A Half-life of Burch
4. Heart Disease
In the years 1978-80, much of Burch’s work concerned coronary heart disease (CHD) and the overlapping category of ischaemic heart disease, introduced after a major change to standard classification in 1968. Nobody seriously claimed that smoking was overwhelmingly to blame for heart disease. It was one of a number of ‘risk factors’ also including obesity, high blood pressure, diabetes and low levels of exercise, and there were not wanting those who favoured a legislative war on supposedly unhealthy lifestyles in general. For that reason, Burch found it easier to get a fair hearing for his ideas. He published two long papers on the topic in important journals in 1978 and 1980, with a number of letters to the journals arising out of them.
Coronary heart disease: tests of an etiological hypothesis
The curtain rose on this period with a guest editorial in American Heart Journal for December 1977, and two more followed. Carl Seltzer must surely have had a hand in this, but we do not know and there is no entertaining inside story of this period. Burch’s letters to Seltzer (as preserved in the archive) drop off around this time, not because their friendship had cooled – far from it – but because they seem to have communicated by long distance telephone call, far more expensive then than now.
The editorial and its two sequels are a general discussion of the problem and how it should be attacked. Some, but not all, epidemiological studies seemed to show that CHD was associated with smoking, but Wald’s recent work had been refuted by Seltzer. The Doctors’ Study was methodologically flawed and research on twins was the most promising avenue.
Early in the following year two exchanges arose in connection with the editorial.
Smoking and the cause of CHD
In the editorial, Burch noted that Ancel Keys’s Seven Countries Study had shown (as of 1970) no significant correlation between smoking and heart disease in five of the countries, including Crete and Corfu (the two were studied separately though both, of course, are in Greece).
Dr Christ Aravanis, who had conducted the study in those islands, reported that he had found an association at last, “after some 18 years of investigation” as Burch noted in reply – in any case, associations do not settle the question of causation.
Why don’t they read the old books?
Dr Otto Neurath wrote that Burch’s criticisms of the ‘risk factor’ approach to CHD “did not go far enough”. Neurath surveyed successive modifications to the story that cholesterol is to blame since it originated in 1913, showed that interesting evidence to the contrary had been ignored and forgotten, and called for risk-factor epidemiology to be abandoned in favour of a clinical and physiological approach.
In reply, Burch cordially agreed with Neurath’s attack on the way that correlations had been elevated to the status of causes under the name of “risk factors”. However, bad statistical techniques should not discredit good ones. Burch’s own analysis, now published in Gerontology, exonerated more or less all the major risk factors in favour of a genetic explanation.
Coronary Heart Disease: Risk Factors and Ageing
This, one of Burch’s two major papers on coronary heart disease, was published in Gerontology in 1978. Burch attacked the problem in the same way as lung cancer, analysing the data in terms of his autoaggressive theory. However there were important differences between his analyses of the two diseases.
- The disease commonly called coronary heart disease was first identified in 1912 and had undergone several changes of name and definition since then.
Deaths rates from CHD were so high that they affect the general death rate, which was not true of lung cancer death rates.
There was little difference between Burch’s theoretical curve and a standard power law distribution.
Burch’s equations predicted age of onset of a disease but data referred to age at death. The correction was simple for lung cancer, more complex for CHD.
To the extent that factors like smoking caused CHD they did so by shortening the period from onset to death (represented by the parameter λ in Burch’s equations), and much of the paper is a discussion of six important risk factors: cigarette smoking, weight, exercise, hypertension, diabetes and cholesterol.
After discussion of his general theory, Burch proceeded to consider changing rates of CHD between 1921 and 1973. There was a large apparent increase which might be the result of a precipitating agent and might be an artefact. Men and women matched the same curve (allowing for a different latent period), though smoking rates were not parallel. There was a similar relationship in rates in US blacks and US whites.
The last part of the paper considered whether each of the six risk factors might be a causal factor, given these patterns. The factors considered were nothing like independent of each other, were probably genetically influenced, and probably none of them were causal factors.
A couple of short publications from 1978 are based on the Epidemiology paper.
Coronary Heart Disease and Smoking
Tabak-Journal, the trade publication which featured Burch’s views on lung cancer in 1974 did the same for heart disease in this article, a plain language summary of the conclusions of Coronary Heart Disease: Risk Factors and Ageing without detailed argument. They were as follows.
CHD accounted for 26% of deaths whereas cancer of all types accounted for only 21%.
