How much evidence do you need to make recommendations about what the public should eat?
It depends really.
“On fair evidence we might take action on what appears to be an occupational hazard. For example, we might change from a probably carcinogenic oil to a non-carcinogenic oil in a limited environment and without too much injustice if we are wrong. But we should need very strong evidence before we made people burn a fuel in their homes that they do not like or stop smoking the cigarettes and eating the fats and sugar that they do like.”
This a quote from a great framework used in public health for making such decisions. It was put forward by Austin Bradford Hill in the 1960s, and has become known as the “Bradford Hill criteria“. It’s a set of conditions that should be met, or tests that should be made, before public health people start to make recommendations about what to avoid and what to do instead.
See also, Austin Bradford Hill, “The Environment and Disease: Association or Causation?”
Proceedings of the Royal Society of Medicine, 58 (1965), 295-300.
So what does this have to do with margarine?
In the previous post, we learned that New Zealanders on average consume around 4.9Kg of butter per capita each year, as well as a similar amount of palm oil, around 8.5Kg of canola oil, and around 2.7Kg of soy bean oil (a total of 21Kg of added fat, similar to the totality of 1966 butter intake). Much of the latter three oils goes into non-dairy spreads (along with smaller amounts of other oils such as corn, olive, rice bran, and sunflower, figures for which were not available). So what do we know about these oils and spreads, and their health effects, and should we be telling people to eat them especially over butter?
What are non-dairy spreads?
Butter is butter; its composition will vary slightly depending on what the animal is fed, so that winter silage produces a paler fat, lower in carotenoids, and the feeding of palm kernel expeller produces a fat in which the beneficial trans, cis fat rumenic acid (or CLA) is partly replaced by other trans fats, the importance of which is still uncertain, but these differences are very small compared to the differences that exist within the categories of margarine and non-dairy spreads. Although we’ll use the terms interchangeably here, food labelled “margarine” (a word few food producers seem to use today), as chemist Laurence Eyres reminded us in the Listener, must by law contain at least 80% fat, whereas spreads are usually lower fat. What Eyres doesn’t tell us is that this can be animal fat – there are budget spreads in the supermarket that contain beef fat as an ingredient. Not that there’s anything wrong with that, but obviously the idea that we can “replace animal fat with non-dairy spreads” is a bit misleading.
It’s curious that no-one who supports the substitution of margarine for butter mentions this, and the reason may be that the substitution exists in their heads as a theoretical one – they don’t actually go down to the supermarket and read the labels on the different products that people are buying and eating, or if they do, they only read the saturated fat information on the label.
A lot is made of the use of partially hydrogenated oils (PHO), a source of trans fats, in margarine, and how these are being removed from the food supply by a voluntary arrangement, with labeling still optional. Other countries have made greater efforts to label and remove industrial trans fats than New Zealand, Australia, and the UK. India introduced mandatory labelling and limits on trans fats from PHO within a short period, and the US FDA withdrew the GRAS (generally recognized as safe) classification from PHO, with a complete ban (barring any exemptions being granted) effective later this year. Note that other sources of saturated fats, and of naturally occurring trans fats, are still GRAS and always will be, so that attempts to combine “trans fat and saturated fats” into some common category of “bad fats” have no validity. But why was PHO included in margarine to begin with?
Butter has a spreadable consistency (at least at the right temperature), partly because its saturated fat content ensures that it is not too runny, and partly because the phospholipids and cholesterol it contains allow the fats to form an emulsion with its small amount of water. The trans fats in PHO were straight chains like saturated fats, so had a similar consistency, while appearing as unsaturated fats when tested in the laboratory (they also, unlike saturated fats, interfered with the conversion of polyunsaturated fats into various signaling molecules, which was a bad thing). So they have had to be replaced with more saturated fats, such as beef fat or palm oil. Lighter spreads that don’t contain these fats need to include emulsifiers and stabilizers; this gives them more in common with other highly processed foods, and also means that you’re paying extra for water. The technology of interesterification means that oils can now be made harder by switching fatty acids around on the glycerol backbone of triglycerides, changing their interactions with one another and thus their consistency, but this technology is only used in some of the spreads on the NZ market.
