Harold Fred Dorn and the First National Cancer Survey (1937-1939): The Founding of Modern Cancer Epidemiology

Abstract

The development of modern epidemiology, particularly cancer epidemiology, is often seen as a post–World War II phenomenon. However, the First National Cancer Survey, conducted from 1937 to 1939 as part of the newly formed National Cancer Institute's initial activities, provided the first data on the occurrence of cancer in the United States. This project was directed by a young sociologist, Harold Fred Dorn. Through Dorn, many of the methodological innovations in sociology, such as the use of surveys and observational study designs, were incorporated into modern epidemiology. I examine Dorn's training and early career in the context of the First National Cancer Survey as a means of investigating the beginnings of modern epidemiology.

THE DEVELOPMENT OF CANCER epidemiology is often viewed as a post–World War II phenomenon, with investigations of the relationship between cigarette smoking and lung cancer fostering the formation and growth of the field.¹ According to this paradigm, modern epidemiology arose, in part, out of cancer epidemiology. Of particular importance are the 1950 case–control studies of Hill and Doll and of Wynder and Graham. The beginnings of cancer epidemiology, however, predate World War II. Although it might be argued that some activity in the early 20th century, such as Frederick Hoffman's report showing demographic contrasts in international cancer mortality, had no sustained influence, the effects of Harold F. Dorn's cancer epidemiology efforts, begun in the 1930s, have persisted into the current era.²

In 1933, Harold Fred Dorn received his doctorate in sociology and demography from the University of Wisconsin.³ In 1936, after completing postdoctoral studies at the London School of Hygiene and Tropical Medicine and gaining some experience in demographic analysis with one of the founders of modern American demography, Frank Lorimer, at the Population Council of America, Dorn found employment as a demographer and survey analyst with the US Public Health Service.⁴ The Works Progress Administration had given the Public Health Service an unemployment relief grant to undertake a survey of the health status of the US population.⁵ The National Health Survey of 1935–1936, predecessor to the current National Health Survey, was the largest noncensus survey undertaken by the federal government up to that time.⁶ A year later, when the National Cancer Institute (NCI) was created, Dorn transferred to the National Institute of Health's Division of Public Health Methods; he would direct the First National Cancer Survey of 1937–1939.⁷ The 70th anniversary of the survey is an appropriate time to assess its impact on the formation of cancer epidemiology in the 1930s.

I consider the activities of Dorn in relation to the activities of the survey. After a review of Dorn's background and training, I discuss the organization of the survey and its operation. I then consider some of its findings. Last, I discuss the impact of the survey, and of Dorn's activities regarding it, from the perspectives of the formation of cancer epidemiology and modern epidemiology.

EARLY LIFE AND TRAINING

Harold Fred Dorn was born on July 30, 1906, in Tompkins County, New York, one of 10 children born to Floyd Dorn, a dairyman, and his wife, Minnie Elizabeth Miller.⁸ After finishing high school, Dorn worked on the family farm for 2 years before attending Cornell University. He received a bachelor of science degree in rural sociology in 1929.⁹ He continued at Cornell for graduate training in rural sociology, receiving a master of arts degree in 1930. Dorn's thesis concerned a typical social survey of a rural New York county.¹⁰

Prior to 1910, during American sociology's formative period, social surveys comprised an enumeration of facts describing a community, neighborhood, or county.¹¹ They were commonplace; however, they offered little insight into sociological phenomena. Sample surveys developed in the interwar period as a means of understanding sociological relationships.¹² The empirical tradition of American sociology progressed with this development. In modern epidemiology, disease registries, exemplified by the US Surveillance, Epidemiology, and End Results System, are the present-day counterparts of social surveys—a complete enumeration of facts regarding a particular population.

In 1930, Dorn enrolled in the rural sociology doctoral program at the University of Wisconsin.¹³ The university's sociology department was directed by John Gillin,¹⁴ an ordained minister, who had received his doctorate in sociology from Columbia University in 1906 under the supervision of Franklin Giddings.¹⁵ Giddings, much influenced by Karl Pearson (the founder of modern biometry, creator of the correlation coefficient and chi-square test for 2×2 contingency tables, and the mentor of the founders of modern English epidemiology), was at the forefront of the empiricism movement in sociology; empiricists sought to quantify sociological phenomena and test hypotheses in a quantitative manner. Gillin played a leading role in the American sociology community's response to criticism that the field was “not scientific.” He was elected president of the American Sociological Society in 1926.

