Falsification

A Half-life of Burch

5. Falsification

The timeline now returns to 1978. Apart from coronary heart disease, Burch’s work in the next few years would focus on his precipitator model of lung cancer and the new randomised, prospective studies of smoking and health. His encounters with the anti-smoking establishment had persuaded him that Doll and his school sought only to confirm their hypotheses, not to refute them. To a man of Burch’s scientific standards, the touchstone which distinguished science from pseudoscience was the search for falsification, and that is his dominant theme during this period.

Whitehall

The year 1978 saw the publication of results from the Whitehall study, the first randomised, controlled study of smoking and health, begun in 1968 by Geoffrey Rose and colleagues. From a pool of civil servants who had previously been studied they selected 1445 middle-aged male smokers at high risk of smoking-related disease and assigned them at random to one of two groups. The 714 men in one group were intensively counselled by a doctor to give up smoking, and they did. Depending on how you correct for men who dropped out of the study, 36% or 58% had given up smoking three years later. The other 731 men were left undisturbed, though smoking rates were falling at this time and 14% had given up three years later. Despite randomisation, those who gave up in both groups tended to have other characteristics in common.

Rose and Hamilton decided on a sample size of about 1500 because they wanted to be able to specify the degree of confidence that could be placed in their results. Any randomised experiment can come up with a freak result, and statisticians distinguish between Type I and Type II errors (‘an alarm without a fire’ and ‘a fire without an alarm’). They expected, on the evidence of an earlier study of postmen, that the intervention groups would have a significantly lower death rate than the normal care group. On that assumption, the probability of a Type I error would be 5% and of a Type II error 20%, standard levels in the soft sciences.

Their expectations were not fulfilled. After ten years, 98 of the intervention group and 94 of the normal care group were dead, a statistically insignificant difference. A breakdown into different causes of death would also have been insignificant and was not given.

Rose already knew, because everyone knows, that smoking is bad for your health. Therefore a Type II error must have occurred. The fact that more of the normal care group gave up smoking than predicted meant that the probability of a Type II error was greater than 20%, and this was the explanation he seized on. The probability that a test will avoid a Type II error is called its ‘power’, and as other randomised controlled studies began to appear with similarly negative results it became a refrain. “The test lacked power. The test lacked power.”

Sources

Progression of tumours and non-neoplastic disorders as manifestations of a common biological origin

During this period, Burch found a new platform in Medical Hypotheses. That journal, founded by David Horrobin in 1975, was designed to provide a home for radical and dissenting views which tended to get suppressed by the peer review system and therefore followed an unconventional policy of accepting papers on the basis of editorial judgement rather than assessment by referees. It was forced to return to the peer review system in 2010 after it published a radical and dissenting paper defending Peter Duesberg’s hypothesis that HIV does not cause AIDS. Burch’s first publication there, not in the tobacco archive, discusses four medical conditions in the light of Burch’s general theory of disease.

Two of the four conditions are familial polyposis and cervical cancer, cases where non-malignant growth sometimes but not always proceeds to malignancy. The other two are Parkinsonism and male baldness, which also proceed through defined stages. Burch thought he could show that all four are brought about by mutations to more than one growth control cell resulting in multiple forbidden clones, the first clone to appear producing the first stage of the condition and the rest the ensuing stages at a constant rate of mutation.

Familial polyposis is interesting for various reasons. It is plainly genetic – familial means ‘running in families’ – it strikes around the age of 20, which means there are few other factors to obscure the age-distribution, and is ‘multifocal’ (it consists of benign polyps all over the bowel wall with no single polyp appearing to be ancestral to the others). It sometimes progresses to malignancy, also multifocal. One of the reasons why it is widely held that cancer originates in a single mutant cell of the affected organ, despite contrary evidence of this kind, is that the tumours are generally identical genetically. Burch showed that on his forbidden clone hypothesis that argument fails, since on his view cancer originates with a mutation in the immune system. Cells with the corresponding mutation in the bowel wall might continuously arise in their hundreds or thousands and be continuously eliminated until the immune system fails.

Sources

Are 90% of cancers preventable?

Burch borrowed the title of this piece from the Lancet editorial which signalled the capture of that journal by the disciples of Richard Doll and occasioned Burch’s debate with Sharp.

Burch’s paper concerns a typically sloppy piece of modern epidemiology, a study of oesophageal cancer by A. J. McMichael. McMichael assembled data on smoking and drinking rates and levels of oesophageal and laryngeal cancer in Britain and Australia in the mid-twentieth century. For the two cancers and smoking he knew the breakdown by age and sex, but not for drinking rates. On that basis, he claimed to show that alcohol was a major cause of both cancers. Tables and diagrams showed that the ratio between the British and Australian drinking rates converged even as the equivalent ratios for the cancers converged. This did not seem to be true for smoking rates, although an anomaly in the convergence rates might be explained by cigarette smoke as a co-carcinogen, since both cancers are commoner in smokers: this was the issue in a brief correspondence involving Cherry and Forbes. The paper supplies nothing in the way of statistical analysis such as significance testing.

The medical correspondent of The Times was sufficiently impressed by the McMichael paper to write it up as definite fact in a ‘science report’. “Detailed analysis of trends in Britain and Australia at the Commonwealth Scientific and Industrial Research Organisation in Adelaide has shown…”

Burch was working on oesophageal cancer at this time, and had been analysing the British data. Although McMichael knew the age-specific death rates from 1910 and was concerned that both oesophageal and laryngeal cancer rates were rising in younger age groups, his analysis used age-standardised figures. Burch presented graphs of deaths from both cancers broken down by age groups (a technique he increasingly favoured). The trends over the course of the twentieth century showed no very obvious correlation with trends in drinking.

Sources

1980

Smoking out censorship

A controversy in New Scientist in 1980 involved Burch only marginally (he made a minor contribution reaffirming his views on smoking and referring to recent work of his), but it arose out of an affair with wider implications. That year saw the appearance of an important work of cigarette scepticism, The Causes and Effects of Smoking by the psychologist Hans Eysenck. Eysenck was a household name in Britain, mainly because of his popular paperback Know Your Own IQ and its sequels. He sometimes appeared on television and his views commanded far more attention than those of the little-known Burch. His book has little to say about the causes of cancer. After expounding Fisher’s views on the matter and Burch’s extension of them, he proceeded to consider the other leg of Fisher’s hypothesis, genetic influences on smoking, for which he had plentiful evidence, a small part being Seltzer’s work on the physical characteristics of smokers.

In response to this threat, Richard Peto circulated a Request to Journalists alleging that Eysenck’s book was “trivially wrong” and demanding that it not be given publicity. “It is therefore undesirable that articles asserting that experts disagree as to whether smoking causes lung cancer should appear throughout the public press (like the articles in 1975/6 when the same fallacious arguments were first propounded by Burch.” The remainder of the Request set out the case against cigarettes more or less as presented at Hammersmith, apart from a passage about the dose-response relationship allegedly based on the 1978 rethink. A section headed “Benefits of stopping smoking” illustrates Peto’s tone:

It is impossible to forgive Eysenck for holding forth at length about inhaling, trends in mortality and so on among British doctors without taking the trouble to seek out the definitive report on that study which was published in 1976 (Doll & Peto, Brit. Med. J. 2, 1525-36); a simple telephone call to any serious scientist remotely connected with this field would have led him to this paper. All the points about trends, inhaling, etc. which he repeats from other authors are replied to there. Contrast the data on relative decrease in smoking and in lung cancer reported there with Eysenck’s conclusions from preliminary analyses of this very study.

This is revealing of how Peto wanted the matter to be seen. His Request cites no research that he was not involved in himself: his study, his analysis, his conclusions are “definitive”. That, indeed, is how many influential people saw him – Todd, Fletcher, Sharp among others. The unanswerable question is how he saw it himself.

The editor of New Scientist rose to the occasion. A leading article, while not really disputing the facts of the matter as expounded by Peto, denounced his Request as “character assassination – for what else are his implications that Eysenck will do anything to sell a book or that he is not an ‘expert’?”

However

The evidence that smoking causes cancer is very strong indeed. It is, for example, far more convincing than Cyril Burt’s evidence – on which Eysenck relied for years – that genetics play the major role in determining intelligence. The evidence is powerful enough to lead legislators, justifiably, to think of actions they can take to discourage smoking… But legislators should – and do – accept a lower level of proof than scientists: they must play safe.

The Cyril Burt affair is worth considering in this context. Burt, an educational psychologist and IQ researcher, was widely believed at this time to be guilty of academic fraud. The charges concerned his implausibly constant estimate of the heritability of intelligence to two decimal places in numerous papers based on unpublished data. They were first levelled after Burt’s death by Leon Kamin, sensationalised by the left-wing journalist Oliver Gillie in the Sunday Times, repeated in the official biography by Leslie Hearnshaw, and further publicised in The Mismeasure of Man by Stephen Jay Gould. Kamin and Gould were Marxists (and members of an informal group which also included the tobacco abolitionist Robert N. Proctor) who made no secret of loathing any use of science which might legitimate the inequalities of a class society.

Eysenck had been a student of Burt’s and was one of his few defenders, denying that he was guilty of anything more than carelessness. This may, after all, be the truth. Burt’s estimate of heritability holds up well in the light of work conducted in the age of the genome, and the most damaging accusation, that he invented fictitious research assistants, is now known to be untrue. Eysenck was himself an object of vilification by many on the left, to the extent that his children were obliged to change their surname, so unpopular was it in some circles. His work on both intelligence and smoking feature in a many a left-wing exposé of the political nature of the scientific enterprise and it is a general truth that political activists tend to be environmentalists in the broadest sense of that word. Burch certainly saw the connection between the two issues. As he wrote to Congressman Preyer, “environmental determinism is so entrenched in contemporary medicine and genetic interpretations are so distrusted (recall the IQ controversy) that the prospects appear dim.”

Another who made the link was the evolutionary biologist William Hamilton. He was not a smoking sceptic and there is no reason to think that he and Burch had heard of each other, but his views are interesting and deserve to be more widely known: they are quoted in an appendix.

Sources

Pathology, inference and carcinogenesis

The paper of this name is a full-length exposition of Burch’s theory of cancer and growth control, the longest accessible free online, with a detailed account of the mathematics involved and the physiological motivation of the theory. The reader who thinks of tackling it should be warned that most of the biological material is hopelessly dated and that the details cannot possibly be correct. For instance, Burch devotes a section to familial polyposis, an inherited condition in which benign polyps sometimes, but not always, progress to cancer of the colon. From analysis of the age distribution, he deduced that three forbidden clones are at work, the first of which produces benign growth, the second malignancy and the third a terminal event resulting in death. After Burch’s death, Bert Vogelstein identified genes on three different chromosomes in the affected cells which undergo mutation as the disease progresses. They are now known to be tumour suppressor genes, a mechanism which was unknown in Burch’s day. They have no known connection to the immune system, although its role in carcinogenesis still seems not to be fully understood. But the real question is whether the striking patterns Burch detected, so reminiscent of the order emergent from disorder which he knew from particle physics, are the clue which may yet unify medicine with physics.

Sources

Stochastic models of stomach cancer

In 1979, K. Manton and others published a hypothetical explanation of how cancer of the stomach originates in a multi-stage process involving random events in the cells of the stomach. They subjected it to statistical tests of significance and compared its performance with other models of stomach cancer – including that of Philip Burch in The Biology of Cancer, which they also subjected to significance testing. There was a good measure of common ground between Manton and Burch on stomach cancer. Both held that the disease takes three forms, one found in men and two in younger and older women. The age curves of these three forms all conformed roughly to a power of 6 or, as Burch held, 7. The appropriate statistical test (measure of goodness of fit) was chi-square.

Chi-square is a comparison of observed values and expected values. In this case, you convert actual levels of stomach cancer at different ages into standardised totals and write them down as a column of figures. From the equation representing your model, you calculate expected values and write them down in a second column. The chi-square formula then gives you a measure of the similarity of observation and prediction, the goodness of fit.

This was undoubtedly a fair-minded, well-designed evaluation of Burch’s general approach to cancer. The data, the general treatment of it, and the appropriate statistical test were not in dispute. Manton and his colleagues held posts at American universities far outside Sir Richard Doll’s network of influence. Burch’s model of stomach cancer failed their test for various plausible values of his parameter k.

As always when somebody engaged with his work, Burch riposted and Manton riposted to the riposte. Manton, said Burch, had tried out various alternative values for his parameter k (the mutation rate) but not alternative values of his parameter S (the proportion of the population at apparent risk). Burch had applied the same test to the same data for alternative values of both k and S and some of them were statistically significant. The values giving the best fit were not the same as those supplied in The Biology of Cancer, which he still preferred for other reasons.

