The Rise and Fall of Modern Medicine Page 7
Bradford Hill’s logical inference from statistical data – his demonstration of smoking’s causative role in lung cancer – was a masterpiece. The simplest of all medical statistics are, as already noted, the ‘vital statistics’ recording the unarguable event of death. When analysed over a defined period, they may display a characteristic pattern such as the rise and fall typical of an infectious epidemic. The collection and interpretation of such vital statistics constitute a form of scientific observation little different in its way from ‘observing’ the effects of disease in an individual. But statistics in this form can only report what has happened; they cannot produce any insights into why it has happened. For this it is necessary to move from simple observation to performing an ‘experiment’, which, like the Randomised Controlled Trial, is fairly straightforward and again essentially involves making a comparison. When the various aspects of the lives of a group of people with a particular disease are compared to those of another group without the disease, differences might emerge which, it might be inferred, could theoretically be the cause of the disease being studied.
In 1947 Bradford Hill, along with Edward Kennaway of St Bartholomew’s Hospital and Percy Stock, the government’s chief medical statistician, were asked by the Medical Research Council to investigate whether smoking might explain the ‘startling phenomenon’ of the fifteen-fold increase in the death rate from lung cancer in Britain over the previous twenty-five years. They were subsequently joined by Dr Richard Doll, who later recalled the division of opinion that reflected the prevailing views of the time:
Kennaway was particularly interested in the possibility of smoking being a factor, but I don’t think anybody else was. Bradford Hill certainly wasn’t particularly keen on smoking as a cause, nor was I, while Stock was particularly keen on the effect of general urban atmospheric pollution. I must admit I thought the latter was likely to be the principal cause, though not pollution from coal smoke which was terrible in those days but which had been prevalent for many decades and hadn’t really increased. Motor cars, however, were a new factor. If I had to put money on anything at the time I should put it on motor exhausts or possibly on the tarring of roads. But cigarette smoking was such a normal thing and had been for such a long time that it was difficult to think it could be associated with any disease.32
The main problem facing Bradford Hill was that 90 per cent of the adult male population were smokers, so clearly it would not be possible to implicate tobacco simply on the grounds of whether someone smoked or not. Rather, it was necessary to identify some biological phenomenon from which it would be reasonable to implicate tobacco. The most obvious is the ‘dose-response relationship’ – the higher the ‘dose’ of tobacco the greater the ‘response’, or incidence, of lung cancer. The statistical method was known as the ‘case-control’ study, where ‘every case’ of lung cancer was compared with a ‘control’ who was similar in every way other than suffering from some other disease. Theoretically, then, if the heavy smokers were disproportionately represented among the lung cancer group compared to the controls, one might infer that smoking was the cause of the disease. Though this seems straightforward in principle, in practice it is quite difficult, mainly because it is so difficult to ensure that the ‘cases’ and ‘controls’ are truly comparable. The investigation, therefore, had to do much more than record how much a person smoked; rather,
a range of potentially relevant factors had to be taken into account: the age, sex, urban or rural residence, and social class of the subject; occupational history; exposure to air pollutants; forms of domestic heating; and the history of smoking including for those who had smoked, the age of starting and stopping, the amount smoked before the onset of illness, the main changes in smoking history, the maximum amount smoked, the practice in regard to inhaling and the use of cigarettes or pipe.
