Statistics and fat


This excellent post just reminded me of a blog I wrote earlier this year.

Both point out the problems with this whole “saturated fat is going to give you heart disease” thing. The data on inspection just don’t add up.

Not only do they not add up, they sometimes show the opposite of what is claimed. But, after statistical adjustment for possible confounders we see something.

Here’s what I mean – the big combined cohort paper from Yanping et al (including Walter Willett as senior last author) claims that saturated fat is still the bad guy and replacing it with either good quality carbs or polyunsaturated fat will reduce the risk of heart disease.

In the blog Doc’s Opinion Dr Sigurdsson combines data in Yapping et al’s Table 2 from the two cohorts used in this study. The numbers below show the combined number of fatal Coronary Heart Disease events and non-fatal MIs (heart attacks) in both cohorts. One after 24 years of follow up the HPFS cohort, and a second after 30 years of follow up – the NHS cohort.

What’s shown here are the combined number of CHD events for the highest and lowest 20% of intakes of various types of nutrients.

You’ll see the actual combined raw data run contrary to their final conclusions on a number of key nutrients.

I especially draw you attention (see table below) to the LOWER CHD in people who ate more fat (1420 v  1641 events), more saturated fat (1434 v 1599), and less total carbs (1378 v 1717).


The thing is these are the actual number of people who had actual heart disease (death or non-fatal heart attack).

But the dietary differences were confounded at baseline. That is, multiple differences co-occurred… “At baseline (Table 1), men and women with a high intake of SFAs as a percentage of energy were slightly younger, had a higher body mass index, had a lower prevalence of physical activity and multivitamin use, and consumed more cholesterol. Participants with higher energy intake from SFAs also tended to have higher energy intake from MUFAs and trans fats and lower energy intake from carbohydrates (Table 1). ”

So they used some statistical techniques to adjust for these differences.

“To obtain overall estimates for both sexes and to increase statistical power, the HRs [Hazard ratios] from the age- and multivariable-adjusted models from the 2 cohorts were combined with the use of a fixed-effects inverse variance-weighted meta-analysis because no significant heterogeneity between the cohorts was observed.

What this means in the end is that they claim that saturated fat causes heart disease. When in reality these effects never occurred. In fact, you’d have to say that the unadjusted data supports the idea that we should eat more total fats including SFA and PUFA, less carbs and sugar, and if you are eating carbs eat good quality (unprocessed) ones. Yet these raw effects are never dealt with.

Remember these raw data are actual people suffering actual heart disease. The rest of the adjusted models are only theoretical, allowing for the confounders as they choose.  I’m all for statistical adjustment, but not when you end up making claims that seem to bear no resemblance to reality, because the original association has somehow been reversed.

What I’d take from this paper – there is evidence that what we eat affects our health.  One indicator is heart disease. Several things seem to interact, which isn’t surprising because when you reduce energy from one nutrient you generally will replace it with another. Likely issues in these two cohorts are too much carbohydrate (and low carbohydrate quality) and not enough fat.

Why can’t there be some agreement on this part?  Why do we revert back to “dietary guidelines are still right and we should avoid saturated fat”? Defence of the status quo under such conditions depends on tolerating the dissonance that this generates. This “vision of the anointed”, or rather lack of vision, is still a major issue for public health nutrition.

Whatever the adjustment and statistical techniques, these data don’t show causation. We need experiments for that. Such dietary experiments are hard (and expensive) to do. In the past, several such trials were done for disease prevention (heart disease or cancer) which involved various versions of the low-fat diet, with mediocre results and no effect on mortality from either disease. Any of these longer term preventive trials could have included a low carbohydrate arm, but none did. However, we do see medium term effects of exactly what the raw epidemiological data in Yanping et al shows – that eating more fat, less and better quality carbs is generally a good thing.

Bottom line: Saturated fats as they occur in whole plant and animal foods represent no serious public health risk worth making recommendations about. 