Soon after Dorn began his doctoral studies at the University of Wisconsin, Samuel Stouffer joined the department.¹⁶ Stouffer, an early advocate of sample surveys, received his doctoral degree at the University of Chicago under the supervision of Howard Ogburn, another student of Giddings.¹⁷ Dorn became one of Stouffer's first doctoral students.

Dorn spent part of the 1931–1932 academic year at the Scripps Foundation for Research in Population Problems at Miami University of Ohio.¹⁸ (Interestingly, that same academic year, Stouffer studied at the University of London; Figure 1.) His interest in demography flowered; he published his first demography paper (dealing with allocation of deaths among nonresidents of states and countries) in 1932 and a follow-up paper with Stouffer the next year.¹⁹ In 1933, Dorn completed his doctoral dissertation on the effect of urban–rural migration on mortality in Ohio, an outgrowth of his work at Scripps.²⁰ (At the time, there was no national vital statistics system in the United States, hence Dorn's use of a single state's vital statistics data.) The work was quickly appreciated by leaders in the field, with Edgar Sydenstricker, one of the chief architects of 20th century epidemiology, citing Dorn's work in his seminal work on health and the environment.²¹

FIGURE 1.

Relationships between Harold F. Dorn and some of the major figures in sociology, statistics, and epidemiology in the first half of the 20th century.

Note. Solid lines denote established pedagogical relationships or working relationships; dashed lines indicate relationships inferred but for which direct evidence is lacking.

Harold Fred Dorn, PhD, 1906–1963.

Source. National Library of Medicine.

Dorn subsequently obtained a Social Science Research Council fellowship to study at the University of London, University College.²² (Interestingly, Sydenstricker was a member of the council awarding the fellowship.) He attended classes and seminars by both R.A. Fisher and Egon Pearson.²³ There is no evidence that he interacted with Bradford Hill, noted English epidemiologist and statistician; however, since he referenced Hill's writings in his thesis and shared an interest in demography with Hill, Dorn likely had some contact with him too. Returning to the United States, Dorn joined the Works Progress Administration.²⁴ He was assigned to the National Resources Committee and was a key contributor to its report, “The Problems of a Changing Population.”²⁵ He also worked on the first National Health Survey, conducted in 1935 and 1936.²⁶ Stouffer was also involved in the survey's conduct.²⁷

Another future leader in cancer epidemiology, William Haenszel, also worked on the survey; we do not know if he had had any contact with Dorn. During this time, Dorn came into contact with the eminent demographer–sociologist Frank Lorimer and with the Population Council. He published papers on many topics, including demography and others more directly connected with his employment.²⁸

THE NATIONAL INSTITUTE OF HEALTH AND THE NATIONAL CANCER INSTITUTE

In 1937, Dorn was transferred to the Division of Public Health Methods (DPHM) at the National Institute of Health.²⁹ The division, newly organized, was directed by Joseph Mountin, a rising star in the Public Health Service, who had been involved with the National Health Survey.³⁰ In 1931, before his assignment to the survey, Mountin had directed the Office of Studies of Public Health Methods in the Division of Scientific Research within the Public Health Service.³¹ After the completion of the survey, in 1937, Mountin was assigned to direct the new DPHM; among his recruits to staff the division was Dorn.³² Comparison of the activities of the DPHM with the office Mountin headed prior to the National Health Survey suggests the effect of his recruits on the DPHM research agenda. There was little in the activities of the office that bore resemblance to those of the division. The former had a strong orientation to public health administration, whereas the latter was epidemiological. One of the division's first tasks was undertaking a national survey of cancer, the First National Cancer Survey.³³ The need for such data had been recognized in 1927, but a full decade would pass before the necessary data were collected and analyzed.³⁴

The First National Cancer Survey, conducted in the United States in 1938 and 1939, included cases (both prevalent and incident) that occurred from 1937 through 1939.³⁵ Results from the National Health Survey indicated the unreliability of self-reports of cancer. Some people with cancer had not been told they had it, and others were unwilling to acknowledge it; even among those who knew they had the disease and were willing to inform an interviewer, sufficient supporting medical information was often unavailable. The First National Cancer Survey therefore included only physician-diagnosed cases of cancer; every physician, hospital, and clinic in a given population area was surveyed.