However, said Burch, he rejected the test which his model had just passed. The reason was the parameter λ, the latent period, which also ought to be given various values in the test. Unfortunately, Burch could see no general pattern to the value of λ and could therefore not specify plausible alternative values to the one he favoured. Any assessment of his model as against others must involve other considerations: these he supplied in the form of four reasons to suppose that the events leading to cancer took place, not in the affected organ, the stomach, but in forbidden clones elsewhere in the body.

Though Manton’s paper and Burch’s reply are heavy on mathematics, the real issue is scientific reasoning. Manton had charged Burch with being like a navigator who could not calculate his latitude accurately. Burch’s answer is that he knew his latitude very well, but alas he had no way of calculating his longitude.

Manton, in reply, objected that the parameter S ought not to vary at all, as it was supposed to be the proportion of the population which is susceptible to stomach cancer. This was indeed the meaning of S, but as Burch stated many times elsewhere, S was subject to diagnostic error, which he had reason to believe was substantial. He did not get the chance to repeat himself here, as the rules of journal debate gave Manton the last word.

Sources

Individuals transplacentally exposed to maternal smoking may be at increased cancer risk in adult life

The campaign against smoking in pregnancy continued throughout this period, and anti-smoking researchers continued to seek evidence of its harmfulness. A paper by Richard B. Everson (Lancet, 19 July 1980) is printed under the rubric ‘hypothesis’, and that is what it is. It describes a number of reasons to suspect that pregnant women might induce cancer in their children by smoking and, in somewhat vague terms, proposes possible tests of the hypothesis. Circumstantial evidence for the hypothesis mainly involved animal studies in which carcinogens crossed the placenta and, of course, the alleged role of cigarette smoking in low birthweight.

Burch wrote on 9 August to observe that the proposed retrospective study would be worthless because of the self-selection problem. A better idea would be a randomised, controlled study like the Whitehall Study, which had shown no effect of reduced smoking on death rates. He told Seltzer (11 August):

Having got the impression that the Lancet had put me on their black list I recently carried out a simple experiment. I addressed a covering letter to the Editor directly and the enclosed letter for publication which will probably appear in the Lancet of August 9. Anyway it is a relief to find that the ban is not absolute and short, simple letters might be more influential than we generally allow.

Sources

Smoking and lung cancer: tests of a causal hypothesis

By 1980, Burch had proved to his own satisfaction that the Doll-Hill account of how cigarette smoking causes lung cancer was untenable and that the only possible causal role for cigarettes could be to precipitate lung cancer at the end of a long sequence of genetic mutations. In the 1980s he published four papers testing that hypothesis, of which this is the first.

He continued to assume a linear dose-response relationship. For that reason he analysed data relating to the entire population of Britain, or England and Wales, since the relevant factor was how many cigarettes were smoked regardless of which people smoked them. Figure 1 of the paper shows how any particular linear dose-response relationship can be expressed as a single quantity D, the number of cigarettes required to double the number of lung cancers relative to the number there would be if nobody smoked. The rest of the figures plot changes in cigarette consumption against changes in the lung cancer rate.

In between comes a page of stiff algebra showing how D can be estimated from that information. In brief, Burch begins by defining D in terms of a function R(x,ti) with x number of cigarettes as input and the corresponding number of excess lung cancers as output. The parameter ti is time, age in years, and does not come into the calculation. Six lines of algebra show that if we know D, we know the value of R for any x, and three more that D, whose actual value is unknown, can be estimated from reported changes in x and R. Burch symbolises the changes with the calculus expressions Δx and ΔR/R but calculus is not involved in the estimation of D, only algebra.

Figures 2 to 10 show that an observed increase in cigarette smoking seldom correlates with an increase in excess lung cancer. The theoretical values of D required to explain the data wander all over the place and are quite often negative. Burch considers whether the freakish behaviour might reflect errors in the data, the hypothesis or the time lag between the precipitating event and time of death, but nothing mends matters.

Lee’s verdict

Lee’s report on the paper to the TRC began

This is an incredibly silly paper in which Burch goes to great lengths to test and refute a hypothesis that is not only basically implausible but is capable of being refuted virtually instantaneously by a glance at the data he uses.

It ended

Having rejected Doll’s model and then the precipitator hypothesis, Burch is left with nothing really but to conclude that, as he cannot think of any other causal models, smoking is either not causally related to lung cancer at all or related to such a minor extent that one cannot pick up anything clear in the background noise due to problems of diagnosis. A conclusion with which I cannot agree.

Sources

1981

Smoking and mortality in England and Wales, 1950 to 1976

A second paper on the Precipitator hypothesis uses some of the same ideas, but it concerns the effect of smoking on the overall death rate, not just from lung cancer. concerns a test of the precipitator hypothesis and uses some of the same ideas, but it concerns the effect of smoking on the overall death rate, not just from lung cancer. Changes in smoking rates and changes in the death rate are calculated as before (Δx and ΔR/R appropriately modified) but they are now analysed as a time series from 1950 to 1976, one series for each age group, men and women treated separately. All in all, there is no very striking correlation between movements in the smoking rate and movements in the overall death rate. Life expectancy improved considerably in all age groups during the period under consideration. A section, and associated diagram, towards the end of the paper demonstrate that most of the improvement, certainly in men, can be accounted for by the near elimination of tuberculosis: no other explanation is required.

Lee’s verdict

Lee, reporting to the TRC, correctly observed that Burch’s argument assumed a linear dose-response relationship and drew attention to the Doll-Peto quadratic relationship of 1978. “In essence, this is a silly paper as it goes to great pains to test a hypothesis that cannot be justified and that virtually no-one believes anyway, namely that cigarettes smoked today are a major determinant of your chance of dying in the next 2 or 3 years.”

Sources

Smoking, lung cancer and hypothesis testing

New data from the 1970s became available while Burch was pursuing the Precipitator hypothesis. He devoted a fresh paper to it because he had reason to believe that diagnostic accuracy had improved in recent years. However, he found that the new data showed no tendency to corroborate the hypothesis.