Starting in April 1948, doctors in twenty hospitals in London notified Doll of any patients suspected of having lung cancer. Doll would then arrange for a ‘lady almoner’, as social workers were quaintly called in those days, to interview both the patient and two ‘controls’, one with cancer of the stomach or colon and one from another of the general medical and surgical wards with a disease other than cancer. He found that 99.7 per cent of the lung cancer patients confessed to smoking, compared to 95.8 per cent of those with ‘diseases other than cancer’. Such an observation by itself obviously proves nothing at all, but when the patients were subdivided into four groups depending on how much they smoked, ranging from ‘one cigarette’ to ‘fifty cigarettes’ a day, then it is possible to discern a trend of a higher risk of lung cancer among the heavy smokers (see page 64). Examining the final set of figures in the table, 4.9 per cent of lung cancer patients smoked fifty cigarettes a day, twice as high a percentage as the 2 per cent of controls – a subtle difference perhaps, but whichever way the smoking habit was investigated, either looking at the amount smoked every day, or the maximum amount smoked, or the total amount smoked over the years, and so on, the same pattern emerged: the greater the amount of tobacco consumed, the higher the risk. For Doll and Bradford Hill the conclusion seemed inescapable: ‘It is not reasonable, in our view, to attribute the results to any special selection of cases or to bias in recording. In other words, it must be concluded that there is a real association between carcinoma of the lung and smoking.’33
Smoking habits between patients with lung cancer and controls
.
(From R. Doll and A. Bradford Hill, ‘Smoking and Carcinoma of the Lung’, BMJ, 30 September 1950, pp. 739–48.)
We now know this only too well, but at the time things appeared very differently. Social habits had been incriminated in lethal diseases before, most notably drinking alcohol as a cause of liver cirrhosis, but this is a fate restricted to a minority of alcoholics. Smoking was different, as virtually everybody ‘indulged’. It was an intrinsic part of each and every social occasion and the offering of a cigarette an integral part of social (and often sexual) intercourse. Its incrimination in a lethal disease was thus a matter of the utmost gravity. The director of the Medical Research Council, Sir Harold Himsworth, strongly advised Bradford Hill and Doll that they should delay making their results public, as Doll subsequently recalled: ‘Himsworth said the finding was so important he did not think we should publish it until we had found it again’ (i.e., repeated the study and found the same results). They duly set to work, this time investigating lung cancer outside London (lest their findings might have been a fluke attributable to some unidentifiable ‘London factor’), but this proved unnecessary when a few months later an American study came to exactly the same conclusions.34
Doll and Bradford Hill promptly published their first study in the British Medical Journal on 30 September 1950, and its distinguishing features merit some comment. Firstly, the ‘dose-response’ relationship between smoking and lung cancer was very subtle and this could readily have been obscured were it not for the rigorous way in which possible sources of bias had been anticipated and eliminated. Secondly, it is impossible to convey, without publishing the paper in full, the lucidity of its exposition, so its weighty conclusion seems unarguable. Put another way, it is very difficult to appreciate the novelty of their paper. The source of reliable knowledge in medicine had always been in the biological and physical sciences. Now, in the face of considerable scepticism, statistical methods had ‘triumphantly’ (one can justifiably say) been demonstrated to be capable of providing a new and genuine insight into the nature of disease.
Nonetheless, it would take more than this for people to stop smoking. Bradford Hill looked around for some other way by which the link could be demonstrated and – in a masterly stroke of imagination – invented an entirely new method of investigation. The ‘case-control’ study he had just conducted was ‘retrospective’, in that it tried to make sense of something that had happened in the past, how the habits of a lifetime may have contributed to one disease in particular. But if the association betwe
en lung cancer and smoking was valid, he should get the same result looking forward, starting with a large number of men and women, asking them pertinent questions about their lives, including their smoking habits, and then sitting back and watching what happened to them over the years. They would die from diverse diseases, but the smokers should die in disproportionate numbers from lung cancer. The elegance of this ‘prospective’ or ‘cohort’ study is the simplicity of the open-ended question – ‘What do smokers die of?’ – to which time will inevitably provide an answer.
Bradford Hill chose as his cohort the 60,000 doctors on the Medical Register, who were likely to be reliable in answering the questions posed to them. There could be no more forceful way of bringing home to the profession the hazards of tobacco – which hopefully would then be passed on to patients – than by incorporating them in this scientific endeavour to provide further proof that smoking caused lung cancer. In November 1951 Bradford Hill wrote a letter to the British Medical Journal which was published under the headline ‘Do You Smoke?’:
Last week I sent a letter personally to every man and woman on the Medical Register of the UK asking them to help me. I asked them to fill in a very simple form about their smoking habits.