For the survey, the United States was divided into three regions: north, south, and west.³⁶ The proportion of the population in cities with 100 000 persons or more relative to that for the entire region was approximately the same for each of the three regions. Since medical students were employed as medical record abstractors, each area had to have a medical school, limiting the survey to metropolitan areas.³⁷ The 10 registries selected, including the counties that each comprised, are shown in Table 1. Two registries, New Orleans and Philadelphia, did not include any suburban counties. San Francisco included both the city of San Francisco and Alameda County (including the city of Oakland), but no suburban counties.

Which counties were included in the survey was determined largely by the availability of population statistics necessary for the calculation of incidence and prevalence rates. The registries selected had such data available in the form of census publications. (This requirement also limited the survey analyses to people residing in the registries.) The surveyed population included 13 million people—approximately 10% of the overall population of the United States and 18% of the metropolitan population of the country.

The survey collected information on all cases of cancer (except for leukemia and Hodgkin's disease) occurring at any of the facilities canvassed. If a local survey office did not receive a physician's response to a request for information, a medical student visited the physician's office and requested the information in person. For each metropolitan area, endorsements of the survey by the state and local health departments and local medical and hospital societies were sought.

The same methods were used in each of the metropolitan areas surveyed. Information collected included demographic information (name, address, gender, race, age, vital status), date first attended for cancer, date last treated for cancer, date last attended, method of diagnosis, and if appropriate, the death certificate diagnosis—including who completed the certificate and whether an autopsy was performed.

The calculation of incidence rates in the survey's early data analyses was challenging. The last census conducted had been in 1930. Formulating county-specific estimates of population was not feasible during the 1930s; demographic and computational methods did not facilitate such formulation. By the early 1940s, however, provisional data from the 1940 census provided estimates of the US population.³⁸

In the overall investigation, 63 555 cases of cancer were identified, of which 48 043 occurred among residents of the designated metropolitan areas.³⁹ Barely half of these (26 085, or 54.3%) were reported as diagnosed during the study year, and some 4144 additional cases (8.6%) were ascertained from review of death certificates. For all newly diagnosed cancers, one fourth of all cases were reported by physicians with no involvement of a hospital or clinic. These results included skin cancers, the majority of which were reported by a physician alone.

A variety of interesting findings developed from the survey data, most of which focused on etiologic clues from the geography of cancer occurrence; papers were published for each of the various regions in which data had been collected. The first paper concerned the incidence of cancer in Atlanta, Georgia.⁴⁰ During the next three years, data for each of the other nine registries were reported, the last being Denver, Colorado.⁴¹ Dorn coauthored the initial report from Atlanta and authored the report for Chicago. He wrote none of the other registry-specific papers.

At least two papers dealing with the epidemiology of cancer of a specific site resulted from the survey, one on gastric cancer and the other on lung cancer.⁴² The former was presented at a meeting on gastric cancer held by the NCI in 1940.⁴³ Attendees included James B. Conant (Harvard), Louis I. Dublin (Metropolitan Life Insurance Company), Perrin H. Long (Johns Hopkins), James Ewing (Memorial Hospital), Thomas Parran (Public Health Service surgeon general), Carl Voetglin (NCI director), and Shields Warren (Harvard), among others. Although Dorn was the paper's senior author, he did not attend. Also noteworthy is that, except for Dublin, Selwyn Collins, who was Dorn's coauthor (and who presented the paper and was also in the DPHM), was the only person present with statistical training.

The paper Collins presented was sophisticated for 1940, including an examination of potential reporting biases. The ensuing discussion was significant for its inclusion of many points that became major themes in cancer epidemiology. Among the paper's commentators was Madge Thurlow Macklin of the University of Western Ontario, one of the founders of modern human genetics.⁴⁴ Macklin observed that some aspects of gastric cancer suggested by Collins et al., such as the socioeconomic gradient associated with its occurrence, had been previously noted. Macklin then commented:

We could study them carefully, tabulate their physical findings, study them for the onset or presence of the five main types of cancer, let us say, cancer of the lung, stomach, and rectum, and in the female, cancer of the breast and uterus too. We should have a complete record of dietary habits, of chronic respiratory diseases, of the amount of alcohol they consumed, of how much they smoked, of how many children the women bore, and whether or not they nursed them, etc. Then when they died we would have an autopsy and record all the pathological findings. The persons with gastric cancer could then be studied with respect to any one factor which might have been a cause of gastric cancer, and could be compared with a control group of the same age and sex distribution that did not develop gastric cancer. Where would you get such a population? I would suggest that the government employees be the experimental population. You might make it a condition of employment that they enter into this scheme. It would be of distinct advantage to them for they would receive all their yearly or semiyearly examinations free, from experts particularly interested in special fields. Their climatic conditions would be more or less controlled, for all of them would be living in Washington; and all would be subjected to the same expertness of diagnosis. From such a study, you might derive much valuable information which cannot be obtained from a study of a population which has died of gastric cancer.⁴⁵