Sources

Passive smoking and lung cancer

Takeshi Hirayama was the designer of a large scale prospective study of smoking and lung cancer in Japan, one of the eight major prospective studies and the only one conducted on a mainly non-white population. (It revealed a relative risk of lung cancer in smokers of 3.8, far lower than in the other prospective studies and suggestive of a strong genetic component to lung cancer.) Hirayama’s study was comprehensive and well designed and provided him with data for a study of passive smoking in the form of exposure to a spouse’s smoke, since Hirayama recorded the relationships between his subjects. From an analysis of 346 lung cancer deaths in women he found that the risk of lung cancer in non-smoking wives of smokers was significantly higher than in wives of non-smokers, and the risk increased with the husband’s level of risk.

Hirayama published these results in the British Medical Journal in January 1981 and a correspondence ensued. Sherridan L. Stock positively welcomed the news, which showed that “it will be remarkable if society continues to sanction such a ubiquitous and damaging form of pollution just so that a habituated minority can gratify themselves with a psychostimulant whenever they please. It is to be hoped that every effort will now be made by the authorities to minimise the incredibly selfish practice of smoking in public.” G. H. Miller noted that a study of his own in Pennsylvania had reached similar conclusions to Hirayama’s. However Burch was not the only dissenting voice.

G. C. Sutton pointed out that smokers tend to marry each other, so that one would not expect rates of lung cancer in husbands and wives to be independent of each other. Martin Rutsch drew attention to the 101 deaths in the 3,1951 unmarried women. Assuming the death rate to be the same in unmarried and married smokers, this implied a death rate among unmarried non-smokers which is higher than that in non-smokers married to smokers. Rutsch also showed that a supposed difference between rural and urban areas identified in the study was not statistically significant. Burch agreed with Sutton on the importance of assortative mating between smokers and smokers. A study of lung cancer in Athens showed a higher relative risk for non-smoking wives of heavy smokers than for smoking wives, and a genetic explanation worked just as well as a causal one.

Hirayama’s paper was the first epidemiological study of passive smoking, but the anti-smoking movement had had it in their sights for years previously as a crucial issue which could be used to denormalise smoking. The topic has been covered by Vincent-Riccardo Di Pierri in his book and website Rampant Antismoking Signifies Grave Danger and by Chris Snowden in Velvet Glove, Iron Fist, the book and website. Passive smoking as an alleged irritant and social nuisance was a running theme at the Fourth World Conference on Smoking and Health in Stockholm in 1979. G. H. Miller spoke about his work at this conference, and di Pierri has uncovered some interesting information about his reaction to questioning by Ted Sterling and Peter Lee. “Miller said that Sterling was able to pull out all of the inconsistencies in the Hirayama study because of the great detail presented. Miller said he wanted to avoid this and keep it simple.” Di Pierri also shows that Hirayama had been an anti-smoking campaigner for years previously, and that some questioned the conduct of his study of active smoking.

Some years later Alvan Feinstein had this story to tell.

In private conversation I recently heard an authoritative leader in the world of public health epidemiology make the following statement: “Yes, it’s rotten science, but it’s in a worthy cause. It will help us get rid of cigarettes and become a smoke-free society.”

Sources

Hickey, causation and the Establishment

Burch’s correspondence with Richard Hickey picked up again around this time. In 1980, Hickey had written to the Director of the National Cancer Institute, Vincent T. DeVita, protesting about the one-sidedness of the Surgeon General’s latest report on smoking and health. This and a succession of further letters were shunted sideways, given a polite brush-off or at best given answers which dodged the points at issue. The correspondence involved DeVita’s subordinates, Diane J. Fink and Joseph F. Fraumeni, and Robert Hutchings, Associate Director of the Office of Smoking and Health. Appeals to their nominal boss, the Secretary of Health and Human Services Robert S. Schweiker, and to Senator Arlen Specter of Pennsylvania proved fruitless.
It was probably naïve of Hickey to expect a barrage of letters, however erudite, to change official opinion. One interesting thing about the episode is how little difference was made by the supposedly decisive transition from the Carter to the Reagan administration (Schweiker and, at that time, Specter were both Republicans). In the 1980s, the campaign against smoking had become entrenched and was no longer subject to public debate.

Hickey passed the correspondence to Burch, who wrote back to congratulate him on his “pertinacity and courage. Your accusations of selective reporting are, of course, unanswerable. You will not be surprised to learn that I have come to take this kind of dishonesty for granted. How else could the Surgeon General put out his reports?” He continued

What really staggers me, however, is Fraumeni’s sentence: ‘Most revealing in support of a direct effect of cigarette smoke has been the fairly rapid decrease in the likelihood of developing lung cancer among men who have given up smoking…’ We can only assume that he actually believes this interpretation and is genuinely unaware of the underlying fallacy. Which, for the ‘Acting Associate Director of the Field Studies and Statistics Program’, is truly pathetic. It is usually more charitable to assume that these people are unenlightened rather than corrupt, but one cannot be sure in all instances as to which is the case.

Hickey seems to have been genuinely baffled that nobody wanted to know about the weaknesses in the case for causation. Burch did not have a ready explanation either. In a further letter he wrote:

It is not easy to account for the rise of the causal interpretation in this country because Fisher had his say in the BMJ and Nature; nevertheless, the environmentalists managed to get away with evasion of the central issues. The causal position is readily appreciated by those who have little sophistication and it is in harmony with general attitudes in medicine where the causal agent needs to be clearly identified. Curiously enough, causality also chimes with Marxist views. That might explain Doll’s dedication but I imagine that it plays little part in the ready acceptance of causality in the U.S. where, perhaps, naivety triumphs. Doll has been extremely influential in this country and many regard him as a demi-God. (My own admiration falls short of idolatry!) The puritanical, anti-pleasure principle, is also important and so I fear we shall find it extremely difficult to change attitudes. We can only appeal to the intellect; the causationists will capture the emotions most of the time.

Sources

At the Brewers Society

Information travelled more slowly in those days. In 1981, Burch was still getting feedback about his lecture to the Royal Statistical Society in 1978, for instance at a conference in Stratford-upon-Avon, where he lectured on oesophageal, laryngeal and colorectal cancer. He reported on it afterwards to Seltzer (26 October) and, separately, Hickey (23 November).

