less alarming than those from previous studies based on patients seeking medical care.

In the mid-19th century, the German hygienist Max von Pettenkofer viewed cholera as resulting from the interaction between a postulated cholera germ and the characteristics of soils. In order to cause cholera, the cholera germ had to become a cholera miasma, but this transformation required prolonged contact of the germ with dry and porous soils when groundwater levels were low. This hypothetical germ-environment interaction explained more observations than did contagion alone. Despite its attraction, von Pettenkofer's postulate also implied that cholera-patient quarantine or water filtration was useless to prevent and/or control cholera epidemics. The disastrous consequences of the lack of water filtration during the massive outbreak of cholera in the German town of Hamburg in 1892 tarnished von Pettenkofer's reputation and marked thereafter the course of his life. von Pettenkofer's complex mode of thinking sank into oblivion even though, in hindsight, germ-environment interactions are more appropriate than is bacteriology alone for explaining the occurrence of cholera epidemics in populations. Revisiting the fate of von Pettenkofer's theory with modern lenses can benefit today's quest for deciphering the causes of complex associations.

Wilhelm Weinberg, German physician, founder, and president of the Stuttgart Society for Racial Hygiene, published in 1913 the results of a large, retrospective cohort study entitled Die Kinder der Tuberkuloesen (Children of the Tuberculous). The exposed cohort comprised 18,212 children whose 3,246 fathers and 2,022 mothers died of tuberculosis between 1873 and 1902. The unexposed cohort comprised 7,574 children whose 1,830 parents died of causes other than tuberculosis in 1876, 1879, or 1886. He found that children of tuberculous parents had higher mortality rates and lower fertility than did children of nontuberculous parents. Because of its size, rigorous design, and meticulous analysis, Weinberg's cohort study stands as one of the major epidemiologic works carried out before 1945.

The authors evaluated the contributions of nine genetic (G) variants (selected from 275 single nucleotide polymorphisms in 11 reverse cholesterol transport pathway genes), five environmental (E) factors (selected from 10), and G x G, E x E, and G x E interactions in explaining population variance of blood lipid concentrations. Total cholesterol, triglycerides, and high density lipoprotein (HDL) cholesterol were measured, and low density lipoprotein (LDL) cholesterol and HDL cholesterol/LDL cholesterol ratio were calculated in a population-based random sample of 1,543 men and women in Geneva, Switzerland, aged 35-74 years in 1999-2001. Explained variances (R2) for HDL cholesterol/LDL cholesterol ratio, HDL cholesterol, and LDL cholesterol, respectively, were 34%, 33%, and 19%, decomposed into main effects of G (6%, 4%, and 5%) and E (25%, 28%, and 11%), with just 3%, 2%, and 3% due to G x G, E x E, and G x E interactions, respectively. Risk factor clustering was only moderate: 70% of study subjects had < or =3 variants, 75% had < or =2 environmental exposures, and 69% had < or =5 of both types of factors. Multiple genes with weak associations, together with more dominating environmental factors, are involved in determining blood lipid concentrations. Interactions added little explained variance. Increasing trends in hypercholesterolemia are attributable to environmental changes affecting populations as a whole. Reducing obesity and smoking and moderating alcohol intake in entire populations should remain the primary strategies for lipid control.

The objective of medical screening is to identify disease in its preclinical, and therefore hopefully still curable, phase. This may have been an old quest in medicine but it became historically possible when at least four conditions were met: the availability of simple, valid and acceptable forms of tests, the discovery of effective treatments, the establishment of a theory of screening, and the wide access to health care. Five selected examples that illustrate the history of medical screening are reviewed: screening for psychiatric disorders in the United States army as it is one of the oldest screening programmes; screening for syphilis as it used one of the earliest screening tests; screening for diabetes as one of the first modern forms of mass screening; screening for cervical cancer using the Pap test as one of the greatest successes of screening; and screening for breast cancer by mammography as this offers a good opportunity to discuss the development of modern evaluation of screening programmes. The evaluation of the impact of screening on human health slowly progressed, from obvious changes in the vital statistics such as the decline in incidence of syphilis, to less obvious changes such as the decline in mortality of cancer of the uterus, to finally more subtle changes, such as the impact of mammographic screening on breast cancer mortality. Methods of evaluation had therefore to adapt, evolving from simple surveys to case-control studies and randomised trials. The history of screening is short, but very rich and mostly still to be written.

This study explored the genetic basis of the combination of extreme blood levels of HDL-C and LDL-C, a well-studied endophenotype for CVD, which has several attractive features as a target for genetic analysis: (1) the trait is moderately heritable; (2) non-genetic risk factors account for a significant but still limited portion of the phenotypic variance; (3) it is known to be moderated by a number of gene products. We exhaustively surveyed 11 candidate genes for allelic variation in a random population-based sample characterized for known CVD risk factors and blood lipid profiles. With the goal of generating specific etiological hypotheses, we compared two groups of subjects with extreme lipid phenotypes, from the same source population, using a case-control design. Cases (n = 186) were subjects, within the total sample of 1708 people, who scored in the upper tertile of LDL-C and the lowest tertile of HDL-C, while controls (n = 185) scored in the lowest tertile of LDL-C and the upper tertile of HDL-C. We used logistic regression and a four-tiered, systematic model building strategy with internal cross-validation and bootstrapping to investigate the relationships between the trait and 275 genetic variants in the presence of 10 non-genetic risk factors. Our results implicate a subset of nine genetic variants, spanning seven candidate genes, together with five environmental risk factors, in the etiology of extreme lipoprotein phenotypes. We propose a model involving these 14 genetic and non-genetic risk factors for evaluation in future independent studies.