Vegetable spreads – what evidence is there for benefit?
So is there anything (causally) harmful or good about these products? Do they have health benefits? Bearing in mind the Bradford Hill criteria which look at the scientific evidence. The evidence should be tested against these criteria:
- Strength (effect size): A small association does not mean that there is not a causal effect, though the larger the association, the more likely that it is causal.
- Consistency (reproducibility): Consistent findings observed by different persons in different places with different samples strengthens the likelihood of an effect.
- Specificity: Causation is likely if there is a very specific population at a specific site and disease with no other likely explanation. The more specific an association between a factor and an effect is, the bigger the probability of a causal relationship.
- Temporality: The effect has to occur after the cause (and if there is an expected delay between the cause and expected effect, then the effect must occur after that delay).
- Biological gradient: Greater exposure should generally lead to greater incidence of the effect. However, in some cases, the mere presence of the factor can trigger the effect. In other cases, an inverse proportion is observed: greater exposure leads to lower incidence.
- Plausibility: A plausible mechanism between cause and effect is helpful (but Hill noted that knowledge of the mechanism is limited by current knowledge).
- Coherence: Coherence between epidemiological and laboratory findings increases the likelihood of an effect. However, Hill noted that “… lack of such [laboratory] evidence cannot nullify the epidemiological effect on associations”.
- Experiment: “Occasionally it is possible to appeal to experimental evidence”.
- Analogy: The effect of similar factors may be considered.
Criteria 1 and 2: Strength and consistency of epidemiological evidence
There is a body of epidemiological research that claims that replacing saturated fat (or carbohydrate) with polyunsaturated fat reduces cardiovascular risk. Unfortunately, the association isn’t consistent (there are populations where no association, or the opposite association has been seen, at least from theoretically replacing saturated fat with polyunsaturated fat).[1, 2] However, even in the populations where this association has been seen, there’s no clear evidence that margarine, or cooking oil, is the source of it. The problem is that many minimally refined foods are also good sources of polyunsaturated fat, especially chicken and pork, nuts and seeds, olives and avocadoes, but also meat and dairy; all whole foods add more polyunsaturated fat to the diet than sugar and flour do. There don’t seem to have been many attempts to isolate the polyunsaturated fat in oils and spreads and compare it with that in whole foods and animal fats.
There have only been a few epidemiological studies comparing margarine with butter. In Framingham, which was the original large population followed to test the lipid hypothesis (it didn’t work out – there was never any neat linear association between saturated fat in the diet, LDL cholesterol, and heart disease), using margarine instead of butter was associated with no change in heart disease during the first 10 years, then an increase over the 10 years following. This was attributed to trans fats, but that doesn’t explain the different effects over 10 and 20 years.
“Adjusted for age and energy intake, the risk ratio for CHD for each increment of 1 teaspoon per day of margarine was 0.98 [95% confidence interval (CI) = 0.91-1.05] for the first 10 years of follow-up and 1.10 (95% CI = 1.04-1.17) for follow-up years 11-21. Butter intake did not predict CHD incidence.”
Recently, a study was published claiming a theoretical benefit from margarine use instead of butter for heart attack (MI) risk over a 13 year period.
“Substituting butter or stick margarine with tub margarine was associated with lower risk of MI (HRs = 0.95 and 0.91). Subgroup analyses, which evaluated these substitutions among participants with a single source of spreadable fat, showed stronger associations for MI (HRs = 0.92 and 0.87). Outcomes of total CHD, ischemic stroke, and atherosclerosis-related CVD showed wide confidence intervals but the same trends as the MI results.”
Unfortunately, this study is one we don’t have access to, and the methods are not obvious; however, no number of events is given in the abstract, so it is possible that “theoretical” just means calculated risk from serum lipids. Still, it’s noticeable that even in this study, substituting one type of margarine for another was associated with more benefit than replacing butter.
(Given the variety of non-dairy spreads on the market, and the inconsistency of their ingredients, it might be more helpful if the experts fought over what sort of margarine people should use, rather than whether they should use it instead of butter).