As befitted an event organised by the Brewers Society, it was followed by a sumptuous banquet at Ragley Hall, a truly magnificent Palladian house owned by the Marquis of Hertford who acted as our (very aristocratic) host and told us something of its history… The hospitality was lavish with champagne on three occasions; at the banquet we had three wines and port but the cuisine did not match them for quality. Overall, an enjoyable affair that did my reputation no harm at all.

More than one attendant at the conference expressed support for Burch’s work.

Allyn Kimball, Professor of Biostatistics and Statistics at Johns Hopkins, was the invited first discussant of my paper and he did me proud. He was more directly critical of the normal epidemiological fallacy of equating association with causation than I was and told the meeting about my JRSS paper on smoking and lung cancer. He pointed out that the Royal Statistical Society had invited me to address it on that subject and that among the 12 discussants, 6 had been in favour, 4 neutral and 2 against.

Afterwards I was greeted very effusively by Walter Somerville, former editor of the British Heart Journal. He told me about a paper he had given recently at the Royal College of Physicians where the audience were predominantly Senior Registrars. In this talk he had the temerity to extol the view of Burch and Seltzer on smoking, risk factors and CHD! He told them to study very carefully every word we had written on the subject! After his talk he was approached by a London epidemiologist (unnamed but obviously senior) who was livid with anger about what he had said in our favour. Somerville told him to go and read my Cardiovascular Research paper which used recognized procedures but arrived, of course, at highly unpopular conclusions. To his credit, this epidemiologist did so and wrote a letter of apology to Somerville!

From this and other episodes it is clear that many people do read our papers and actually agree with us but dare not in the present climate of opinion, state their agreement in print.

Sources

Strategy of prevention

A short letter from Burch in the British Medical Journal, 27 June 1981, was partly intended to draw attention to his paper on heart disease in Cardiovascular Research and partly in criticism of a paper Strategy of Prevention: Lessons from Cardiovascular Disease by Geoffrey Rose. That Rose had not been swayed by the results of his Whitehall study on the effects of giving up smoking can be gathered from these extracts. (He was a Methodist lay preacher and keen proponent of supposedly healthy lifestyles.) Rose began by noting low rates of cardiovascular disease in Japan and dramatic falls in Australia and the US, as against a lack of improvement in the United Kingdom.

In Britain, then, we are failing to prevent a preventable disease. If we had shared in the Australian and American decline each year in England and Wales there would be upwards of 25 000 fewer coronary deaths… We do not know why the Australians and Americans have done well in their control of coronary heart disease, or whether (if we did know) we could have shared their good fortune. Yet surely, as a profession, we should at least feel deeply disturbed by the problem and involved in it. We have a professional responsibility for prevention, both in research and in medical practice. When ordinary doctors do not accept that responsibility then prevention is taken over (if at all) by uncritical propagandists, by cranks, and by battling commercial interests.

Burch:

When we do not know how the outcome was accomplished the use of the word “control” is surely inappropriate; if factors beyond our control were responsible prevention would not be possible.

Rose:

There has been a gratifying decline in smoking by male doctors in Britain in the past 20 years. In most cases the motivation has probably not been the intellectual argument that in the end some will obtain health benefits; it has been social pressure. Being a smoking doctor is uncomfortable these days for your colleagues either pity you or despise your. Not smoking may be easier. Social pressure brings immediate rewards for those who conform…. To influence mass behaviour we must look to its mass determinants, which are largely economic and social.

So that is the purpose of intellectual argument.

Sources

Cigarette smoking and Parkinson disease

A letter by Burch commented on a paper in Neurology by Robert J. Baumann and others. A comparison of the smoking histories of 237 patients with Parkinson’s disease and 474 controls matched for residence, age, race and sex showed a strong negative correlation between the disease and smoking. They considered some possible explanations: smokers die too soon to contract Parkinsonism, an infection causes both Parkinsonism and smoking, and the first symptom of Parkinsonism is a change in personality which reduces the desire to smoke.

Burch made his familiar point about the three possible directions of causation and pointed out that the authors had ignored the possibility that smoking behaviour and predisposition to Parkinson’s disease might be constitutional. He repeated an observation first made by K. Westlund that if smoking prevented Parkinsonism then rates of the disease ought to fall with a rise in smoking levels and this had not happened.

Sources

1982

Controversy with Reif

In his later career as a smoking sceptic, Burch increasingly found himself engaging with opponents who would have been unworthy of his attention in the days before he was branded a scientific heretic. One such was Arnold Reif of Boston City Hospital and Boston University School of Medicine. (“Arnold Reif who hailed from Yorkshire. We met some years ago and bear no rancour although his anti-smoking attitude appears to border on fanaticism”.) Reif’s paper Effect of Cigarette Smoking on Susceptibility to Lung Cancer purported to rebut “Fisher’s fallacy” and “Burch’s extension of Fisher’s views”, together with a restatement of orthodoxy on smoking and cancer. He was concerned that “Medical scientists who debate Tobacco Industry executives on radio or TV (as I have done) should be aware of the the scientific fallacies they are likely to encounter and prepared to rebut them definitively.”

Since the paper criticised Burch extensively by name, it might be thought that he was entitled to defend himself in the journal Oncology which printed it. Burch had no such expectation.

If they can publish Reif’s essay without shame it is unlikely that they will look with favour on mine. Although my effort was almost certainly wasted, the writing afforded me some amusement mixed up with a sense of shame at attacking so pathetic an opponent. Bully! However, to be bracketed with Fisher – even an evil Fisher whose logic can be faulted – is flattery indeed!” (to Hickey, 4 September 1981).

(“Bully”: probably “Oh what a bully I am” rather than “Bully for me”.)

He was right. “My suspicions about Oncology have been duly confirmed. Needless to say the referees made no attempt to fault my arguments; my paper contained ‘no novelty’ and my ideas ‘had been widely reported elsewhere’.” (to Hickey, 20 October 1981).

Hickey suggested publication in the Journal of Chronic Disease, now co-edited by Alvan Feinstein, but Burch intended another paper for that outlet and sent his reply to Reif to Medical Hypotheses, where it appeared in 1982.

Burch did not bother to consider Reif’s theory of smoking and cancer (a variant of Doll’s view of cigarette smoking as a promoter of lung cancer), but concentrated on the issue of causation. Reif’s list of 12 facts supposedly disproving the constitutional hypothesis and Burch’s refutation of the refutation deserve quotation in full. The rest of the paper discussed the possible genetic basis of the correlation. Reif raised the familiar objection that cancer genes doubling as smoke genes must be implausibly coordinated, and that Burch assumed the truth of his autoaggressive theory. Burch answered these points and considered evidence for the genetic theory: some of this is new, for instance a report by Brian Henderson on mainland China and a study of Mormons by James Enstrom. A full version of Reif’s facts and Burch’s counterarguments is here.

