The Makarska Conference Panel on Causality in Medicine
During a great pause in the research directions of CVD prevention following NIH rejection of a Diet-Heart Trial, a conference was convened by principals concerned with the state of CVD epidemiology and public policy, a joint meeting of Working Groups on Mass Field Trials for the Prevention of Coronary Heart Disease. All took place in Makarska, the former Yugoslavia, in September 1968, under the auspices of the Research Committee of the International Society of Cardiology. Panelists in a Symposium on Causality in Medicine were Richard Remington, Professor of Biostatistics, University of Michigan; Ancel Keys, Professor and Director, Laboratory of Physiological Hygiene, University of Minnesota; Jeremiah Stamler, Professor, Northwestern University; Jerry N. Morris, Professor of Social Medicine, London School of Hygiene and Tropical Medicine; and Frederick Epstein, Professor of Epidemiology, University of Michigan.
Remington: “Those of you who have studied philosophy and the philosophy of science know that this word ‘cause’ has troubled philosophers for hundreds of years and that there are extensive writings on the subject. Here I want simply to make you doubt the concept of cause a little bit and to question it, to sense perhaps that the word is ambiguous and that we need to look at the structure that lies behind the concept of cause. I like to give several examples that may begin to shake your faith in the concept:
Infection with the polio virus and clinical polio is an example of a ‘necessary cause’: infection with polio virus is a necessary cause in the development of paralytic polio. You don’t get paralytic polio unless you are infected with the polio virus. The effect implies the cause, but most people infected don’t develop paralysis.
A major reduction in calorie intake is followed by weight loss. In this kind of causal pathway the cause implies the effect. It is a sufficient cause, sufficient to cause the effect, but not a necessary cause.
Infection with measles virus almost invariably produces clinical measles, thus it is both a necessary and sufficient cause.
Let’s look at lung cancer and cigarette smoking. Smoking is neither a necessary nor sufficient cause for lung cancer. There is a causal link but smoking is different from any of the other causes. Here we invoke the [now classic] criteria for causal inference.  First, cigarette smoking precedes lung cancer in time, thus there is temporality. Second, lung cancer is far more probable in the presence of smoking than in its absence. That shows both prediction and strength in the relationship. Third, the influence of underlying variables that affect both lung cancer and smoking must be ruled out or understood. Fourth is congruence with logical mechanisms. Fifth is coherence: the relation should fit existing knowledge and not contradict a well-established body of knowledge.”
Keys: After discussing experimental manipulation of variables, Keys says, “The Berkson School would maintain that only an experiment in manipulating smoking to cause lung cancer would establish causality. Now, of course we will agree that the greatest part of natural science is not susceptible to any such examination. There would be no science of geology or astronomy. But how are we going to make this more specific?”
Stamler: Also rejecting the Berkson concept that there is no proof without active intervention to test a hypothesis, Stamler says, “I agree with Ancel Keys that some of the richest experience in the history of science is in areas in which a wealth of theory has been developed and validated in a variety of ways without that [experimental] form of proof.”
Remington: “There will always be a core of science in any scientific area that is controversial, that is susceptible to argument— on which we will differ— whether the criteria for causal inference have been satisfied.”
Stamler brings up the causal significance of repeated sets of data yielding consistent evidence of association: “The notion from the original retrospective data that men with coronary disease had higher levels of serum cholesterol was a tenable notion from a variety of theoretical points of view as early as the first or second decade of this century.” He traces the sequence of knowledge about lipid in the walls of arteries and ingesting lipid that caused varying amounts of lipid in the blood, when already in the early 20th century a hypothesis could be put forward about diet influence on cholesterol and coronary disease. This evidence was bolstered, he showed, by more carefully done work post-WW II with retrospective evidence indicating the association between diet, serum cholesterol, and coronary disease and then by the prospective studies reporting a consistent relationship of blood cholesterol and coronary risk as reported widely in the late 50s and early 60s.
Stamler states: “When data come forth that are not consistent, so-called exceptions to the rule, there should be an ability to account for them.” He asks whether the exceptions purported to exist among the Masai or in Somali or in the Swiss Alps could be explained rather than their negating the rule, as claimed by George Mann: “Does an exception negate a rule? An exception does not necessarily negate the rule, but exceptions must be accounted for either by other underlying variables, by hereditary differences, or by differences [in the case of these primitive tribes] based on exercise or parasitic infestation or whatnot.”
Stamler addresses the criterion of congruence or consistence among epidemiological, clinical, laboratory and experimental data. Dietary cholesterol is an example when an apparent exception to the rule is not an exception after more careful work is done.
Stamler quotes Pasteur: “There are not two sciences; only science and its applications.” He points out further that a randomized trial or a mass field trial is not in itself final proof. “The final proof is a radical change in the course of the disease in the population. Even that is complicated to establish.” He gives the example of Darwin’s theory being accepted so rapidly in Britain because there had been 30 or 40 years discussion of it but that Lamarck had no reasonable mechanisms to propose whereas Darwin did.
Stamler discusses the plausible pathogenic sequence between diet, blood levels, and atherosclerosis and the lack of such a plausible sequence from salt ingestion to hypertension. “With common diseases we are dealing with causes that are neither necessary nor sufficient. We deal rather with multiple causes, none of which is necessary or sufficient but which make peculiar patterns with each other. I like the strict formulation in terms of the facts that are presently available. We advise: ‘On the basis of these facts we think it’s reasonable to try the following measures and to see what happens.’ I don’t think we have authority to tell the public any more than that.”
Epstein indicated that the evidence on cardiovascular risk factors was sufficient to recommend action to public health agencies. “But one can only tell the public that there is very strong circumstantial evidence for a relationship between these risk factors and coronary disease and we believe that their modification will lead to a reduction in incidence. Fortunately, in the case of the risk factors for coronary heart disease and ways of modifying these risk factors, there is no reason to suppose that what one recommends could do any harm, whereas there is a good chance that it would do good.”
Stamler suggests reformulating the question away from the individual case to epidemiology and the population: “Is there a cause without which. . ., on a large scale, premature atherosclerotic disease will never occur. Then we may speak about diet as a necessary cause, with other causes that are contributory.” He suggests that in the absence of the dietary factor, hypertensives in heavy cigarette smoking populations do not frequently get premature atherosclerotic disease. Neither will high physical activity [in the presence of the diet factor] protect from severe epidemic atherosclerosis. Reconstituting the issues in that way, he maintains, the human and animal data fit.
Jerry Morris analyzes the diet evidence differently, saying: “We end up postulating something so slight [the dietary relationship] that one comes out the other way and it is inconceivable that this [diet] is in fact the necessary and sufficient cause.”
Keys: “I couldn’t disagree more. The more carefully one looks the more secure is the information. I say this categorically: no one has ever been able to find a population subsisting on a diet even moderately low in saturated fat having a high frequency of severe atherosclerosis or coronary heart disease.”
Morris: “This is not my interpretation of the literature, but of course this is one of the cases that we want to discuss here. So let’s be off to our high cholesterol, high saturated fat dinner.”
[ed. Jerry Morris when interviewed 35 years later had not changed his mind one whit against diet being a “necessary factor.” (Henry Blackburn)]
To view the original Markarska document click here
Causal inference criteria for statistical associations are found in the introduction to the 1964 Surgeon’s General Expert Committee on Smoking and Health.