The risk of CHD was higher in smokers but less so with advancing age. Correlation was not causation and the disease was known to be partly genetic.
Early results from studies of twins bore out the predictions of the constitutional hypothesis.
Diagnostic standards made analysis difficult but studies of the entire population (not just smokers) shed light on the matter.
Smoking was not related to initiation of CHD, only to development.
There was a latent period of 15 years in men, 24 in women, but its length was shorter in heavier smokers: however, long-term change in rates of CHD showed no relation to changing smoking rates.
Causationists appealed to correlation and the Doctors’ Study, but Seltzer had rebutted those arguments. “Little or no causal action is involved.”
An end to heart attacks?
According to an editorial in the popular magazine World Medicine, there was no doubt that giving up smoking reduces the risk of coronary heart disease in middle age, but until now it seemed that cutting down on fatty foods to reduce cholesterol was a matter of too little too late. However, Bill Castelli had recently claimed that “atheromatous plaques” can be completely removed from the arteries if cholesterol is reduced to starvation levels, and preventive medicine could be prescribed with a vengeance.
Burch, dissenting from this in a letter to the editor, noted that the evidence that giving up smoking reduces the risk of heart disease was far from “unequivocal”. Arguments from the Doctors’ Study, beloved of preventive medicine, collapsed upon examination. Self-selection and diagnostic error were huge problems, but Burch’s paper in Gerontology examined evidence for which they are less important. The association with smoking appeared to be non-causal and to wane with advancing age.
Coronary disease: risk factors, age and time
Burch contributed this, another guest editorial, to the American Heart Journal in April 1979. It is a short and, by Burch’s standards, accessible summary of the results presented in Gerontology the previous year. He restated his conclusions that smoking, cholesterol and other ‘risk factors’ do not cause CHD and that health policies which treat them as such “are likely to create unnecessary misery and anxiety”.
The tumorigenic theory of atherosclerosis
A brief correspondence arose out of a paper by Earl P. Benditt and John M. Benditt presenting evidence that atheromata – the plaques which obstruct the arteries supplying the heart with blood in atherosclerosis – are monoclonal, that they originate as a single cell and are akin to tumours. To Burch, this was big news, since it provided support for his concept of autoaggressive disease which unified the apparently unrelated categories of cancer and degenerative disease. Burch wrote a long letter to the editor expounding that theory, appraising the Benditts’ observations in its light, and drawing attention to his recent work on coronary heart disease which exonerated cholesterol and other alleged causal factors. Burch’s letter referred in passing to a previous letter by P. Constantinides which controverted the Benditts’ reasoning. Constantinides replied to Burch, noting various features of atheromata which still needed explaining, and asserting that the recent fall in coronary heart disease proved the value of a low cholesterol diet.
Lee submitted an unsympathetic review to the Tobacco Research Council. The first sentence was “Having, apparently, got bored with lung cancer, Burch has now extended his all embracing theory of growth to atherosclerosis.” The last was “The net effect of Burch’s extension into atherosclerosis will, I imagine, be to undermine his credibility regarding smoking and lung cancer.”
Ischaemic heart disease: epidemiology, risk factors and cause
The second of Burch’s two major papers on heart disease was published in Cardiovascular Research in 1980. It has much the same structure and argument as the Gerontology paper: first Burch expounded his general approach and his theory of disease and then applied them to actual data. He again concluded, from a consideration of age-specific mortality ratios, that none of the six main risk factors for heart disease is a causal agent. The main difference between the two papers involves the change in diagnostic criteria with the introduction of ICD8 in 1968. The Gerontology paper discussed data collected subject to the old criteria, while this one used data from the ICD8 period. On the basis of the new data, Burch presented evidence that two of the risk factors – hypertension and cholesterol levels – were themselves autoaggressive conditions with genetic causes.
Two passages in the paper are replies to important criticisms of Burch’s approach: his neglect of curve-fitting (statistical significance) and his apparently circular practice of describing data in terms of the hypothesis which they are alleged to corroborate. They are worth quoting in full.
In a rigorous analysis based on a least squares fit and: (i) using data for death rates with only random (and no systematic) errors; (ii) knowing the distribution of the latent period; and (iii) knowing the population distribution by age within 5-year age groups, we could in principle compute chi-squared from recorded and predicted numbers of deaths and evaluate P, the probability of the null hypothesis. Because these several stringent requirements of rigorous statistical analysis cannot be met, either by the data or the theory, we have to abandon the idea of calculating confidence limits; and approximate graphical comparison of observed and calculated age-specific death-rates must needs suffice. This, however, is adequate for our present purpose which is to explore the possibility of accounting plausibly for the quantitative (though imperfect) mortality data in terms of a mathematically approximate but biologically well-founded theory.