Burch sent his reply to Reif to Marvin Kastenbaum, a statistician at the Tobacco Institute and apparently that organisation’s Burch-watcher (see his report on Lee’s examination of Burch and on the Royal Statistical Society address). Whether they had other contact is unknown.

Lee’s verdict

Lee summarised Burch’s reasoning in a report to the TRC and adjudged him the victor on every point except number 6, male-female sex ratios. On this matter, Lee reproduced Peto’s comparison, presented at Hammersmith, of the generations born in 1850, 1890 and 1930 as evidence for the causal theory. He dismissed Burch’s alternative account as “extremely implausible”.

Sources

Cigarette smoking and coronary heart disease

An article in the Canadian Medical Association Journal by W. A. Tweed, Does cigarette smoking cause coronary heart disease, was an attempt to refute the views of Seltzer and Burch on the issue. Burch replied in a letter to the editor. Tweed seems to have known their views mainly from their two relatively non-technical articles in American Heart Journal for 1980: Seltzer, What are we to believe? and Burch, Smoking and coronary disease.

Tweed had read this material and understood it. He grasped the point about self-selection and argued that Gary Friedman’s ‘baseline’ study has refuted it. He saw the importance of randomised prospective studies and twin studies and accepted the admission of Rose and Cederlöf that their studies were inconclusive.

It is implicit in the logic of all this that the causal role of smoking in coronary heart disease was completely unproven until 1981, when Friedman put it beyond all doubt. On that supposition Tweed asserts

“Although our knowledge is incomplete, we must continue in our efforts to eliminate smoking.”

Burch’s reply cited fresh material from his papers in Gerontology and Cardiovascular Research. These were clearly new to Tweed, and he seems to have made an honest effort to understand them. The reply, and Tweed’s reply to the reply, were printed on 15 July.

Tweed dismissed Burch’s curves, and the equations which they picture, on the grounds that the variables and constants involved “have no measurable biologic status”. The only one with a basis in observation was time, so “Burch’s demonstration that age-dependent disease increases in a predictable manner with age is simply a tautology.” Like Friedman, Tweed held that epidemiology should confine itself to observable (“biologic”) factors and rejected hypothetical entities like Burch’s forbidden clones for being hypothetical.

Tweed refused to believe that Burch was an honest man.

The health hazards of cigarette smoking are too well documented to be questioned for any reasons other than perversity or self-aggrandizement. Although some evidence, including that of Burch, suggests there may be subsets of the population more susceptible to smoking-related diseases, there is no evidence that anyone is immune. Our goal, therefore, should be to abolish the offending factor; we should not be diverted by academic game-playing.

Abolition was now the ultimate goal of the anti-smoking movement.

Sources

The Waxman bill

In March 1982, US Congressman Henry Waxman introduced a bill requiring new, more strongly worded health warnings on cigarette packets. The tobacco industry was swift to mobilise its congressional allies, led by Congressman Thomas Bliley (who gave his name to an important tranche of the Tobacco Documents), and to provide them with ammunition supplied by sympathetic scientists, including Seltzer and, at Seltzer’s prompting, Burch, who drew up a ten-page memorandum on smoking and health and sent it to Bliley with offprints of some of his papers.

According to the preamble of the bill,

The Congress finds that –

(1) Cigarette smoking is the largest preventable cause of illness and premature death in the United States and is associated with the unnecessary deaths of over three hundred thousand Americans annually

Other findings blame smoking for lung cancer, emphysema, heart disease, premature birth and stroke in women taking the pill. The ninth and last finding proclaims the need for a new strategy “to educate and provide information to the American public to allow them to make informed decisions as to whether or not they should smoke”.

Burch’s submission consists of a survey of his work on smoking, lung cancer and cardiovascular disease up to 1981, including Smoking and Mortality in England and Wales and Ischaemic Heart Disease, which he enclosed in the form of offprints. He wrote to Seltzer. “Dear Carl, I thought you might like to see my submission to Congressman Bliley. How will it be used, I wonder? What are the chances for this ghastly bill?”

Seltzer wrote to another correspondent, “… it is much too long and detailed. However the points he makes are very good.” Bliley read the entire submission into the Congressional Record on 11 August 1982.

Sources

1983

Submission to Senator East

The Waxman bill having passed in the House of Representatives, Senators Orrin Hatch, Robert Packwood and Edward Kennedy brought corresponding legislation in the Senate in 1983. Opposition to the bill was led by Senator John Porter East. Burch again joined in the lobbying: his contribution was substantially the same memorandum as he submitted to Bliley, incorporating a few new developments such as the failure of MRFIT. This document, and statements by other sceptical scientists, were included in the Congressional Record, but to no avail. The bill passed and was signed into law by President Reagan on 12 October 1984.

Tutored no doubt by Seltzer, Burch opened his letter to the senator thus: “Dear Senator East, I have the honour to enclose a critique of bill S. 772 before the Senate and House of Representatives of the United States of America.” It is interesting to contrast the required formality with the familiar, old boy network tone in which Burch addressed David Owen and Nigel Lawson, ministers of the crown and future peers of the realm.

Sources

Oslo and MRFIT

Two new randomised prospective studies were published during this period: the small scale Oslo study at the end of 1981 and the much larger MRFIT in September 1982.

The Oslo study by I. Hjermann and others was part of a larger survey of all men in that city. 1232 of them were identified as having a high risk of coronary heart disease and randomly assigned to an intervention group of 604 men or a control group of 628. The intervention group were strongly advised to modify their diet and give up smoking while the controls were not. 25% of them gave up smoking completely as against 17% of the controls. After five years, the intervention group had significantly lower levels of CHD than the controls, but death rates were not significantly different. Cancer deaths were few and not statistically significant. Reduced cigarette smoking seemed to account for little of the improvement in CHD rates because, so the researchers claimed, “the statistical power of the test was too low”. The power was intended to be 60%, but the behaviour of the control group meant that it turned out to be lower.

