Jump Over Left Menu
Large Scale Prospective Medical Surveys
Richard Peto and M.C. Pike
Lung cancer is much commoner among cigarette smokers; deep vein thrombosis and pulmonary embolism (serious blood clot) is commoner among women who use the pill; cirrhosis of the liver is commoner among heavy drinkers. Work as well as pleasure can be harmful: nuns have an increased risk of breast cancer, lung cancer is commoner in certain occupational groups in gasworks and bladder cancer is commoner among persons whose job involves exposure to certain chemicals (benzidine and β-naphthylamine). These discoveries, and others like them, while not necessarily of immediate practical use, can make a substantial contribution to controlling these diseases.
The establishment of an association between a certain factor and a certain disease does not prove that that factor causes that disease, but it may lead to research that will. For example, the observed association between cigarette smoking and lung cancer led to work that showed that certain cigarette smoke extracts are highly carcinogenic (cancer producing) for laboratory animals. This is an important part of the evidence that cigarette smoking is a direct cause of lung cancer, rather than, as Eysenck (1965) had suggested simply being commoner in the 'type' of person who develops lung cancer. On the other hand, an observed association between a factor and a disease may not be because the factor causes the disease but rather because both are related to some other factor. For example, a negative association has been established between oral contraceptives and peptic ulcer. This, however, did not show that the pill protected women from peptic ulcers but rather that women with peptic ulcers had less sexual activity and hence less need for the pill. Care is clearly needed.
Once a possibly important association between a certain factor and a certain disease is suspected, we will wish to determine its correctness especially after allowing for the possible association of both the factor and the disease with nuisance variables such as age or sex. A quick way of doing this is often to do a retrospective (or case-control) study. If the factor causes the disease, then the factor is more likely to be found in persons with the disease than in persons without the disease. Thus, we take a group of patients with the disease (cases) and for each case we choose another person or persons (controls) without the disease who is identical in all 'relevant' respects (e.g. age, sex, marital status, social class) to the case; if the factor is significantly more frequent among the cases than among the controls, this is evidence that it is possibly involved in the aetiology of the disease and that the association warrants further research. This case-control method is an extremely powerful tool once the art of selecting proper controls is mastered, and it provides fairly rapid answers to a set and strictly limited question.
For example, soon after the introduction of oral contraceptives, reports of women who had suffered a deep vein thrombosis or pulmonary embolism while on the pill began to appear. This disease is, however, also known to occur in women who do not use oral contraceptives and so this did not provide really valid evidence of association. These reports did pose a set and limited question, however, and a number of retrospective studies were set up to answer it. In particular our colleagues Doll and Vessey (then at the Medical Research Council's Statistical Research Unit) investigated married women aged 16-40 years admitted to 19 large hospitals during 1964-1967 with deep vein thrombosis or pulmonary embolism without evident predisposing cause. The results of this study (see Table) suggest that the risk of these diseases is substantially increased by the use of the pill (Vessey and Doll, 1969).
Similarly (and historically earlier), the association was established by this case-control method between the smoking of tobacco, especially cigarettes, and lung cancer (Doll and Hill, 1952).
Deep vein thrombosis or pulmonary embolism (Vessey and Doll, 1969) - each patient was matched with 2 controls.
|Diagnostic Group||Oral Contraceptives|
The much more general questions: What is the effect of oral contraceptives on morbidity from all diseases? and What is the effect of smoking on mortality from all causes? cannot, however, be answered by this approach. We need instead to do prospective studies, that is, find out for a large number of persons whether or not they use the factor under study, and then keep them under surveillance for some time to see what happens to them. These require far more time, subjects, money and effort than case-control studies and are not entered into lightly. Three such studies are being conducted in our department at present.
(1) The Medical Research Council is sponsoring a study in which it is intended to follow for up to 10 years 15,000-20,000 women attending Family Planning Association clinics who are using a variety of methods of birth control. This will permit us to obtain a balanced view of both the good and the bad effects of the various methods of contraception. This study was begun on a pilot basis in May 1968, and to date it has recruited over 12,000 women.
In addition to asking about contraceptive practices, much general information and a full medical history is recorded on each woman on entry to the study. Subsequently, morbidity (illness) is measured by noting all hospital attendances. The results of cervical smears, and the outcome of each pregnancy (planned or unplanned) are also recorded, as are any changes in contraceptive practice. All this information is stored on magnetic tape at ATLAS. The tapes are updated monthly and analysed using our SIGMA 2 link.