Walter Clement Alvarez, from the Mayo Clinic, followed Macklin's comments with the observation, “I think we ought to begin soon to gather statistics to show whether the increase in the amount of smoking done by women is increasing the incidence in them of carcinoma of the upper parts of the respiratory and digestive tracts.” In reply, Collins noted, “A living study in which a group of people would be followed throughout life would be ideal. However, cancer is relatively rare and such a large number of people must be followed through to obtain statistically significant cancer rates that such a project is a very large undertaking.” The discussion of the paper then closed.⁴⁶

Three years later, Dorn published a paper on lung cancer occurrence in the United States.⁴⁷ He summarized the data from the 1937–1939 incidence survey, observing the increasing prevalence of respiratory cancer generally and lung cancer in particular among the older age groups. He noted the increasing mortality from lung cancer and the excess mortality among men compared with women. By contrast, overall, women tended to die from cancer (all sites) at a greater rate than did men. Dorn suggested the potential role of air pollutants in the etiology of lung cancer, but he then cast doubt on the hypothesis by noting the relatively low rates of lung cancer in Pittsburgh compared with other less polluted cities. He also noted the lack of variability in the occurrence of lung cancer in women among the cities surveyed even though there was considerable variability in the presence of air pollution.

In 1944, Dorn published the final paper reporting data from the First National Cancer Survey; it summarized the survey's findings.⁴⁸ Several demographic themes were discussed, such as male–female differentials in cancer occurrence, the pattern of age-specific incidence rates of cancers in different organ systems, the abundance of nonmelanotic skin cancer, and the impact of differences in access to health care among Blacks and Whites on the observed risk of specific cancers. The specifics of the individual rates themselves are not as noteworthy as the sophistication of the analysis and the general approach shown in examining cancer incidence data.

DISCUSSION

In this essay, I discussed the development, conduct, and analysis of the First National Cancer Survey. The survey was the largest epidemiological exercise regarding cancer undertaken in the United States up to that time. It was also one of the first activities in support of the recently constituted NCI. The circumstances surrounding the survey are important for an under standing of the development of modern epidemiology generally, and of cancer epidemiology specifically, for several reasons. First, the survey marked a major commitment to cancer epidemiology by the NCI. Whereas the National Health Survey of 1935–1936 was a Works Progress Administration–funded project to relieve unemployment during the Great Depression, the 1937–1939 survey was funded entirely by the NCI.

Second, the cancer survey established the precedents for conducting such studies. Since 1937, through two additional cancer incidence surveys and the establishment of the Surveillance, Epidemiology, and End Results System, the basic construct for cancer surveillance (i.e., the use of active surveillance, a focus on histological confirmation as an index of data quality, and the use of trained health care professionals as abstractors) has remained unchanged.⁴⁹ The addition of outcomes, as well as the increased precision afforded by information technology advances, began in the early 1970s; it did not undercut the work initiated in the 1937–1939 survey.

Third, the methodological reasoning required for the advancement of modern epidemiology was in place prior to World War II. The cohort and case–control observational study designs were developed by sociologists in the period 1910 to 1930.⁵⁰ Converse has described the development of population-based surveys as a means of obtaining the data needed in the execution of these designs during the early and mid–20th century.⁵¹ Dorn's training at the University of Wisconsin, and his subsequent involvement with The American Soldier project during World War II (supervised by his thesis adviser, Samuel Stouffer, at the Pentagon), underscore his familiarity with such methodological reasoning.⁵² In specific instances, it is possible that Dorn was unaware of some methodological advances. Regarding the gastric cancer paper presented in 1940 by Collins, however, it seems unlikely that Dorn would not have known of Macklin's comments on a paper that he had coauthored.⁵³ Significantly, in the discussion of the paper at its presentation, no one had contested what Macklin said. Indeed, the only concern that was voiced was the number of persons needed in such a study, presumably because of statistical power.

A related question arises in Dorn's use of veterans’ life insurance holders for his cohort study on cigarette smoking and lung cancer.⁵⁴ Veterans were, at one time, government employees, and the Veterans Administration provided the means for following a cohort of such individuals. Did the idea for that study originate with Macklin's comments? There is no evidence to support such a claim; the origin of Dorn's study will remain unknown until further evidence is brought to light.

Fourth, in reviewing Dorn's work with the First National Cancer Survey, we see the emergence of themes defining cancer epidemiology in the latter half of the 20th century, prior to World War II. The effect of the aging population on cancer incidence was already understood; Dublin, the statistician from Metropolitan Life Insurance, had observed in 1926 the decline in infectious disease mortality in the United States and its effect on increasing risk of cancer.⁵⁵ Omran would generalize Dublin's observation 45 years later under the rubric “The Epidemiological Transition.”⁵⁶ Cigarette smoking's role in the occurrence of respiratory cancer had been put forward prior to World War II, and the distinctive demographic correlates of specific cancers were the subject of discussion. Diet had been proposed as a cause of digestive cancers, an idea put forward by Hoffman⁵⁷ in the 1930s and subsequently refined by others, but it did not reach fruition until the past two decades. The relation between exposure to the sun and skin cancer was noted, as was the possibility of misdiagnosis and missed diagnoses in some races over others.⁵⁸ Indeed, the nonhereditary basis for most cancers was understood in the mid-1920s.⁵⁹

Last, the success of the cancer survey suggested the promise of cancer epidemiology in identifying groups at high risk for cancer; it also indicated that there were potential etiologic agents for specific cancers (such as sun exposure for skin cancer) and showed that such surveys were cost effective. A related aspect of the survey's success was the platform it provided to Dorn. During World War II, Dorn would direct the Medical Statistics Division in the Office of the Surgeon General of the US Army, which allowed him access to large-scale data analyses similar to those he would undertake as part of the Veteran's Life Insurance study.⁶⁰ By successfully completing the survey, Dorn was able to build an epidemiology program at the National Institute of Health immediately after World War II and to provide the leadership needed for the field to grow and prosper.

In 1947 and 1948, around the time of the Second National Cancer Survey, Dorn hired five statisticians from elsewhere in the federal government: Jerry Cornfield, Sam Greenhouse, Nathan Mantel, Jacob Lieberman, and Marvin Schneiderman.⁶¹ None of them had any prior experience with epidemiology, although within the next decade they would become leaders in the development of the advanced statistical methods epidemiologists would rely on for the next half century. The impact of Dorn on their careers was enormous. Their backgrounds were for the most part in surveys, an area of interest to Dorn because of his training in demographics. Yet it is difficult to imagine Dorn assembling this group in the absence of the 1937–1939 survey.

Although Dorn's work in the 1935–1936 National Health Survey may have brought him into contact with such leaders in epidemiology and demography as Sydenstricker, it was during the cancer survey that Dorn worked for and with Mountin. Although the initial paper reporting survey methods and results (for Atlanta) listed Mountin as first author,⁶³ it seems likely that it was written by Dorn. All of Mountin's other papers concern public health administration and do not focus on data.⁶⁴ The details of Dorn's interactions with Mountin are unknown. None of the Dorn papers at the National Library of Medicine provide any insight into Dorn's association with him. Mountin played an important role in the development of the Framingham Heart Study a decade later; Dorn could have influenced his conception of the study, and such involvement may have paved the way for Dorn's eventual move in 1957 from the NCI to the National Heart Institute.⁶⁵ There is no evidence, however, to support such assertions. The diverse influences on Dorn, and his subsequent impact on others, are shown in Figure 1.

In summary, Harold Dorn, a University of Wisconsin–trained demographer, worked on the first National Health Survey in the United States during 1935 and 1936. He then went to work on the newly formed, NCI-supported First National Cancer Survey from 1937 through 1939. This survey provided the first estimates of cancer incidence in the United States. It piloted techniques of active surveillance that would be used in two subsequent surveys (in 1947–1948 and 1970). The data analyses addressed such issues as the identification of high-risk groups, potential etiologic factors, and potential sources of bias. Discussions of the data derived from the survey suggest a sophisticated understanding in the pre–World War II public health community of cancer epidemiology and its potential on cancer control.

By the time of Pearl Harbor, many of the themes defining cancer epidemiology in the latter half of the 20th century (e.g., smoking as an etiologic factor, diet as a source of carcinogens for gastrointestinal cancers, bias in surveillance, geographical differences in cancer incidence, solar radiation as a carcinogen, and increasing occurrence of lung cancer) had been brought to the medical community's attention for further discussion and research. The methods used in furthering those themes, adopted from sociologists, had also been established by then.⁶⁶ The field of cancer epidemiology specifically, and of modern epidemiology generally, had taken form by the time of the nation's entry into World War II.