Since 2013, my task for the American Journal of Epidemiology has been to continue the regular publication of “Snippets From the Past,” a section that George Comstock carried out between 1991 and 2005. In addition, I have served as the editor of a new section of the Journal entitled “Epidemiology in History,” which is dedicated to invited and unsolicited papers (1).

The analyses of observational longitudinal studies involving concurrent changes in treatment and medical conditions present difficulties because of the multitude of directions of potential relationships: past medication influences current symptoms; past symptoms influence current medication; and current medication is associated with current symptoms. In the context of a long-term study of non-randomized pharmacological treatment of schizophrenic relapse, we present an analysis of bivariate discrete-time transitional data with binary responses in an attempt to understand the transitional and concurrent relationships between schizophrenia relapse and medication use. A naive analysis does not show any association between previous medication and current relapse. However, we provide evidence suggesting that current treatment may impact current relapse for those who have previously taken medication, but not for those who haven't taken medication in the past. When univariate models are specified to assess these associations, the bivariate nature of the problem requires a choice of which response, relapse or medication, should be the dependent variable. In this case, the choice of relapse or medication as a dependent variable does matter. Hence, our results derive from models where both relapse and medication are treated as dependent variables. Specifically, we specify a bivariate log odds ratio for current relapse and current medication use and a separate univariate logit component for each of these outcomes. Each of these components contains transitional associations with previous relapse and medication. Such models represent extensions of univariate transitional association models (e.g. Diggle et al. (1994)) and correspond to bivariate transitional models (e.g. Zeger and Liang (1991)). We incorporate changes in transitional associations into the full-data parametric model for final inference, and investigate if these temporal changes are due to learning effects or the impact of drop-out. We also perform residual analyses and sensitivity analyses in the context of missing data patterns.

The occurrences and timing of reproduction-related events, such as menarche, first birth, and menopause, play major roles in a woman's life. There is a lack of comparative information on the overall patterns of the ages at and the timing between these events among different populations of the world. This study describes the variability in reproductive factors across populations in Europe, the Americas, Asia, Australia, and Africa. The study sample consisted of 18,997 women from 13 centers in 11 countries interviewed between 1979 and 1988 who comprised the control group in a World Health Organization international, multicenter case-control study of female cancers. All were surveyed with the same questionnaire and methodology. Overall, a typical woman in this study reached menarche at age 14 years and delivered her first live child 8 years later, at age 22. She was 50 years old at natural menopause and had had 36 years of reproductive life. The median ages at menarche varied across centers from 13 to 16 years. For all centers, the median age at first livebirth was 20 or more years, with the largest observed median (25 years) occurring in China. The median delay from menarche to first livebirth ranged from 5 to 11 years. Among the centers, the median age at natural menopause ranged between 49 and 52 years. In most populations, younger women had a first birth at a later age than did older women. This tendency was more accentuated in some populations. These results reveal, perhaps for the first time, the variability of reproductive histories across different populations in a large variety of geographic and cultural settings. Except for menopause, international variability is substantial for both biologically related variables (age at menarche) and culturally related variables (age at first birth). There is a generational effect, characterized by more variability of age at first birth and delay to first birth in the younger than in the older generations. Am J Epidemiol 1998;148:1195-1205

The New York City neighborhood of Harlem has mortality rates that are among the highest in the United States. In absolute numbers, cardiovascular disease and cancer account for the overwhelming majority of deaths, especially among men, and these deaths occur at relatively young ages. The aim of this research was to examine self-reported smoking habits according to measures of socioeconomic status among Harlem men and women, in order to estimate the contribution of tobacco consumption to Harlem's remarkably high excess mortality. During 1992-1994, in-person interviews were conducted among 695 Harlem adults aged 18-65 years who were randomly selected from dwelling unit enumeration lists. The self-reported prevalence of current smoking was strikingly high among both men (48%) and women (41%), even among highly educated men (38%). The 21% of respondents without working telephones reported an even higher prevalence of current smoking (61%), indicating that national and state-based estimates which rely on telephone surveys may seriously underestimate the prevalence of smoking in poor urban communities. Among persons aged 35-64 years, the smoking attributable fractions for selected causes of death were larger in Harlem than in either New York City as a whole or the entire United States for both men and women. Tobacco consumption is likely to be one of several important mediators of the high numbers of premature deaths in Harlem.

The relation between body mass index expressed as weight in kilograms divided by height in meters cubed (weight (kg)/height (m3)) and plasma total cholesterol was assessed in a sample of 11,389 US schoolchildren, aged 5-18 years, between 1984 and 1989. Among whites and Hispanics, a statistically significant nonlinear relation was found between weight (kg)/height (m3) and plasma total cholesterol. Mean cholesterol values increased exponentially above the 50th percentile of body mass index. Additionally, mean cholesterol levels as well as the odds of hypercholesterolemia (total cholesterol value greater than 180 mg/dl) were highest among children with weight (kg)/height (m3) values above the 95th percentile. Among black children, weight (kg)/height (m3) did not appear to be related to plasma total cholesterol. Possible explanations as well as public health implications of these findings are discussed.

The relation of serum retinol and carotenoids to airway obstruction was investigated in a population-based study. Respiratory symptoms and the pulmonary function of 83 white males were assessed in 1979 5 years after their blood had been drawn. Airway obstruction in 1979, defined as a forced expiratory volume in 1 second which was equal to or less than 75% of the forced vital capacity, was associated with a decreased 1974 level of serum retinol.