Behold, a new diet-health observational paper appears! (What, this one, again?)

There are two long term observational studies of diet and health that get more press coverage than all the others combined. New papers on these studies, which say the same things with minor variations, are published every few months in high-impact journals. The editors of these journals don’t seem to mind that the papers they get are unoriginal and the repetitive choice of the same studies amounts to cherry picking, because the authors are the rock stars of the nutritional establishment, and their university, Harvard, has a cachet that guarantees publication, just as an education at Eton guarantees influence in the British political establishment no matter how much of an idiot one happens to be.

Was that too critcal? You be the judge.

The studies are the Nurses Health Study (NHS), which had run for 30 years and is all-women (being a male nurse must have been an exclusion criteria for this study) and the Health Professionals Follow-up Study (HPFS) which had run for 26 years (and, you guessed it, included no female health professionals) when the results were analysed yet again to create this new paper.

Association of Specific Dietary Fats With Total and Cause-Specific Mortality
Dong D.Wang, MD, MSc; Yanping Li, PhD; Stephanie E. Chiuve, ScD; Meir J. Stampfer, MD, DrPH; JoAnn E. Manson, MD, DrPH; Eric B. Rimm, ScD; Walter C. Willett, MD, DrPH; Frank B. Hu, MD, PhD.
JAMA Intern Med. Published online July 05, 2016. doi:10.1001/jamainternmed.2016.2417

The mortality rate in these two studies, given their long duration, was quite high – 33 304 deaths among 126233 people. After adjustment for known and suspected risk factors, dietary total fat compared with total carbohydrates was inversely associated with total mortality (hazard ratio [HR] comparing extreme quintiles, 0.84; 95% CI, 0.81-0.88; P < .001 for trend). The HRs of total mortality comparing extreme quintiles of specific dietary fats were 1.08 (95% CI, 1.03-1.14) for saturated fat, 0.81 (95% CI, 0.78-0.84) for polyunsaturated fatty acid (PUFA), 0.89 (95% CI, 0.84-0.94) for monounsaturated fatty acid (MUFA), and 1.13 (95% CI, 1.07-1.18) for trans-fat (P < .001 for trend for all).

Taken at face value, this tells us that people who ate a higher-fat, lower-carb diet overall were less likely to die, over three decades, than people who didn’t. The amount of saturated fat people ate didn’t have much, if anything, to do with this.

So how was this reported in the media?

People who eat more saturated fat have a higher risk of an early death, according to a large study that contradicts recent claims that “butter is back”.

The study from the Harvard TH Chan School of Public Health in the United States has been following 126,000 people for three decades to assess the impact of their diet on their health and lifespan. The researchers claim it is the most detailed and powerful examination to date of the effects of eating different types of fats.

Its findings run counter to those of a study published by the National Obesity Forum in May, which said people should eat more fat and fewer carbohydrates and rubbished the nutritional guidelines from Public Health England, which recommend that people should eat less butter and red meat.

According, that is, to The Guardian. Last time we read the National Obesity Forum document, it was recommending that people at risk of diabetes and obesity eat more fat, mainly from wholefoods, while using olive oil and butter for cooking, and restrict carbohydrates, especially refined carbohydrates.

This is a dietary pattern that, if eaten by the people in the NHS and HPFS, would have increased fat in place of carbs (HR 0.84), increased MUFA (HR 0.89), increased PUFA (0.81), and decreased trans fats (HR 1.13). In other words, following the National Obesity Forum recommendations should have significantly decreased mortality. In this context, the HR 1.08 for SFA (a tiny 8% increased risk) is relatively meaningless. We’ll have a look at how it was arrived at next (spoiler, it was arrived at poorly).
But first it’s worth talking about trans fats. Trans fats are formed when unsaturated vegetable oils are partially hardened by hydrogenation to create spreads and shortenings that can be used in place of animal fats like butter, lard, and tallow. These industrial trans fats (mainly elaidic acid, the trans isomer of oleic acid) are associated with an increased risk of CVD mortality at relatively small intakes and are probably best considered toxic in all but naturally occurring trace amounts.
Industrial trans fats only exist in the food supply because of the widespread fear of natural saturated fat and animal fats. The people who warn us against trans fats today include some of the same people who made us eat them in the first place.

How does the Guardian article get around this inconvenient truth? By treating trans fats and saturated fats as if they are part of the same problem.

