Article Text


From cross-sectional survey to cohort study
  1. Elsebeth Lynge
  1. Elsebeth Lynge, Institute of Public Health, University of Copenhagen, Denmark; elsebeth{at}

Statistics from

Before the Second World War, case reports had indicated that working with certain chemicals might lead to the development of cancer. After the war, these occupational risks were investigated more thoroughly. Arsenic had long been suspected to be carcinogenic, and in 1945 the British Factory Department set up a committee to investigate the possible relationship between arsenic exposure and lung cancer. A Bradford Hill and E Lewis Faning investigated mortality among workers in a factory producing arsenic powder. Their report published in the British Journal of Industrial Medicine in 19481 is the first description, I have come across, of how to undertake a proper historical cohort study of an occupational group. According to the authors, information required for the study included: “(a) a list of the male employees, divided according to their occupation, for each year from 1943 … back to 1900; (b) the approximate dates of birth of these men so that the age constitution of the population at risk could be computed at different points of time between 1900 and 1943; (c) information regarding all deaths occurring in that period of time”. Therefore, “[g]iven these data the recorded numbers of deaths from different causes could then … [be] compared with the deaths which would have been expected to occur in a population of the stated size and age constitution if it had been experiencing during the passage of years the national, or local, rates of mortality”. Unfortunately, these required data were not available from the arsenic factory being investigated, so Hill and Faning instead undertook a proportional mortality study equally well described in their four-page paper.

Richard Doll was the first author to publish a historical cohort study in the British Journal of Industrial Medicine. His paper on causes of death among gas workers was published in 1952.2 He followed male pensioners from a large London gas company from 1939 to 1948 for each year, counting the number of men in each 5-year age group (he was clearly an investigator who knew his cohort members): “[a] pensioner who was in receipt of pension before the beginning of a year and who was alive at the end of the year was counted as one unit; a pensioner who received a pension before the beginning of the year but who died during it … counted as half a unit …”. Doll acknowledged advice from A Bradford Hill in the preparation of the paper. Doll used the same methodology when he reported on the mortality from lung cancer among long term workers from a large asbestos works. These data were published in the British Journal of Industrial Medicine in 1955.3

Working along the same lines, R A M Case developed reference tables for the percentage of men expected to die from bladder cancer given that they started to be observed at a given age and at a given point in time. His tables published in the British Journal of Industrial Medicine in 19534 were a tool for “medical officers … to investigate whether or not there is any reason to believe that a particular industry … gives rise to tumours of the urinary bladder”. Case’s method became the analytical basis for the famous investigation of bladder cancer in the British dyestuff industry that he published together with colleagues in 1954 in the British Journal of Industrial Medicine.5 6

Less known is the use of Case’s method in the analysis of lung cancer in chromate workers, but this study is worth mentioning because it represents a nice turning point in the understanding of a workplace carcinogen. In the British Journal of Industrial Medicine in 1951,7 P Lesley Bidstrup published a clinical and radiographic survey of workers from the chromate producing industry. The study was initiated based on previous case reports on lung cancer in workers exposed to chromium. “One case of carcinoma of the lung was found among the 724 workers interviewed and radiographed…. This observation may be compared with the results obtained among men in Great Britain in 1949 by mass radiography units of the Ministry of Health … it is seen that 0.41 cases were expected, whereas, in fact, one was found”. Bidstrup rightly concluded that “the numbers are clearly too small for any definite conclusions to be drawn” and “it is hoped that a follow-up study over several years will provide data from which the true incidence of carcinoma of the lung in workmen in the chromate-producing industry in Great Britain may be assessed”.

Bidstrup acknowledged the assistance of Richard Doll in assessing the results of the radiographic survey. It was, however, in collaboration with R A M Case that he carried out a follow-up investigation of lung cancer mortality among the investigated workmen. The Bidstrup and Case study was published in the British Journal of Industrial Medicine in 1956.8 Of the 723 originally disease-free workers, 12 were known to have died from lung cancer between 1 November 1949 and 31 August 1955. However, the vital status and eventual cause of death were unknown for 217 workers who had left their employment during the follow-up period. A modification of the original Case method was therefore needed, including for the 217 workers only “the time that elapsed before they did leave”. This calculation resulted in 3.3 expected deaths, and the authors concluded that “an excessive mortality from carcinoma of the lung must therefore be admitted”.

In this way the methodology of the historical cohort study in the workplace setting was gradually established in the 1950s. A Bradford Hill, Richard Doll and R A M Case all contributed to its development. Case’s methodology was, however, more difficult to follow than that presented in the papers by Hill and Doll. It is therefore not surprising that it is the latter approach to occupational health research that has survived into the present. The historical cohort study is still the gold standard when it comes to assessing the possible carcinogenicity to humans of workplace exposures, and the data sources have spread beyond the original use of company records to, for example, union data,9 health insurance records10 and census data.11


View Abstract


  • Competing interests: None.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.