The MRFIT trial involved 12,866 men at high risk from CHD recruited at 20 locations in the USA, randomly assigned to an intervention group or a control group. The intervention group was intensively counselled to consume less cholesterol and give up cigarette smoking; they were also given personal treatment for high blood pressure if they had it. The control group were left undisturbed. After seven years there was no significant difference in the overall death rate or in deaths from CHD. There were slightly fewer cancer deaths from lung cancer in the control group than the intervention group (53 as against 58) but not significantly so. The power of the test was intended to be .88 but turned out to be .75, contrary to expectation derived from the Framingham study, and again low power was blamed for the negative result.

The results of MRFIT were published in the Journal of the American Medical Association. The editor, George D. Lundberg, did not hide his disappointment nor to fail to stress that the study was inconclusive through lack of power. In publishing, the journal had met one of its goals, which was “to foster responsible debate”. However, he broadmindedly gave first place in the correspondence columns to Burch and Seltzer.

Burch outlined the methodology and results of his two major papers on CHD, noted that MRFIT was consistent with them, and called for the real precipitating agents of heart disease to be identified. Seltzer congratulated JAMA on publishing a negative result, which in the light of Whitehall and Oslo was no aberration. The whole question of what causes CHD ought to be reopened. In other letters, Dr Kurt Oster argued that the expected result of MRFIT was derived from a wrong analysis of the Framingham study which overemphasised the role of lipids (dietary fat) and treated CHD in isolation from atherosclerosis in general. David Gordon and Weldon Walker made technical points about the treatment of hypertension. I. D. J. Bross had an explanation of the “anomalous” lung cancer rates in both intervention and control group: those who gave up smoking must have died of CHD instead.

The main author of the report, Dr Jeffrey A. Cutler, replied to Burch that the genetic explanation did not fit studies of migrants and unrandomised studies of quitting and to Seltzer that there was overwhelming evidence that the risk factors are causes, and possible reasons for the unexpected result were given. To the other correspondents he replied that the misinterpretation of Framingham was no such thing, the point about treatment of high blood pressure was under consideration and that it was too soon to expect reductions in lung cancer.

Sources

Radiation-Induced Cancer

Burch described this as “my shortest publication and, perhaps, the least important”. It reads, in its entirety:

I am disturbed to see that the misleading expression “radiogenic cancer” is becoming increasingly popular; it even appears in the title of an article published recently in this journal. Admittedly 14C and 40K nuclei – among others – in cancer cells emit ionizing radiations and hence in one rather contorted sense cancers might be said to produce radiation. Seldom, if ever, is this the intended meaning. In the great majority of instances editors would be justified in substituting “radiation-induced cancer” for “radiogenic cancer”. May I, sir, respectfully plead with you to do so?

To a traditionally-educated Englishman like Burch, ‘radiogenic’ ought to mean ‘causing radiation’ just as ‘carcinogenic’ means ‘causing cancer’. “My wife came across ‘anthropogenic’ recently – not in the context of reproduction!” he told Hickey.

Sources

Cigarette smoking and ulcerative colitis

A paper by Herschel Jick and Alexander Walker presented evidence strongly suggesting that cigarette smoking protects against ulcerative colitis and provoked a letter from Burch. A study of sufferers matched with controls found that it was far less common in smokers than in non-smokers, including ex-smokers. However, “At present, a causal association between smoking and a reduced risk of ulcerative colitis remains a highly speculative explanation of the data observed.”

An editorial in the same number by John C. Bailar III MD identified four possible explanations of the findings: random chance, biased data, constitutional difference between smokers and non-smokers and a prophylactic role for cigarettes. But, he warned:

Evidence for a straightforward causal effect on lung cancer was clear to unbiased observers almost two decades ago, though because many people find the truth so unwelcome, some research on whether causation is involved still continues…

Although informed skeptics about the overall health effects of tobacco are now few and on the fringe, enough traces of the former controversy remain to unsettle the public and its legislative representatives. These lingering echoes of past battles remind me of the last visible trace of the Cheshire cat, though the grin has become a risus sardonicus. A critical public and a responsible tobacco industry should not attempt to redress unreasonably strict standards of proof that cigarettes cause lung cancer with unreasonably lax proof that cigarettes protect against ulcerative colitis.

A correspondence followed. S. N. Gyde and R. N. Allen reported a similar result, Evelyn Bowers, Elaine Allen and Richard Hickey favoured a causal explanation and Stephen Hanauer and others supported Bailar’s agnostic outlook. Burch commended Jick, Walker and Bailar for their caution but Bailar was wrong to insinuate that lung cancer sceptics are “not unbiased” and “find the truth… unwelcome”. In reply, Bailar acknowledged the points made as valuable and important and affirmed that no personal attack was intended on Burch.

Burch, politely feline in a journal publication, was openly sarcastic in a private letter to Seltzer.

The negative association between smoking and ulcerative colitis has obviously given much cause for concern. The caution with which this is being interpreted is exemplary. However, I took offence at Bailar’s comments in the Editorial columns and have despatched the enclosed letter to the Editor.

Seltzer replied

I was amazed at the extreme rigour of his examination of possible explanations, epidemiological weaknesses, et cetera, which he employs in his caution to accept a cause and effect conclusion with respect to the significant negative association between ulcerative colitis and cigarette smoking. To my knowledge, nothing like this has appeared before in connection with any finding of a positive association between cigarette smoking and a disease.

Sources

The Surgeon General’s “Epidemiological Criteria for Causality”: a critique

The reports of the Surgeon General of the United States continued to appear yearly, as they had since 1965. In a paper published in the Journal of Chronic Disease for 1983, Burch subjected the scientific methodology and statistical reasoning of the 1982 report to critical examination. As he noted, it was the first time such a thing had been done since Brownlee’s negative assessment of the original 1965 report. It begins with a full-length statement of the ways in which correlation between a habit and cancer (H and C) may reflect causation.

I H causes C the causal hypothesis
II C causes H the converse causal hypothesis
III Other factors cause H and C the common cause hypothesis
(or where these are genetic) the ‘constitutional hypothesis’
IV I, II and III in combination

For Burch, a statement that smoking causes cancer ought to be backed up by a consideration of all these possibilities with an estimate of how they combine to explain the correlation and how confident of the explanation we can be.

Successive Surgeon Generals’ reports since 1964 presented the evidence under five headings, the supposed criteria of causation:

  • consistency
  • strength
  • specificity
  • temporal relationship
  • coherence

which Burch considered in that order.

Those who publish extensively on the same subject often repeat themselves even when they have something new to say, either because they are writing for a fresh audience or because the structure of the argument demands it. Burch was no exception, and many of the facts in this paper are familiar from his earlier work. His purpose was to evaluate them in the light of the Surgeon General’s methodology and his own.

To the Surgeon General, a table of risk ratios in 35 retrospective studies, all of them positive, was evidence that different studies taking different approaches display consistency. To Burch, their different sizes, ranging over two orders of magnitude, suggested the opposite. A statement about Swedish twins, also cited as evidence of consistency, proved to be unfounded.

Risk ratios in different populations were also relevant to the criterion of strength, and here Burch had new data relating to Asia. Risk ratios in Asians were very significantly lower than those in whites. A report by B. E. Henderson from mainland China, then opening up after the Cultural Revolution, claimed a risk ratio as low as 1.57 in that country and reported that Chinese scientists were very sceptical of the causal hypothesis. Burch’s material on dose-response relationships contradicted the Surgeon General’s assumption that they reflected the strength of a causal relationship and actually supported the constitutional hypothesis better.

The sheer number of smoking related diseases told against the idea that smoking was a specific cause of anything, and Burch’s analysis of different forms of cancer and of Parkinsonism showed that no specific mechanism was at work.

All Burch’s work on long term trends in smoking and lung cancer was relevant to the criterion of temporal relationship, and he presented it once more.

Most of what Burch had already said was relevant to the criterion of coherence, including the modal age of onset, which the Surgeon General included under this head. This section of the report laid great stress on low lung cancer rates in groups such as Mormons who forbid smoking. These groups consist of self-selected converts and their descendants, and Burch’s severest criticisms of the report centred on its neglect of self-selection, the need for randomisation and the importance of the randomised Whitehall study. He also discussed new work on Mormons including a study by James Enstrom (later known as a passive smoking sceptic) which showed that lung cancer rates in Mormons “are not clearly related to their smoking habits”.

Burch concluded that “in the context of lung cancer: (a) reported associations are inconsistent; (b) the reported strength of associations ranges widely; (c) the association has no specificity; (d) the temporal relationship shows many anomalies; and (e), because of (a) to (d)… the association lacks coherence.”

Exchange with Lilienfeld

The same issue of the Journal of Chronic Disease contained a Criticism of Burch’s Critique by Abraham M. Lilienfeld and the next one a Reply to Lilienfeld by Burch. Lilienfeld was known to smoking researchers for his work with George Takuhata on the incidence of smoking and lung cancer within families. He was also the brother-in-law of Jacob Yerushalmy, though they clearly had different approaches to epidemiology. His Criticism begins “In the history of science, there are always a few individuals who are never convinced when a new concept appears.” His approach emphasises “the totality of the evidence” for “a scientific judgement that there is a sufficiently high probability that the causal hypothesis is true, and that it is sufficiently true to warrant taking preventive action”.

He proceeded to take issue with Burch’s principal points in the order he made them, that of the Surgeon General’s five criteria, though he stated in passing that he preferred Bradford Hill’s ten point elaboration of them in 1965. Lilienfeld was completely out of sympathy with Burch’s Popperian approach to science, to the extent of finding his attempts to falsify hypotheses evidence of bad faith. “It came as a complete surprise to note that Burch has incompletely presented the findings of several studies, or has omitted discussing others more fully. This is, needless to say, most disturbing.”

Burch’s “Reply to Lilienfeld” began “New concept or old hat?” and continued with an account of how he originally accepted the causal theory and came to question it. “Should I respond to [Lilienfeld’s] imputations of bad faith by descending to counter-accusations?” These “might entertain the reader – and are difficult to resist – but I doubt whether they would illuminate the methodologic problems.”

There is no point going over every issue the two men debated. The interesting thing is what Lilienfeld regarded as important and what he did not. He went into considerable detail about Asian lung cancer rates. When cancers are analysed by histological type, Asian and Western rates were “really not that different,” to which Burch retorted that reported rates vary so widely that “it would have been rather remarkable had they been ‘really different’”. This is mild seeing that Lilienfeld saw fit to add “disregard of the high relative risk with the epidermoid type raises serious questions about the author’s inference and the omision of the detailed results by Burch raises other kinds of issues”.

On the other hand, “the findings of a negative association with Parkinsonism and possibly other diseases is not relevant to the present issue. However, how Burch can infer that such a negative association is due to genetic factors puzzles me.” In reply Burch restated his reasoning, but perhaps this is missing the point. Lilienfeld was concerned with identifying and reducing risks to health and regarded explanation as a means to that end. Where no risk is apparent, no explanation is required. Similarly, the negative results of Rose’s Whitehall study went to show that his approach was fruitless and ought not to be tried again.

Lilienfeld devoted a lot of space to questioning the data which went into the Four Tides diagram and regarded the supposed 30-year lag between uptake of smoking and the appearance of lung cancer as “pretty straightforward reasoning”. But as for the remarkable Twin Peaks correlation, “I’m not sure why it is even necessary to discuss the rate of change of frequency of smoking in the different sexes as presented by Burch. His argument is irrelevant.” Lilienfeld simply misunderstood Burch here: the Twin Peaks diagram plots the change in the speed of change in lung cancer rates and brings to light a pattern which, on the 30-year lag theory, ought not to exist.

On the subject of Mormons, Lilienfeld had intriguing new data. In the progressively higher ranks of the Mormon priesthood – Aaronic, Melchizedek, Seventh Priest and High Priest – lung cancer is progressively rarer. “Burch would probably state that these results are due to selection, but I think he would be pushing that explanation a bit too far.” The word “explanation” is telling. Lilienfeld saw selection as a confounding factor to be investigated in the event of an unexpected result. He just does not see that random sampling is the essence of statistics and that without it there is no saying whether a result is expected or not.

Burch was the most courteous of debaters, but towards the end of the exchange he put his true opinion of risk factor epidemiology beyond doubt. Criticised by Lilienfeld for ignoring the Lilienfeld-Tokuhata study of possible genetic factors in lung cancer, he retorted that the Surgeon General had ignored it too, along with many other studies on similar lines which might have been cited. “Perhaps by that time I was tired of pointing to a neglect of genetic considerations that would have done credit to Lysenko.”

Sources

6. Preventive Medicine

Continued