The new paper, published in the prestigious US journal Jama Internal Medicine, says eating trans-fats and saturated fats, including those from red meat and butter, is linked to higher mortality rates compared with the same number of calories from carbohydrates.

More importantly, they say, they found that death rates dropped by between 11% and 19% among people who substituted saturated fats such as butter, lard and red meat for unsaturated fats such as olive oil, canola oil and soybean oil.

The first claim is false; the trans fats in red meat and dairy are widely believed to be beneficial and weren’t the trans fats measured in the studies. Let’s look at that latter claim – it might surprise you to know that no-one in this study substituted any food for any other food (or rather, if they did, this wasn’t looked at). This is all imagination; and when these substitutions were tested for real, they didn’t result in lower mortality. Also, there’s a lot of unsaturated fat in butter, meat and lard. The unsaturated fat that’s supposed to be good for people in this study. It’s also important to note that Americans eat very little butter (it was 4.8% of total fat in 1955, before the scare started); the contribution of butter and lard to saturated fat intake in these studies would have been small, and in a recent meta-analysis of butter alone, the first ever done, it has zero correlation with cardiovascular disease.

How were the data analysed?

But how were the results obtained? (If interested, you can find detailed, and somewhat scathing, critiques, by other experts, of the methods used here on the PubPeer website, which go into much more depth). Participants in the study were asked, every few years, to remember what food items they had eaten in the previous year. Really. “In each SFFQ, we asked how often, on average, the participant had consumed a specified portion size of each food during the preceding year.” Oh dear. The NHS participants have reported, by this method, eating only an average ~1,500 kcal a day for 32 years. For the purposes of our analysis, we’ll have to reluctantly take these data at face value, but seriously? (an ongoing problem in nutrition research really, especially these types of cohort studies)

In the raw NHS data, women in the lowest quintile for saturated fat intake (average age 48.2 years) had more than double the death rate of women in the highest quintile (average age 45.9 years) (5852 vs 2332). So saturated fat intake was spread unevenly by age; younger women ate more of it (and people with low cholesterol ate more); or so it seems from the raw data (though the paper says otherwise); but would an extra 2.3 years really make such a huge difference? When both studies were combined, and the deaths were adjusted for age, the HR for saturated fat was 1.71 (1.65-1.78). This is quite a correlation – but there were a lot of differences, besides age, between the upper and lower quintiles for saturated fat intake. More smokers and less exercise in the upper quintile for saturated fat, for instance. When the data were adjusted for “known and suspected risk factors” the correlation was bumped down to the minimal HR of 1.08. The difference between 1.08 and 1.71 is +0.63. That is, there are things here that are associated with a 63% increase in mortality, and the authors had to pare those away to arrive at something associated with an 8% increase, which is what they decided to warn us about.
I don’t know about you, but I’m more interested in this 63% risk, which dwarfs anything any of these fats might be doing to us.

After age, these adjustments were made; this was the 63%.

Adjusted for white race, marital status, body mass index, physical activity, smoking status, alcohol consumption, multivitamin use, vitamin E supplement use, current aspirin use, family history of myocardial infarction, family history of diabetes, family history of cancer, history of hypertension, history of hypercholesterolemia, intakes of total energy and dietary cholesterol, percentage of energy intake from dietary protein, and menopausal status and hormone use in women.

Smoking  (for example) wasn’t evenly distributed among quintiles, and passive smoking, which was among the many possible “known and suspected” risk factors not measured, probably has a similar distribution to smoking. So does sugar consumption; though measured in these studies and previously associated with both saturated fat consumption and CVD mortality, this also wasn’t adjusted for. What’s interesting here is that these authors think that vitamin E supplement use is worth adjusting for. Polyunsaturated vegetable oils are the main source of vitamin E in modern SAD-type diets.

Why do smoking, inactivity, and no doubt sugar consumption and other unhealthy behaviours associate with saturated fat intake? After all, there’s no natural explanation for this. Steak doesn’t make you crave cigarettes, coconut oil doesn’t make people stop exercising. But this is a study involving two groups of health professionals, followed during the years 1980-2012. These people were TOLD, over and over again, that foods that were sources of saturated fat were unhealthy and foods that were sources of polyunsaturated fat were healthy.  We ask whether it’s ever possible to do statistical adjustments to model the independent associations of single variables (e.g., saturated fat) by various statistical techniques, given the public health messaging over the decades which inextricably confounds these results?
The people who cared about being healthy ate less butter and red meat, and used polyunsaturated oils and spreads. They exercised more and probably did many other things we don’t measure well or even at all. The people who couldn’t be bothered made less of these changes, were more likely to keep smoking and care less about second-hand smoke, probably drank too much and for all we know were more likely to have unprotected sex and use drugs – and on the list can go. They included the risk-takers, the contrarian ignorers of advice, the fun-lovers, and the pessimists and fatalists. This is called “healthy user bias”, and it’s always a problem when analysing observational studies of behaviours that society has strong opinions about.