So far, only 2 new conditions have been found to be associated strongly enough with contraceptive practice to merit further study (one for the pill and one against). However, as the study progresses many more will probably come to light. Analysis of these will be complicated by the major differences that exist between women who choose different contraceptive methods: women using 'the pill' smoke about twice as much as do women who are using other methods, and women with previous histories of blood-clotting disorders are very likely to avoid using the pill. Although we will be able to make allowance for many of these, we will probably eventually generate several questions which will have to be answered by special case-control studies.
(2) In 1951 a questionnaire was sent by Doll and Hill to all doctors in the United Kingdom asking about their smoking habits. Their answers allowed us to classify nearly 35,000 male doctors according to age, type and amount of tobacco smoked, and whether they were currently smoking or had given up. The survivors amongst these men were questioned again about their smoking habits in 1957 and 1966; and we are soon to send out a final questionnaire, bringing our follow-up to 20 years. Information about the date and cause of death of the doctors who die is notified to us by the Registrars General of the United Kingdom, the General Medical Council and the British Medical Association (this is one of the reasons for choosing doctors as the study population). Our follow-up at the time of the third questionnaire showed that as of 1 November 1965 (14 years after the study began) we know for all but 21 of the 34,445 men (0.06%) whether they were alive or dead, and if they had died their date and cause of death.
The replies to the questionnaire (age, smoking and inhaling type, change of habits, date of starting smoking, date of stopping smoking, date and cause of death) have all been coded and are on magnetic tape at ATLAS. Updating of the data on deaths is done on a regular basis (so far nearly 10,000 men have died).
This prospective study has allowed us to investigate the relationship between smoking and all diseases. Some of the main findings are:
- The death rate from all causes amongst cigarette smokers smoking 25 or more cigarettes a day is more than 60% greater than among non-smokers, with a fairly steady increase of rate with amount smoked (doctors smoking 1-14 cigarettes per day had a 25% excess). This excess is more marked at younger ages so that the risk of dying in the years from age 45 to 55 (given that one is alive at age 45) is 1 in 27 for a non-smoker and this increases steadily to 1 in 10 for a smoker of 25 or more cigarettes per day.
- When cigarette smokers stop smoking the difference between their death rate and that of non-smokers decreases steadily. This effect is found at all ages.
- Smoking is related not only to lung cancer but also to coronary disease, chronic bronchitis, upper respiratory cancers, peptic ulcer and possibly some other diseases. (For an up-to-date report, see Royal College of Physicians, 1971.)
(3) Persons with chronic bronchitis suffer from persistent phlegm, recurrent infections and airways obstruction. These appear together so uniformly that it is difficult to decide which causes which, and it is probably important to know this for effective interference with the disease process. Ten years ago it was supposed that irritation from cigarettes and air pollution (both causes of chronic bronchitis) produced extra phlegm, that this was a good environment for infections to flourish in, and that just as an infected wound can leave a scar, so these infections scarred and obstructed the fine airways of the lung. One way to investigate this proposition is to measure rates of change of lung obstruction and relate them to current levels of phlegm, infection and obstruction. A prospective study doing this was started in 1961. Over 1,000 London working men have had these variables regularly measured twice a year over a 7-year period. The complete data are on magnetic tape at ATLAS and preliminary analysis has revealed that this postulated causal chain is entirely wrong. Infection docs no permanent damage, but merely flourishes when permanent damage has occurred. New ideas arc now required. The study was under way just after the Clean Air Act began to be enforced in London; the production of chronic phlegm of the men in the study changed dramatically - they stopped getting worse (what was expected) and actually began to improve.
Large scale prospective studies of the type illustrated here have been made much more manageable by the availability of fast computers with a reliable large backing store, and it has been possible to analyse the results in a much more detailed and useful way. Analyses involving the interaction of many variables can be done and, in particular, the effect of stopping smoking has now been clarified.
These studies have been greatly aided by being able to make use of the really excellent ATLAS program advisory group and from the kind help (in particular from D. G. House) in times of urgency, in gaining ready access to ATLAS.
1. Doll, R., and Hill, A. B. (1952). A study of the aetiology of carcinoma of the lung, British Medical Journal, 2, 1271.
2. Eysenck, H. J. (1965). Smoking, Health and Personality, London: Weidenfeld and Nicolson.
3. Vessey, M. P., and Doll, R. (1969). Investigation of relation between use of oral contraceptives and thromboembolic disease: a further report, British Medical Journal, 2, 651.
4. Royal College of Physicians (1971). Smoking and Health Now, London: Pitman.