That is to say, the problem is partly with the quality of the available data, but there is a more fundamental difficulty. The observed curves, on Burch’s theory, arise from the interaction of two different theoretical curves, one for the true onset of disease and one for the latent period during which it goes undetected, and the mathematical tools for tackling the problem did not exist.
On the use of hypothetical entities in his reasoning:
We have seen that all the above “risk factors” associate with the latent period, λ, and that some, probably, associate with S, the proportion of the population at risk with respect to IHD. Our major problem is to decide whether disorders such as hypertension, hypercholesterolaemia, obesity, etc. help to cause IHD by shortening the “natural” latent period between the initiation of the disease process and death; and/or whether they can precipitate the growth of a forbidden clone which, otherwise would remain suppressed by the host’s endogenous defence mechanism. (I cast the problem in terms of the theory of auto-aggressive disease, although it could be presented, somewhat laboriously, in a purely mathematical formulation without reference to any explicit biological mechanisms.)
The (dwarf) planet Pluto and the element germanium were called Planet X and ekasilicon before their existence was confirmed. Burch might, perhaps, have explained his theory in terms of hypothetical entities called by some such name as ekasomes and only later gone on to identify them as forbidden clones. However, this would, as he says, have been laborious, and it would also have been untrue to the biological origins of his ideas in the work of Macfarlane Burnet.
Lee, commenting on the paper for the Tobacco Research Council, had some perceptive criticisms (his thinking on lung cancer at any one time generally mirrored that of Richard Peto, but heart disease was not subject to a party line from “the Oxford workers”). One criticism was that Burch analyses ischaemic heart disease as three different diseases in men and two more in women, and that he fitted five different curves with different parameters to the data. A second was that deaths from a common disease modify the structure of the population affected by it (the phenomenon of ‘removal from the population’). A third concerned the invariance of Burch’s parameter λ with rising age. To Burch, this was evidence that the risk factors for IHD are genetic, not causal: Lee concluded that he had not argued this point convincingly.
(In defence of Burch it may be said that the five different types of IHD are distinct diagnostic categories in the ICD and were not invented for the occasion. As for removal from the population, Burch had discussed in in Coronary Heart Disease and pronounced it unimportant in the age groups below 75. On the invariance of λ, Lee may well have a point.)
Heart disease in different ethnic groups
An editorial in the British Medical Journal in August 1980 noted with regret that the organisers of the forthcoming 1981 census had dropped proposals for a question about ethnic origins. Many interesting facts hinted at a genetic role in the causation of heart disease, but first-generation immigrants were different from their descendants and only time would tell what factors were environmental.
Recent migrants from one country congregate together, sharing the same social and dietary habits and occupying particular jobs, which will differ from those of the host population and those of other migrant groups…. A generation or so later immigrants’ descendants may be so well integrated that much of any remaining differences might well be genetic, but the more complete the integration the greater the practical, ethical, and political problems of attempting to define those variations.
Burch wrote in (13 September) to publicise some conclusions of his own in Ischaemic Heart Disease. Different populations did indeed have different susceptibility to acute myocardial heart infarction, but a precipitating factor was required for the disease to develop in those at risk. Once it does, different ethnic groups have different prognoses, but an entirely different set of genes was now involved. The so-called risk factors were not causes but manifestations of the genes at work in this second phase.
Smoking and coronary disease
This non-technical paper, containing no new results, considered whether smoking causes CHD and how one would go about proving it. Burch began with a critique of current epidemiological standards, as exemplified by the recent report of the Surgeon General (1979), for ignoring the direction of causality, genetic factors and self-selection. The Rose and Hamilton (‘Whitehall’) study and results so far from the Swedish twins’ study (Cederlöf) did not support a causal role for smoking. Another possible approach was the analysis of long-term trends. Burch had undertaken such an analysis in his Epidemiology paper and found that there is no reason to blame smoking (or for that matter saturated fats) for CHD.
Burch seems to have come across the work of Alvan Feinstein during this period. He several times quotes Feinstein on the contrast between hard and soft scientific methods in epidemiology. “A ‘licensed’ epidemiologist… can obtain and manipulate the data in diverse ways that are sanctioned not by the delineated standards of science but by the traditional practice of epidemiologists”. Feinstein submitted a positive response to the Royal Statistical Society paper and published Burch in the Journal of Chronic Disease. The two men admired each other’s work although, perhaps surprisingly, they only met once.