There may be other evidence that using margarine, specifically, is associated with beneficial outcomes, but if so we haven’t been able to find it.
Criteria 4 and 6: Temporality and Biological plausibility
What about oils? There’s an interesting paper from the Nurse’s Health Study that looked at two different sources of the omega 3 fat ALA. A higher consumption of ALA from (soy) oil-and-vinegar salad dressing was associated with a lower risk of fatal heart attacks, RR 0.46 (0.27, 0.76), whereas a higher consumption of mayonnaise was not, RR 0.84 (0.50, 1.44). Yet mayonnaise is the richer food and contributed more ALA to diets than oil-and-vinegar salad dressing (16.7% vs 12.2%). This study didn’t look at intake of margarine as a source of ALA (6.8%). A relevant point is that much of this population was probably deficient in omega-3 fats (American’s don’t eat much fish, and median daily energy-adjusted ALA intake ranged from 0.71 g in the lowest quintile to 1.36 g in the highest quintile), so we are probably looking at the effects of correcting a deficiency of an essential nutrient. Of course, if you’re using lots of oil-and-vinegar dressing, or, to a lesser extent, mayonnaise, you’re also eating a very different type of diet from someone who isn’t.
In this case we have a plausible mechanism, and a suggestion of temporality – correcting a historical deficiency of omega-3 fats in a population with very low intake of them would be expected to improve the blood clotting aspect of fatal heart attacks. Other foods could have supplied the ALA, but soy oil happened to be the source available. The association was only significant when the oil was used in salads, and stronger in women taking vitamin E supplements, but in these cases it satisfied the Bradford Hill criteria for strength, and is broadly consistent with the RCT analysis of Ramsden et al.
This example seems to satisfy all Bradford Hill criteria to some extent, if considered as an correction of a deficiency analogous to the correction of a vitamin deficiency, but much less so if considered as the effect of a substitution for saturated fat (especially considering the body of evidence that a substitution for carbohydrate would be at least as beneficial).
On balance we’d say there isn’t a case as far as Bradford-Hill’s criteria are concerned to say anything about spreads and benefit.
Vegetable oils and spreads – some evidence for harm?
So – is there any evidence that non-dairy spreads and cooking oils have harmful effects? You’ll find plenty of mechanistic arguments and non-human experimental evidence that they do in the literature. But remember Bradford Hill’s criteria – an epidemiological association is convincing when it’s attached to mechanisms, but is also strong (RR approaching 2 or greater is best, though 1.5 is usually accepted), consistent (not contradicted by directly comparable studies), and has a dose-response (more seems worse).
Criteria 3, 5, and 7: specificity, biological gradient, and coherence
Age-related macular degeneration is a common cause of visual impairment and blindness in older people.
A case-control study gave the following results.
Higher vegetable fat consumption was associated with an elevated risk for AMD. After adjusting for age, sex, education, cigarette smoking, and other risk factors, the odds ratio (OR) was 2.22 (95% confidence interval [CI], 1.32-3.74) for persons in the highest vs those in the lowest quintiles of intake (P for trend,.007). The risk for AMD was also significantly elevated for the highest vs lowest quintiles of intake of monounsaturated (OR, 1.71) and polyunsaturated (OR, 1.86) fats (Ps for trend,.03 and.03, respectively). Higher consumption of linoleic acid was also associated with a higher risk for AMD (P for trend,.02). Higher intake of omega-3 fatty acids was associated with a lower risk for AMD among individuals consuming diets low in linoleic acid, an omega-6 fatty acid (P for trend,.05; P for continuous variable,.03). Similarly, higher frequency of fish intake tended to reduce risk for AMD when the diet was low in linoleic acid (P for trend,.05). Conversely, neither omega-3 fatty acids nor fish intake were related to risk for AMD among people with high levels of linoleic acid intake.
This was followed by a prospective study by the same authors looking at AMD progression (important because temporality, the presumed cause preceding the effect, which cannot always be shown in case-control studies, is another of the Bradford-Hill criteria), which confirmed the associations above, but also found that nut consumption was protective. This is interesting, because nuts are a good source of linoleic acid, indicating that linoleic acid in whole foods (where some of it is in phospholipid form) may behave differently from linoleic acid in vegetable oils, where the phospholipids have been removed by refining.