This, I think, is a new method of approach. May I therefore repeat my appeal through your column? If every doctor, whatever his field of work, will spare only a moment or two this research can be founded on a firm basis and in time give, I believe, firm and important answers. I am, etc.35
In a short period, a mere two and a half years, Bradford Hill had his answer. Of the 40,000 doctors who replied to the questionnaire, 789 had subsequently died, a mere 36 from lung cancer. But when the smoking habits of the deceased were tabulated (see below), lung cancer was the only disease in which there was a clear dose-response relationship – the more tobacco smoked, the greater the death rate, rising from 0.48 per 1,000 doctors smoking 1g of tobacco daily, to 0.67 for those smoking 15g, to 1.14 for those smoking 25g or more, compared to those who had died from ‘all causes’, in whom there was no gradient with increase in smoking habit.36
Mortality rate per 1,000 male doctors in relation to the most recent amount of tobacco smoked
(From R. Doll and A. Bradford Hill, ‘The Mortality of Doctors in Relation to Their Smoking Habit’, BMJ, 26 June 1954, pp. 1451–5.)
The final verdict on the statistical proof of the causative role of tobacco in lung cancer can be found in a reply from Bradford Hill to the following small item in The Lancet of 14 December 1957:
Yesterday the morning post brought an embarrassing revelation from my husband’s past. It was an innocent-looking letter from the Medical Research Council and ran: ‘Dear Doctor: In 1951 you stated that you smoked an average of three cigarettes a day . . .’
Three cigarettes a day! When I met him, around that time, thirty-three would have been a conservative estimate. The mean had hovered around there ever since, plus or minus a few standard deviations. We sat dumbfounded, our bacon cooling uneaten. I broke the silence first. ‘Why you hypocritical old . . .’
Then I was aware that this was superfluous. My husband stared before him, automatically buttering toast.
‘How could I,’ he began brokenly, ‘how could I say such a thing?’
But already it was obvious. My husband is a heavy smoker except when Giving Up Smoking. This happens three or four times a year and lasts anywhere from half an hour to two horrible weeks. During these interludes he is very virtuous, and impossible to live with. Clearly the questionnaire had caught him whilst he was Giving Up Smoking or, more accurately, tapering off.
There it is. Three cigarettes a day. It makes you think. I mean, statistical methods are so reliable these days. Isn’t it appalling that they have to depend on people?
Bradford Hill’s reply was couched in his customary elegant prose:
Sir: I must hasten to correct any marital disharmony that in my innocence I may have promoted. If in November 1951 the husband of your correspondent was temporarily smoking three cigarettes a day, then he truthfully answered the question I put to him. For I had asked him, clearly and deliberately, what are your present habits? Before popping the question I had realised that the present is not invariably a certain guide to the past (or future), that manic and depressive phases may succeed one another, but I had also realised, as alas, your correspondent does not, that these ‘errors’ of classification would inevitably reduce not exaggerate any association between smoking habits and mortality that I might subsequently find. In short the observation in this enquiry that mortality from lung cancer in ‘heavy’ smokers has been some 20 times the rate in non-smokers understates the facts of life. A pity but there it is.37
In fact ‘twenty times’ turned out to be just about right. In 1993 Sir Richard Doll, during a special celebration to mark his own eightieth birthday, summarised the results of the famous Doctor Study forty years on. Almost half – 20,000 – of the doctors who had answered the original questionnaire back in 1951 had died, of whom 883 had succumbed to lung cancer. There is a memorable simplicity in the final conclusion. Those smoking twenty-five or more cigarettes a day have a twenty-five-fold increased risk of lung cancer compared to non-smokers.38
Bradford Hill’s twin achievements of 1950, demonstrating the curability of tuberculosis and the preventability of lung cancer, are impressive enough on their own account, but the true significance – which became ever more apparent as the years passed – was even greater. He was naturally – if, as ever, modestly – conscious of the nature of his legacy, which he discussed in two public lectures fifteen years later in 1965, which might justly be considered his apotheosis: ‘Reflections on the Controlled Clinical Trial’ and ‘The Environment and Disease: Association or Causation?’39,40