Exercise and the heart
A paper of this name by Professor J. N. Morris of the London School of Hygiene and Tropical Medicine reports on a prospective study of coronary heart disease in male civil servants. The men who exercised regularly had a lower incidence of CHD, even in the presence of ‘risk factors’ such as smoking and hypertension. On this basis, vigorous exercise was claimed to have a protective effect against heart disease.
On 3 January 1981 letters appeared from Dr R. D. Jarrett and from Philip Burch. Jarrett complained that Morris had ignored the problem of self-selection: for instance, men with the symptoms of angina are unlikely to take vigorous exercise. Burch pointed out that causation can operate in more than one direction and that correlation is not causation. Morris and his colleagues replied on 31 January. In answer to Jarrett he said that his point about angina was not borne out by the data. To Burch he said that reverse causation is possible and in the short term angina might be a case in point, but the benefits of exercise appeared to be long term. In other words, Morris could not see beyond risk factor epidemiology to the real problems of proving causation.
Mortality in cigarette smokers and quitters
This correspondence in Medical Intelligence, to which Burch made a short contribution, is part of a long but important story. It was occasioned by a paper of the same name by Gary Friedman, a former associate of Carl Seltzer, and others using data made available by the Kaiser Permanente health insurance scheme. Members of the scheme, who constituted a large and fairly representative sample of the American population, were encouraged to complete periodic ‘multiphasic health checkups’ (self-administered questionnaires) which furnished a wealth of data on their changing health and lifestyles over time. Most recently, Friedman and Seltzer had collaborated on a study of four subgroups of people, classified by their smoking habits, who had completed three checkups. They were
- A Persistent smokers
- B Temporary quitters (these had given up and started again)
- C Persistent quitters (these had successfully given up)
- D Never smokers
A comparison of the four subgroups ‘at baseline’, the point before the quitters quit, showed significant differences between them: for instance, on one measure of body weight and numerous other relevant factors, the future quitters resembled never smokers more than persistent smokers.
That is to say, the four classes of subject group like this: A (BCD). The study was published as Characteristics predictive of coronary heart disease in ex-smokers before they stopped smoking in 1979.
Then Friedman and Seltzer parted ways. The unsympathetic version is given by Robert N. Proctor (the CTR was the industry-funded Council for Tobacco Research).
From 1971 through 1980, for example, Gary Friedman, an epidemiologist at Kaiser Permanente in Oakland, California, had been handsomely supported by the CTR for his work exposing complications in some of the epidemiology linking smoking and heart disease. The CTR liked his skeptical bent, feeding as it did their hope that something – anything – other than smoking would explain the ills experienced by smokers. (CTR’s governing lawyers in 1978 were still confiding that they “must find skeptical scientists.”) Friedman obtained about $100, 000 a year from this source, resulting in a number of papers co-authored with Harvard anthropologist Carl Seltzer, one of the industry’s “no proof” ideologues. Friedman was not an ideologue, however, and in 1979 he made the fatal mistake of noting in one of his papers that “most scientists now agree that cigarette smoking is an important factor in causing death”. Friedman shortly thereafter had his funding cut off, as was often the fate of “admitters” of this sort (honest scientists, one could say).
The new paper was Mortality in Middle-Aged Smokers and Nonsmokers. Proctor truncates Friedman’s words in it slightly: the full version is “most scientists now agree that cigarette smoking is an important factor in causing death from lung cancer and chronic pulmonary disease”. Whether it was an important factor in death, without qualification, depended largely on whether it caused coronary heart disease.
Friedman’s method of determining this was to ‘control’ for factors such as body weight: in essence, he adjusted the numbers for body weight to equalise them and adjusted rates of heart disease in the same proportion. Although the pattern A (BCD) was still seen with respect to some factors after adjustment, the pattern for smoking was now (AB) (CD), which meant that smokers were more prone to CHD than non-smokers. Hence he concluded that “this study failed to support the counter-hypothesis that constitutional factors or other environmental exposures explain the relation of cigarette smoking to total mortality or to mortality from coronary heart disease in middle-aged persons.”