The prospective study found that all fats were associated with AMD – however, there was no dose response in the case of saturated fats. Dose response is an important part of the Bradford Hill criteria – if something is truly harmful, more should (usually) do more harm than a little.
One study, NHANES II, did not find a significant correlation between dietary fats and AMD, however this study failed to separate vegetable oils from other sources of fat, and did not provide a sufficiently detailed breakdown of its results.
This is an example that fulfills many of the Bradford Hill criteria. Indeed it already appears in textbooks as an example of the role of peroxidation in oxidative stress diseases. It has specificity because the retina of the eye is exposed to concentrated UV radiation, and plausibility because linoleic acid and ALA are uniquely prone to peroxidation in vivo. It is known not to be the only factor in AMD, yet shows a strong association, increasing our confidence that it is a factor increasing risk.
One of the predicted harms of higher vegetable oil intake is increased cancer risk, due to the peroxidisabilty of polyunsaturated fats, as well as their use by some types of cancer cells. However, this correlation is not usually a strong or consistent one, because exposure to carcinogens and diet both vary by occupation and socioeconomic class, and because cancer, unlike heart disease, subdivides into many discrete diseases, the risk of which is relatively rare. Further, just agreeing to be in a diet study has a significant effect on reducing future cancer risk. However, there are some exceptions.
The use of vegetable oils for high temperature cooking is fairly consistently associated with lung cancer risk in women using them, for example in this case-control study of women in Gansu, China using rapeseed oil (as far as we can tell, this is the name for canola oil in modern China; at any rate, it comes from the same species and is thought to be healthy) for wok cooking.
A useful feature of this study was that rates of two important confounders, deep fryer cooking and smoking, were very low.
The odds ratio (OR) for lung cancer associated with ever-use of rapeseed oil, alone or in combination with linseed oil, was 1.67 (95% CI 1.0-2.5), compared to use of linseed oil alone. ORs for stir-frying with either linseed or rapeseed oil 15-29, 30 and > or =31 times per month were 1.96, 1.73, and 2.24, respectively (trend, P=0.03), relative to a lower frequency of stir-frying. Lung cancer risks also increased with total number of years cooking (trend, P<0.09). Women exposed to cooking fumes from rapeseed oil appeared to be at increased risk of lung cancer, and there was some evidence that fumes from linseed oil may have also contributed to the risk.
In a Hong Kong case-control study in which deep frying was a common style of cooking, the risk was even higher – indeed, similar to that of smoking. Peanut, corn, and canola were the main oils used (there was little difference between them in this study).
The ORs of lung cancer across increasing levels of cooking dish-years were 1, 1.17, 1.92, 2.26, and 6.15. After adjusting for age and other potential confounding factors, the increasing trend of ORs with increasing exposure categories became clearer, being 1, 1.31, 4.12, 4.68, and 34. The OR of lung cancer was highest for deep-frying (2.56 per 10 dish-years) followed by that of frying (1.47), and stir-frying had the lowest OR (1.12) among the three methods.
The first study is curious because linseed oil, an oil very high in ALA, and one which no-one in the West would use for cooking, seemed safer than rapeseed oil. In the recent Harvard study on dietary fat and mortality, based on Nurse’s Health Study data, ALA was associated with cancer mortality (HR 1.12; 95% CI, 1.04, 1.20). There was both a dose-response and a temporal response (the association was only seen with longer exposure). This association is weak, but lung cancer is only one type of cancer, canola oil only one source of ALA, and ALA rich foods can be used in various ways. The Harvard research analyses foods, rather than the ingredients they are made with, so there’s no information on how much canola oil was being used, but in NZ it seems to be the most common cooking oil. However, the idea that cooking oil fumes cause lung cancer seems to be accepted; yet, here in New Zealand, we’re told to use canola oil in place of animal fats or coconut oil (saturated fat wasn’t associated with cancer mortality in the Harvard study).
This is a case that not only fulfills Bradford Hill criteria for causality (the product of heating the ALA in these oils protects lung cancer cells against apoptosis in vitro), but also fulfills his example of a practical intervention. We could look for less probably carcinogenic oils and fats and introduce (or re-introduce) them into the limited environments of kitchens without too much injustice if we are wrong, provided they are no more harmful once eaten.
Since the removal of PHO trans fats from the food supply, the use of palm oil, the vegetable oil highest in the main saturated fats palmitic and stearic acid, has increased in NZ. Palm oil consumption is equal to that of butter, and most of it goes into non-dairy spreads and other processed foods. The “Myth Buster” section of the Listener article we responded to in the last post failed to mention palm oil use in spreads. Palm oil contains a greater amount of the supposedly “bad” saturated fat palmitic acid than butter, thus defeating the point of using margarine. (However, as the amount of palmitic acid in the blood is controlled by the amount of carbohydrate we eat, and we usually eat much more carbohydrate than palmitic acid, it makes more sense to reduce carbohydrate first, and avoid palm oil, before looking at butter. Coconut oil is very low in palmitic acid). Palm oil production has become a far greater environmental disaster internationally than dairying is said by its worst critics to be for New Zealand. To make matters worse, the European Food Safety Agency recently issued a warning that palm oil contains higher levels of the carcinogen glycidyl, caused by refining at temperatures above 200 degrees C, than those found in other processed oils. New Zealand is still awaiting legislation for the mandatory labelling of palm oil, which often appears on food labels as “vegetable oil”. Remember, the New Zealand population has only been exposed to trans fat, palm oil, and interesterified fat products because locally produced animal fats like butter have been supposed, on inadequate evidence, to be harmful, and because a massive international industry exists to profit by supplying us with artificial substitutes.
So there could be a future increased health risk, as well as the ongoing environmental disaster, caused by our rush to consume palm oil, all in the name of avoiding butter and animal fats. However, there’s no epidemiological study looking at the impact of palm oil in Western countries (nor any for interesterified fats). Bradford Hill can’t help us when studies don’t ask the right questions. Most of the studies we’ve looked at weren’t designed to tell us whether getting our polyunsaturated fats from refined oils instead of real foods was a good idea or a bad one, though we can be pretty sure that they’re not as healthful as they were once heated above 200 degrees C, whether in cooking or refining.
We love the way that Bradford Hill’s thinking does more than just help us decide whether to say yes or no to the question of “does X cause (or prevent) Y?”- it can also lead us into a deeper understanding of what the evidence really means.
The case for benefit from PUFA oils and spreads isn’t met, except for the case for correcting a deficiency of LA or ALA – directly analogous to correcting a vitamin deficiency (analogy). Regarded in this way only, the case for benefit should be clear, but can only apply to oils and spreads in cases where oils are the only source of these fats – this doesn’t have to be the case for ALA, and of course it is not the case for LA in normal diets.
Anyone eating a higher fat, real food diet is likely to have an optimal intake of LA and at least a sufficient intake of ALA or other omega-3 fats without relying on these supplementary sources.
For harm from oils and spreads, there is no strong general case for harm, but there is a case when some oxidising stressor is introduced. This includes UV light (in the eye) for AMD, and heat (in the pan) for lung cancer. PUFA manifests this harm in free oil form (there is an analogy here with experimental alcoholic liver disease, where high-PUFA oils in the diet are convincingly harmful, but PUFA in phospholipids can be protective, as are saturated fats, in a context of oxidative stress; we haven’t covered this evidence because its epidemiological component is so far minor).
And then there are the harms untested – novel foods should be monitored better than they are. Absence of evidence isn’t evidence of absence. The effect of heat on palm oil fats, and to a lesser extent other oils, during refining predicts harm for human health, but epidemiological evidence that confirms (or refutes) this may take decades to appear.
A low-butter spread you can make at home.
Supposing you like the taste of butter, want to avoid potentially harmful refined oils, but are also interested in reducing the amount of saturated fat you eat. Or maybe you want to do your bit to reduce the environmental impact of dairy farming. We suggest adapting a recipe for “Sunbutter” from Gayelord Hausers'”Treasury of Secrets” (1963 edition).