  1. I dug down into the studies and what I found makes the whole exercise even more murky.
    They have combined two studies which started a long time ago so we can forgive the sexist assumptions behind them – the Nurses Health Study is the cohort of women, and the Health Professionals Follow-up Study is that of the men.
    The associations for these two studies are actually quite different – opposite, in most things – so why did the Harvard boffins pool them?
    Saturated fat is protective in the women (NHS), and Willett actually wrote a famous paper about this, but not in the men. Whole grains are protective in the men but not the women. (I wonder why? could it be that wholegrains, which are very low in iron and also iron-chelating, are healthier for men than refined grains, which are iron-fortified – but not necessarily for women? Men tend to get high iron levels easily, women are prone to iron-deficiency).
    Another thing that muddies the water in NHS and HPFS is that saturated fat intakes dropped by half over the course of the study in these populations because of dietary guidelines and cholesterol phobia. Yanping et al don’t seem to mention this. They do say they came up with some way of averaging intakes over the years, but who knows if it was the best choice.
    If we look at the Malmo Diet and Cancer Study, which frankly I would rather do as the data collection methodology is much better, it’s men, not women, who have a protective association between total fat and CHD mortality, and it’s quite a big one (and men get most heart attacks in all populations, so this matters more to population health statistics). So the gender difference seen by comparing NHS and HPFS in the US is not supported by studies in other populations.
    Another thing I don’t get is Yanping et al saying we need to substitute this extra 5% energy from PUFA for this, that, and the other thing.
    If you look at NHS and HPFS, sure there is a protective effect of PUFA, but almost all the participants (except the bottom quintile or lowest 20%) are well within 5% of the amount associated with most benefit. The population average of PUFA consumption in most countries today is well within 5% of the consumption in the upper quintile of Yanping et al. If everyone increased PUFA intake by 5%, we would be at a level of PUFA intake that is untested by this Harvard epidemiology. In Malmo, on the other hand, there’s no association between PUFA and heart disease, maybe because intakes are generally adequate overall.
    Also what’s lacking in Yanping et al is information about all-cause mortality or other causes of death. We would ideally want to know that we’re not swapping one risk for another.
    What if the various fat types are just markers of dietary quality? Where people eat most saturated fat in dairy, SFA has beneficial associations, where people eat most SFA in baked and sweet food it has negative associations. Where people eat MUFA in olive oil and nuts it’s good, where they eat it in cheap margarine it’s not. This might help to explain some of these inconsistencies. Certainly, the Harvard crew think that this is the case with carbohydrate sources, so why aren’t they as open to the idea of it also being true of sources of fat, including sources of saturated fat?

  2. Being a skeptic, I almost always attempt to find out who funds the doctors making claims from studies. In particular, who is providing the grants. The outcome of any study will be aligned with the interest of the funding source. I ran a couple of Google searches on Yanping funding, but found only limited info.

  3. Spittinchips · · Reply

    Excellent post, and you can always count on George to add a frivolous and totally non-helpful comment…but curiosity is killing me – what is added to Beale’s lard for stability?

    Have a good one.

  4. It says “oxygen interceptor added to improve stability”, in other words an early antioxidant.
    I reckon it’s BHA – Butylated hydroxyanisole

    “Since 1947, BHA has been added to edible fats and fat-containing foods for its antioxidant properties as it prevents rancidification of food which creates objectionable odors.”

    It’s supposed to be a carcinogen, but the evidence for this is inconsistent.

  5. […] Blame For Your Acne 7. Paleo Pub Favorite: Fish and Chips! 8. Recipe: Baked Potato Shakshuka 9. Statistics and fat 10. Why I Choose To Skip Breakfast Every […]

  6. Hi! I am looking to start a keto diet (asap!) but I have quite a few questions because I don’t eat meat. I would love if you could help me in anyway that you could! Thank you🙂

    1. Totally doable without meat esp if you are eating fish eggs etc?

  7. […] Saturated fat is bad (if you ignore the raw data). […]

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Richard David Feinman

Richard Feinman, the Other

The Science of Human Potential

Understanding how to be the best you can be. Professor Grant Schofield.


A topnotch site

LCHF, Diet & Health


Eat real food. Enjoy real health.

%d bloggers like this: