The Essence of Burch
3. Burch on coronary heart disease
Burch found it far easier to get a hearing for his work on heart disease than on lung cancer, partly because nobody, not even Doll, had claimed that smoking was the main cause. (That role had been allotted to a high-cholesterol diet, due to the work and political influence of Dr Ancel Keys.) Burch had already identified coronary heart disease as an autoaggressive condition in 1968. He returned to the question in two major journal papers published in 1978 and 1980.
Coronary heart disease, risk factors, ageing and time (Gerontology, 1978)
Ischaemic heart disease:epidemiology, risk factors and cause (Cardiovascular Research, 1980)
The diagnostic category of coronary heart disease, failure of the coronary arteries around the heart, was first identified in 1912, though particular manifestations such as angina were known long before that, and the category underwent various revisions thereafter. The most important one occurred with the transition from the seventh to the eighth International Classification of Disease in 1968, involving a major reanalysis and a new name, ischaemic heart disease.
The 1979 paper mainly discusses data sets from before 1968: the smoking studies of Kahn and Hammond, British government statistics from 1921 to 1973 and American statistics from the 1960s. The 1980 paper discusses post-1968 data, British and American government records again and also a dataset from Japan in 1975.
Burch’s general approach to CHD is exactly the same as with lung cancer. He sets up a hypothesis in the form of a curve derived from his autoaggressive theory of disease and compares it with actual data with a view to falsifying it. However, the specific issues discussed in the heart disease and lung cancer papers are very different, possibly illustrating the general power of the theory to unify apparently unrelated diseases, or possibly showing that it was so flexible that it could impose a pattern on anything.
Burch had little difficulty in fitting his data to the Weibull distribution with constant values for the parameters n, r and (for individual data sets) k. He also had empirical evidence for involvement of the immune system in the form of low leucocyte counts in subjects studied just before a heart attack. He did not need to invoke different values of n and r for men and women as he had in his work on lung cancer. None of the Burch curve diagrams in the two papers display a modal peak, a fall in death rates in high ages groups, but the curve is not quite a straight line, since it starts to tail off in old age. One difference between heart disease and lung cancer is that heart disease kills far more people, and their removal from the general population has to be taken into account. Burch’s analysis of the ICD7 data assumes a single type of CHD, but in the ICD8 data he claimed to detect different curves for five different types, including a rare early onset type affecting women only and a rare late onset form affecting Japanese males.
The big difference between the two investigations involves the invisible parameter λ, the latent period. Burch found that he could fit his curves to the data for lung cancer by taking λ to be an offset of 1.5 years. In his work on heart disease the matter is much more complicated. To a first approximation, Burch found that λ was an offset of 10-11 years in men and 20 years in women. This pattern, a latent period twice as long in women as men, characterised many of his claimed 200 autoaggressive diseases and he explained it in terms of the X and Y chromosomes, Say’s organ and other considerations which may or may not be correct. However the complications do not end there.
Coronary heart disease is statistically correlated with a number of ‘risk factors’ which modern epidemiologists of the school of Sir Richard Doll were apt to identify as causes. The main ones are
- hypertension (high blood pressure)
- hypercholesterolaemia (high blood cholesterol)
- body weight
In this context exercise means leisure exercise only, not the performance of hard physical work.
Most of the graphics in Burch’s two papers, and the data they are based on, seem to show that these risk factors correlate with fluctuations in the underlying latent period λ of about 10 years in men and 20 in women. Some of them also correlate with variation in the ‘susceptibles’ parameter S.
If Burch is right so far and CHD is fundamentally of genetic origin, what of the risk factors? On a causal interpretation, those that appear to increase the parameter S – these are cholesterol, hypertension and diabetes – precipitate a Rubicon event after which disease is inevitable. Those that appear to shorten the latent period λ seemingly hasten the onset of disease after the die is cast. However, they cannot all operate in the same way, since the apparent effect of body weight and exercise on λ grow stronger with age and that of the others weakens. The interactions between S and λ are intriguing. In subjects with both diabetes and hypertension the two factors seem to increase S twice over but λ no more than either does on its own. Exercise is good for you in that it seemingly keeps you out of the ‘susceptibles’ subset of the population, but if you are in it, the more you exercise the more likely you are to succumb to CHD.
To Burch, all this was evidence that the risk factors are not causes at all. He was struck by the fact that their correlation with CHD varied considerably between ethnic groups, which indicated that genetic factors were at work. Just as importantly, the risk factors are present in varying degree at different ages. From an analysis of the age distribution of hypercholesterolaemia and hypertension he concluded that they fitted Weibull distributions of their own and might well be autoaggressive diseases themselves. He also held, though he does not press the point in the heart disease papers, that smoking is an autoaggressive disorder. When he wrote “I regard compulsive smoking as a disease” he was not indulging in rhetoric, he meant it literally. Changing rates of smoking and fat consumption showed no obvious correlation with rates of CHD, and nobody had suggested a 30 year time lag like the supposed delay between onset of smoking and onset of lung cancer to complicate the matter. (This also made reverse causation a real possibility.)
Fluctuations in S needed separate explanation. As with lung cancer, misdiagnosis was a serious problem, but it was not essential to Burch’s argument. He cited research by others indicating that influenza was a very good candidate for a precipitating agent, and that it was the reason for falling incidence of heart disease in the 1970s in the USA but not Britain.
Burch knew that his work on heart disease raised important theoretical questions. One is that his observations on the statistics of the risk factors, and the conclusions he drew from them, are independent of the autoaggressive theory. In expounding his ideas he first presents his hypothetical mechanisms of forbidden clones, then sets up theoretical curves and tests them against empirical data. To the unwary reader it sometimes appeared that he was assuming what he set out to prove. An alternative strategy might have been to present the impressive fit of curves to data, explained them in terms of unknown entities to be hunted down like a new planet and then proceeded to identify forbidden clones as plausible physiological candidates.
The other point is Burch’s refusal on principle to assign confidence limits to his curves. The reason is the preliminary ‘tweaking’ he has to perform on the data to get a fit to the curves because of the latent period between onset and diagnosis. It is clear that there must be a latent period, that Burch treated data relating to the same disease consistently. However, the tweaking is different from one disease to another and Burch could offer no general theory of latent periods – indeed it is possible that there is none to be had. If, as he suggested in connection with CHD, the latent period itself fits a Weibull distribution, the problem of testing the general theory against the null hypothesis may well be intractable.
Burch’s studies of lung cancer and CHD apply the same general methodology to two very different diseases in very different ways. The question is whether, as he believed, he operated within severe constraints to unify seemingly unrelated phenomena, or whether he unwittingly allowed himself sufficient latitude to explain any data.