Gayelord Hauser was a diet adviser to the stars in the golden age of Hollywood, who advocated (among other things) low carb diets for weight loss, and the use of what later became known as “health food supplements” such as brewer’s yeast, molasses, and wheat germ oil. His books contain a simple recipe for a low-saturated fat spread, using a pound of butter and a cup of sunflower oil. In our new updated version, you’d heat and blend together equal parts butter and extra virgin olive oil. The result is spreadable straight out of the fridge, and tasty. This might just be the main benefit of the recipe. NZ refrigerators used to have a butter conditioner – a warm section of the fridge, just the right temperature to keep your butter spreadable. These went out with the anti-butter, saturated fat will kill you campaigning of the Heart Foundation and Rod Jackson in the 1990s. Such a loss…
Unlike the original Sunbutter, this mixture will have almost exactly the same fatty acid composition as your own fat stores (credit to Steve Phinney for this insight), and that’s exactly the kind of healthy fat you want to be running on.
And you don’t need to use anything to make it that you wouldn’t normally want in your food.
 Praagman J, de Jonge EA, Kiefte-de Jong JC, Beulens JW, Sluijs I, Schoufour JD, et al. Dietary Saturated Fatty Acids and Coronary Heart Disease Risk in a Dutch Middle-Aged and Elderly Population. Arterioscler Thromb Vasc Biol. 2016; 36(9): 2011-8.
 Praagman J, Beulens JW, Alssema M, et al. The association between dietary saturated fatty acids and ischemic heart disease depends on the type and source of fatty acid in the European Prospective Investigation into Cancer and Nutrition-Netherlands cohort. Am J Clin Nutr2016;103:356-65.
 Margarine intake and subsequent coronary heart disease in men. Gillman MW, Cupples LA, Gagnon D, Millen BE, Ellison RC, Castelli WP. Epidemiology. 1997 Mar;8(2):144-9.
 Liu Q, Rossouw JE, Roberts MB, Liu S, Johnson KC, Shikany JM, Manson JE, Tinker LF, Eaton CB. Theoretical Effects of Substituting Butter with Margarine on Risk of Cardiovascular Disease. Epidemiology. 2017 Jan;28(1):145-156.
 Hu FB, Stampfer MJ, Manson JE, et al. Dietary intake of α-linolenic acid and risk of fatal ischemic heart disease among women. Am J Clin Nutr. 1999; 69(5): 890-897.
 Use of dietary linoleic acid for secondary prevention of coronary heart disease and death: evaluation of recovered data from the Sydney Diet Heart Study and updated meta-analysis
 Seddon JM, Rosner B, Sperduto RD, et al. Dietary fat and risk for advanced age-related macular degeneration. Arch Ophthalmol 2001;119:1191–9.
 Seddon JM, Cote J, Rosner B. Progression of age-related macular degeneration: association with dietary fat, transunsaturated fat, nuts, and fish intake. Arch Ophthalmol 2003;121:1728–37.
 Heuberger RA, Mares-Perlman JA, Klein R, et al. Relationship of dietary fat to age-related maculopathy in the Third National Health and Nutrition Examination Survey. Arch Ophthalmol 2001;119:1833–8.
 Metayer C, Wang Z, Kleinerman RA, et al. Cooking oil fumes and risk of lung cancer in women in rural Gansu, China. Lung Cancer. 2002 Feb;35(2):111-7.
 Yu IT, Chiu YL, Au JS, Wong TW, Tang JL. Dose-response relationship between cooking fumes exposures and lung cancer among Chinese nonsmoking women. Cancer Res. 2006 May 1;66(9):4961-7.
 Wang DD, Li Y, Chuive SE, et al. Association of specific dietary fats with total and cause specific mortality. JAMA Intern Med. Published online July 5, 2016.
 Panel on Contaminants in the Food Chain. Risks for human health related to the presence of 3- and 2-monochloropropanediol (MCPD), and their fatty acid esters, and glycidyl fatty acid esters in food. EFSA Journal 2016;14(5):4426 [159 pp.].