We start with ‘Reflections’. ‘It is not far off twenty years since the MRC published the results of the [first] trial of streptomycin that set off the population explosion of clinical trials,’ Bradford Hill observed, adding, ‘over the last twelve months alone they have extended from a treatment for herpes simplex to a low-fat diet in myocardial infarction, from drugs in the treatment of alcohol withdrawal syndrome to prophylactic penicillin for comatose patients.’ The popularity of the randomised controlled trial obviously lay in its unique ability to provide answers to the sort of questions that doctors ask themselves every day – ‘Does this treatment work better than that?’ – but crucially the questions were posed and resolved in a manner – the experiment – almost synonymous with science itself. The RCT thus came to be seen as the only ‘scientific’ way of resolving these questions and so, almost by definition, superior to any other form of acquiring knowledge, such as ‘clinical experience’. In this way the RCT became the ‘dominant’ discourse of post-war medicine. This, as Bradford Hill acknowledged from his own personal experience, was not necessarily a good thing, as statistics have an equal, or even greater, capacity to variously mislead, obscure or in some other way subvert the truth as to clarify it.
I am faced with trials [of drug treatment] on such an ill-defined or undefined pot-pourri of patients that I can but hopelessly speculate on who got what, when and usually why. These poorly conducted trials not only tell us nothing but may be dangerously misleading – particularly when their useless data are spuriously supported by all the latest statistical techniques and jargon.
Such aberrations apart, Bradford Hill maintained that there was simply no alternative to the RCT in evaluating new treatments and challenging the efficacy of the old.
Nonetheless there have been dissenting voices, particularly recently, about the validity and especially the trustworthiness of the conclusions from such trials. They are, it is argued, insufficiently sensitive to variations in the range of symptoms of disease and thus the responsiveness to treatment. There have been many instances when they have produced the ‘wrong’ result, which was subsequently overturned, but not before the powerful influence of the original false verdict had misdirected medicine down a b
lind alley, often for decades. There is concern, too, about the habit of aggregating the results of many trials to produce a definitive verdict, as if numbers alone could cancel out the falsehoods inherent in poor scientific data. One observer has described this as ‘A new kind of alchemy . . . arcane, esoteric and mesmerising, that promises not to turn base metal into gold but to transmute statistical sow’s ears into scientific silk purses.’ Regrettably not all those who have followed in Bradford Hill’s footsteps have been gifted with the same degree of intelligence or fastidiousness and so sometimes – perhaps even often – the ‘clinical wisdom’ of doctors assessing the efficacy of treatment based on their own personal experience may, after all, be a better guide to medical practice than the ‘objectivity’ of the clinical trial.41
In the second lecture given by Bradford Hill in 1965, ‘Environment and Disease: Association or Causation?’, he elaborated with his customary lucidity on the importance of his discovery of smoking as a cause of lung cancer. This had, after all, established a precedent of the utmost importance, which naturally raised the question as to how many other common diseases – strokes, heart diseases, diabetes and so on (the list is virtually limitless) – might similarly be caused by some aspect of the environment or an individual’s ‘lifestyle’ which, if identified and modified, would prevent it.
Possible clues were assiduously sought in thousands of studies looking for ‘something’ to distinguish patients with a given disease from healthy controls. Inevitably they turned up interesting observations. For example, people with multiple sclerosis are more likely to be cat lovers, and those with cancer of the pancreas drink more coffee than average, and so on. Given the large number of different diseases and the numerous measurable aspects of an individual’s lifestyle, it is possible to generate a virtually infinite number of hypotheses about causation.