Search This Blog

Tuesday, 2 May 2017

Bradford Hill is rolling in his grave

Austin Bradford Hill was, as should be well known, the father of modern epidemiology, who played a key role in determining a causal relationship between smoking and lung cancer.
His 9 criteria (or viewpoints, as he called them) for evaluating epidemiological evidence were only ever a suggestion, and intended to have adaptable interpretations strongly guided by logic and good sense in any given context, but have stood the test of time despite the best efforts of epidemiologists to abandon and undermine them.
Initially an attempt was made to reduce the criteria to a smaller number of more malleable points with more room for guesswork and consensus, in the name of getting on with the business of identifying risks however small.

More recently, perhaps due to criticism, the full 9 criteria have been revived, and two recent efforts see them ticked off pedantically - in contexts which might well have bemused Bradford Hill.

Firstly, and I will only touch on this briefly, we have the "LDL is causal in CVD" paper.[1] Bradford Hill probably never considered that a class of biological particles present in every human being could be the cause of a common disease that is seen in individuals with widely varying levels of these particles. It's a little bit like finding platelets causal in thrombosis.

But even so, the paper commits a cardinal error.
None of my nine viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as a sine qua non. What they can do, with greater or less strength, is to help us to make up our minds on the fundamental question – is there any other way of explaining the set of facts before us, is there any other answer equally, or more, likely than cause and effect?

Is there any other explanation? To determine this, you need to also test the likelihood of the known alternatives. This the authors of the LDL paper do not do. Their paper does not mention insulin, ferritin, or the differing atherogenicity of the different classes of LDL particle and other lipoprotein particles such as VLDL or small, dense HDL, nor the oxidation status of the LDL particles. This is as if Hill had looked at a factory where the workers had a high rate of an unusual cancer, had been told that the workers were exposed to three or more novel chemicals, but had only decided to test the associations for one of them (perhaps the chemical that the company paying his wages made an antidote for). They seem to be arguing for the existence of a biological pathway, which few doubt has some relevance, but overlooking much that is also relevant, such as that the risk associated with LDL will not be decreased if the number is lowered by a method that increases the atherogenicity of the particles, that the association with LDL becomes protective as people age, and that lower LDL levels predict decreased survival in hospital after a heart attack, which may be the reason the FOURIER trial found absolutely no benefit in terms of mortality from extreme LDL lowering.

I have no wish, nor the skill, to embark upon philosophical discussion of the meaning of ‘causation’. The ‘cause’ of illness may be immediate and direct; it may be remote and indirect underlying the observed association. But with the aims of occupational, and almost synonymous preventive, medicine in mind the decisive question is where the frequency of the undesirable event B will be influenced by a change in the environmental feature A.With this in mind, we turn to our second new paper, which seems to risk making an opposite set of mistakes.[2] In this paper, in which the causality of foods and nutrients in cardiometabolic diseases is considered using the Bradford Hill criteria, every possible factor is tested, and most of them are found to be causal.
Perhaps if you can use the Bradford Hill criteria to assert causation for 17 different factors in the same disease you have also refuted each of them individually.
But what's interesting is that, even with this drift-netting approach, saturated fat is no longer making an appearance. Unfortunately we seem to lack the analysis that actually shows saturated fat failing the Bradford Hill criteria, the whole thing's a bit hush-hush for some reason.
We also see that the strength of the association is rated weak for PUFA, which is as it should be.
However red meat gets into their sights, which is unfortunate as people don't eat nearly as much red meat as they used to, yet diabetes, one of the conditions attributed to it here, is very much on the rise.

Their interpretation of temporality in general is weak; as well as one thing preceding another, it ought to take into account where possible the effects of duration of exposure on a disease; there are aetiological aspects to temporality (such as latency in cancer diagnosis) that are more complex than a simple longitudinal relationship. Diabetes is a disease of civilisation and red meat is an ancient food, an aspect of temporality which we probably also need to consider.  

The analogies given in Table 2 are not all convincing, many of them seem to refer to other relationships in the table or other associations that are still unproven. Bradford Hill's idea of an analogy was thalidomide and birth defects, an undeniable example of causation.

If we look at the reference list, we see a fair few Mediterranean diet papers and Harvard epidemiology papers featuring cohorts who were told that avoiding red meat was a healthy behaviour; in fact the sole evidence for the "red meat/processed meat and diabetes" claims is the Pan et al paper from 2011.[3] However 3 of the 10 studies in the Pan et al meta-analysis are their own NHS, NHS2 and HPFS studies, which use a cumulative averaging system that may give false results and data from a population of health professionals known to be influenced by advice about healthy behaviours (including advice given publicly by the study authors). If we remove (or combine) these 3 studies (all published together in this one paper) and combine the two Steinbrecher papers for males and females in the same population, we have 2 of 6 (or 7) favourable studies and 4 (or 5) unfavourable, a ratio which no longer meets the authors' test of consistency. In any case meta-analysis is a way of forcing the appearance of strength and consistency where neither may exist; it is probably most useful where exposures in a number of small, underpowered trials are identical (e.g. the same dose of the same drug for the same condition), and much less useful in diet epidemiology, with its already large populations and its data collection uncertainties.