Can we show that healthy user bias or residual confounding or the considerable potential for error built into the SFFQ system accounts for the 1.08, the remaining 8% of risk not seen in better quality studies of less easily biased populations? Saturated fat is believed to increase heart disease risk (the largest part of CVD mortality) by raising LDL cholesterol, you know the story. This is not a mechanism that anyone thinks plays a role in other causes of death; yet the Harvard study also found a positive correlation between respiratory disease mortality and saturated fat (HR 1.56; 95% CI, 1.30-1.87) and an inverse correlation with PUFA. This is a condition we’d expect to relate directly to smoking and air quality, even more so than CVD. It is likely to be an indicator of residual confounding (the authors describe this association as “novel”, so there’s no support for it in any other studies). The study also found this same pattern of associations (SFA mildly bad, MUFA and PUFA good) with all of the diverse causes of death studied, including cancer and neurodegenerative diseases; this is impossible to explain in terms of the lipid hypothesis, but easy to explain if high saturated fat consumption in this Standard American Diet population was associated with lower micronutrient and antioxidant intake, greater risk taking (including exposure to infections), higher sugar and refined grain intake, and greater exposure to smoking and environmental pollutants.

This reminds us of the study that found red meat consumption was significantly associated with accidental death in men. So beef makes you clumsy, or drive too fast, or not wear a seatbelt. D’you think? It’s much more likely that this indicates some residual confounding around risk-taking behaviours that weren’t properly measured and accounted for, men being overall bigger risk-takers than women. Unfortunately, there’s no accidental death data in the Harvard paper to compare.

We also found an earlier 20-year follow up study from the NHS alone which found no correlation between saturated fat and heart disease incidence in this group – HR 0.97 (95% CI 0.73, 1.27).

Is canola oil consumption associated with cancer mortality?

If you look at supplement 1 for this new Harvard study, which gives the results for men (HPFS) and women (NHS) separately, they are sometimes quite different, and are less significant, than the correlations in the main paper. A really interesting result that’s in the supplementary data is that omega-3 alpha linolenic acid (ALA) consumption is significantly associated with cancer mortality.
The HR for cancer mortality between low and high quintiles of ALA intake (table 14, p62) is 1.12 (1.04, 1.20), which is greater than the HR between saturated fat and all-cause mortality. ALA is in healthy foods like leafy green veges, pulses, some nuts and seeds, and whole grains, in small amounts. It’s not likely that eating these increases the risk of cancer. But ALA is present in larger amounts in canola oil and soybean oil, and these are the main sources of ALA in the US diet.
Go figure, there could be many reasons for this, but it’s odd that this correlation, which runs against the grain of any likely healthy user bias or residual confounding, appears tucked away in the supplement to a paper which is being used to support the idea that we should replace butter and lard with canola spread and soybean oil.
The correlation is mentioned briefly (without figures) in the main paper, which tries to minimise it by saying that the correlation in the “most recent” fat consumption table (table 7) is non-significant. But surely the whole point of a 32-year study like this is to try to capture the effect of dietary factors over a significant part of the lifespan.

It’s predictable that Harvard are beginning to confirm the benefits of a high-fat diet (yes, even in the context of the confounders of this study), but annoying that they harp on the evils of the saturated fat found in naturally occurring and nutritious foods, and continue to promote industrially refined vegetable oils and spreads (which is how trans fats got in our diet in the first place), based on evidence collected by methods which don’t seem that reliable to begin with. In the USA they don’t require labelling of trans fats in oils and spreads and even in NZ this is still voluntary. Even going by Harvard’s own research into the NHS and HPFS cohorts, there’s a better way to get any health benefits of PUFA and vitamin E.

Overall ? Just eat real food with a mixture of fatty acids. Avoid trans-fats and advocate for more funding for intervention research, where cause and effect is more easily determined and confounding exists less because of randomisation.