Seltzer’s version of the rift with Friedman was this:
From the very inception of the above project, I was an equal co-investigator, a planner of its inception and procedures, a recipient of some of the data in process, and when I was suddenly presented with this manuscript by Dr. Gary Friedman through the mail without any prior consultation, I had no choice but to withdraw my name as a co-author on the grounds that there was too much in it that was scientifically unsound. (Private letter to the editor of the New England Journal of Medicine, 10 February 1979, requesting space for a criticism of Friedman’s paper).
Seltzer’s next publication was Smoking and coronary heart disease: what are we to believe? It concerns three topics: the Kaiser Permanente baseline study, the Surgeon General’s 1979 report, and Rose’s Whitehall study. The conclusions he drew from the baseline study were that “smokers destined to quit (ex-smokers) showed characteristics at baseline indicative of lower CHD risk than smokers destined to continue.” More generally, ex-smokers were not a representative sample of smokers because they are self-selected. As for the Surgeon-General’s report, the evidence it cited failed to establish the verdict it pronounced: and the Whitehall study clearly constituted evidence for Seltzer’s views.
Friedman struck back against Seltzer (though only actually naming him in a footnote) in Mortality in cigarette smokers and quitters. In his previous study he had controlled for factors such as body weight in subgroups A, B, C and D using the most recent data (i.e. after the quitters quit). He now controlled for the same factors as they were at baseline in persistent smokers and persistent quitters (A and C). His conclusion was that “Quitting smoking appears to result in a substantial reduction in coronary and total mortality that cannot be explained by the characteristics of quitters before they quit.”
Here Burch comes into the story with a letter condemning Friedman’s procedure and conclusions as fallacious, because he had controlled for factors such as body weight as if they had the same value in each subgroup. In fact the subgroups were self-selected and ought to be treated as samples of different populations. For this and related reasons, Friedman had come nowhere near to disproving the involvement of constitutional factors. A letter from Seltzer made the same point, contrasting the statistical methods of Fisher with the ‘black box’ methodology practised by Friedman (and demolished by Alvan Feinstein).
Friedman’s reply began “Investigators should do more than ask questions and raise doubts; they have a responsibility to attempt to answer their questions.” To Seltzer he said that the multivariate analysis he used was the only practical method of doing so. Not all the risk factors were associated with CHD, but that may have been due to “low statistical power”. To Burch he replied that “one must think about the biological correlates of self-selection” and that was the point of controlling for the risk factors. As for Whitehall, “before a negative result can rule out a benefit of quitting smoking, the experimental group must have a marked and sustained relative reduction in the level of smoking”.
Two issues are involved here. One is that Friedman is only prepared to include observed, quantified factors such as age and alcohol intake, not hypothetical ones only revealed to exist by statistical analysis. The other is that Burch saw Whitehall as a failure to falsify the null hypothesis and Friedman saw it as a failure to verify the alternate hypothesis. Friedman’s methodology is not Popperian falsificationism: but then, for those with eyes to see, risk factor epidemiology never had been about falsification.
As for the funding of the studies, it is true that industry funding of research generally comes with strings attached, but the same can be true of government funding. Seltzer recounted an experience of his own in a letter of 1979, indirectly relating to the Friedman affair.
I was the first person who found that there was a negative relationship between moderate alcohol consumption and the incidence of coronary heart disease. I discovered this relationship in the Framingham Heart Study data and drafted a manuscript with Drs. Kannell and Dawber for publication… Since the Framingham Heart Study data belonged to the U.S. Public Health Service, the manuscript was submitted to this organization for final approval for publication. Astonishingly, approval for publication was denied by D. William J. Zukel, Associate Director of the National Heart and Lung Institute, in part, on the grounds that the article “openly invites the encouragement of undertaking drinking. . . and socially undesirable . . . In other words, the National Institute of Health suppressed important information from their own study which was done in cooperation with their own top staff. The net result was that this draft manuscript was relegated to my files, where it has been ever since.
A few months later I informed Dr. Gary Friedman of the Kaiser-Permanente Group of my findings of alcohol and heart disease in the Framingham data and suggested he look into the matter in his own data. The net result was that in 1976 and 1977 Friedman published papers on this same subject giving the same result I had found previously.
Lee, reporting on the Friedman-Seltzer-Burch exchange to the TRC, judged Friedman to have had the best of it against Seltzer, since the factors he emphasised could not be causal confounding factors. As for Burch, Friedman was right to reject his “ludicrously stringent” methodology, and if Friedman did not comment on Burch’s work on CHD, “Probably Friedman did not want to get into an everlasting verbal battle on the issue”.