If we turn to table 4 we see something alarming.[2] The recommended intake of PUFA is set at 11% of energy. This necessitates the use of oils. Yet only one country in the world has a PUFA intake this high - Bulgaria, where the age adjusted death rate for CHD is 188.45 per 100,000 of population ranking Bulgaria #21 in the world. Poland, a somewhat comparable country, sets a recommended PUFA intake of 3% (real intakes are higher) and has 136.72 CHD deaths per 100,000, placing at #40. The Tsimane' indians of Bolivia have very low PUFA intakes and experience a very low rate of cardiovascular disease, as do the Kitavans and as did the Tokelauan Islanders; high PUFA intakes are unusual in hunter-gatherers free from cardiometabolic disease. A PUFA intake of 11% is an unproven intervention, even the AHA doesn't recommend more than 10%.
The recommended meat intake of one serving a week is only met in Armenia and Georgia - two countries with very high CHD death rates. This is also a meat intake that will not supply nearly enough iron for women of childbearing age, ffs.
Barbados has the highest fruit consumption, as recommended, but diabetes is a major cause of death there.
This sort of arbitrary decision is not one that the use of Bradford Hill criteria allows anyone to make, especially when it is contradicted by this evidence supplied in the same table.

Such insanity aside, the dietary etiology Bradford Hill paper is probably intended as a well-meaning attempt to justify asking Americans to eat beans, nuts, and fish, which won't do them any harm; its danger is that it polishes up the Bradford Hill criteria into yet another tool that ideologues can use to suppress uncertainty, or justify the use of foods in contexts where they are biologically inappropriate (e.g. wholegrain products in the treatment of diabetes). If you don't respect the uncertainty in diet-health science, and the importance of context, you can't be right.

There's an earlier Bradford Hill dietary paper, by Andrew Mente and colleagues, which makes an interesting contrast with the current one.[4] Although in general agreement, albeit tougher, some associations that satisfy the criteria are for individual nutrients - vitamin E and vitamin C. In fact the vitamin E association is stronger than that for PUFA. Oils and other foods high in PUFA are generally good sources of vitamin E.

It may well be that sourcing expensive (or risky) foods and following exotic dietary patterns can protect us from disease. It may also be that the protective factors in foods are the ones we've always known about - the vitamins and minerals, electrolytes and trace elements, protein, essential fatty acids and so on, and that they do us most good when we find them in foods that won't dump energy into our bloodstreams any faster than the foods our ancestors ate thousands of years ago (which means that sourcing nutrients from fortified foods won't be optimal even if we could get the number of them and their balance right, which is far from being the case today). It may also be that other things in foods act as mild pseudomedicines (the polyphenols and other phytochemicals) or make up for deficiencies in our individual metabolisms (the carnochemicals).

This is what I propose as the null hypothesis of nutrition and health - that simple good feeding will give us most of the protection we need, that wandering away from it first with food refining and depletion, then with food processing (defined as the synthesis of replacements for degraded foods from more and more complex aggregations of equally refined food and non-food ingredients), is the cause of our modern cardiometabolic ills (insofar as these are due to diet and not other genetic and environmental factors) - not the fact that we instinctively cling to eating meat - the last surviving nutritious real food in all too many diets today.


[1] Ference BA, Ginsberg HN, Graham I et al. Low-density lipoproteins cause atherosclerotic cardiovascular disease. 1. Evidence from genetic, epidemiologic, and clinical studies. A consensus statement from the European Atherosclerosis Society Consensus Panel. Eur Heart J. 2017 Apr 24. doi: 10.1093/eurheartj/ehx144

[2] Micha R, Shulkin ML, PeƱalvo JL, et al. Etiologic effects and optimal intakes of foods and nutrients for risk of cardiovascular diseases and diabetes: Systematic reviews and meta-analyses from the Nutrition and Chronic Diseases Expert Group (NutriCoDE). PLOSOne April 27, 2017

[3] Pan A, Sun Q, Bernstein AM, Schulze MB, Manson JE, Willett WC, et al. Red meat consumption and risk of type 2 diabetes: 3 cohorts of US adults and an updated meta-analysis. The American journal of clinical nutrition. 2011;94(4):1088–96. pmid:21831992

[4] Mente A, de Koning L, Shannon HS, Anand SS. A systematic review of the evidence supporting a causal link between dietary factors and coronary heart disease. Arch Intern Med. 2009 Apr 13;169(7):659-69. doi: 10.1001/archinternmed.2009.38.


Blogger said...

New Diet Taps into Innovative Concept to Help Dieters LOSE 15 Pounds in Just 21 Days!

Galina L. said...

While researching materials about cancer for my blog article, I noticed that smoking was associated with a cancer ONLY when smokers ate refined carbohydrates. I titled the article "Cancer is just one of deceases of western civilization, or side effects of an English custom to drink a sweet tea"