On Nurses, Pork Fat and Potatoes

Dr. Walter Willett is a guy who almost gets it. The Harvard physician is maybe the most outspoken mainstream researcher in questioning the tenets of the USDA Food Pyramid. The results of his studies frequently call into question the wisdom of the dietary guidelines provided to us by the USDA, and Willett is not afraid to speak up about what he sees as flawed dietary advice. For instance, he questions whether we should be relying on carbohydrates for so many of our calories. And he says the pyramid's insistence that all fat is bad is wrong - he emphasizes the benefits of polyunsaturated fats and says the only fats we should be avoiding are saturated fats and trans fats. Which is where we differ. Not on the trans fat thing, of course. Trans fats are deadly and you shouldn't eat them. Throw out your margarine and Crisco. But Dr. Willett continues to hang onto the incorrect notion that you should avoid saturated fat despite the results of his research.

Dr. Willett lays out his anti food pyramid argument in a 2003 "Eating Well" article called "Razing the Pyramid": "We actually see in the Nurses' Health Study that people who use full-fat salad dressings most days have a 40 percent lower risk of a fatal heart attack ... In some sense we've created a whole national feedlot on this low-fat, high-starch diet ... Probably millions of people have died prematurely from all the trans fats that have been included in our diet ... Unless you are extremely lean and extremely active, you can't tolerate [potatoes]." I've got no qualm with any of that, the occasional plate of butter-fried new potatoes notwithstanding.

NOTE: Eating Well conveniently removed that article from their website as I was publishing this. You can still order the PDF of that article if you want. This PBS article is just as good. I still have the same complaints with his logic.

But when asked what he thinks about the Atkins diet he makes a very interesting statement: "Atkins is partly right in that high intake of carbohydrate, particularly if it is refined, has adverse effects, and that we saw no increase in risk with butter consumption (and only a small increase in risk with saturated fat compared to carbohydrate). However, high intake of saturated fat is not part of an optimal diet because risk of cardiovascular disease and diabetes will be reduced if saturated fat (or carbohydrate) is replaced with either polyunsaturated or monounsaturated fats. Thus, Atkins has mainly compared two unhealthy forms of calories [saturated fat and refined carbohydrates]. The answer is to emphasize healthy fats."

So Dr. Willett is arguing that eating saturated fat is not really worse for you than eating carbohydrates, but you shouldn't eat it anyway, because there are choices that might give more benefit in terms of avoiding heart disease. This argument is wrong-headed on several counts. First, his research clearly shows that refined carbohydrates are far more dangerous for your heart than saturated fat. Second, you have to eat something, right? We can't replace all saturated fat and carbohydrate with polyunsaturated fat. Soybean oil shots, anyone? Most subtly, though, his argument that we should replace saturated fat with unsaturated fats is misleading. For instance, although you wouldn't know it from the mainstream press, 61% of pork fat is, by Dr. Willett's definition, "healthful". It turns out that eating "saturated fat" isn't actually a choice. Read on.

The rest of the article will explore these points. First off, an introduction to Dr. Willett's research:

"The Nurses Health Study" is Dr. Willett's study of the habits and disease patterns of over 80,000 nurses that has been ongoing since 1980. The study gives nurses questionnaires about what they eat, how much they exercise, whether they smoke, etc. The nurse's medical records are tracked and then correlations can be made between lifestyle choices and disease risks. Nurses were chosen as the study group because they are presumably reliable and have good medical records, among other reasons.

NOTE:If you're interested in learning about the statistical methods used by the Willett group, see the statistical discussion at the end of this paper.

Now onto Dr. Willett's statement. The first part is, "Atkins is partly right in that high intake of carbohydrate, particularly if it is refined, has adverse effects, and that we saw no increase in risk with butter consumption (and only a small increase in risk with saturated fat compared to carbohydrate)."

Willett is clearly justified in suggesting that a high intake of refined carbohydrates is detrimental to heart health. This is shown in his paper, "A Prospective study of dietary glycemic load, carbohydrate intake, and risk of coronary heart disease in US women."

The paper reports on the risk of eating a diet with a high "glycemic load". The glycemic load is a complicated subject in and of itself, but in a nutshell, it is the measure of your diet's total effect on blood sugar levels. It takes into account both the type and amount of carbohydrates you eat. Basically, foods that are high in carbohydrates and that cause a very fast blood sugar response have a high glycemic load. Foods that are low in carbohydrate or are more slowly absorbed have a lower glycemic load. Interestingly, a potato has a vastly higher glycemic load than an apple despite their similar amounts of total carbohydrates. This is because potato carbohydrate is largely starch, which is easily converted to blood sugar and apple carbohydrate is largely fructose, which is absorbed into the bloodstream only slowly. Conversely, the carbohydrate in carrots is easily absorbed into the blood, but their glycemic load is small because there is relatively little carbohydrate in a carrot. In a nutshell, this describes the difference between the "glycemic index" - how easily the carbohydrate of a food is converted to blood sugar, and "glycemic load" - how easily the carbohydrate of a food is converted to blood sugar times the total amount of carbohydrate in a serving of that food.

You can look at the definitive source on the glycemic load to find the glycemic load of your favorite foods. The first number after the food name is the glycemic index, the next number is the serving size and the last number is the glycemic load, which is what we're largely concerned with here. To give you a rough idea, though, a large slice of white bread has a glycemic load of 10. It's additive, so two large slices of white bread would have a glycemic load of 20. 120 grams of apple has a glycemic load of 6. A 150 gram baked russet potato has a glycemic load of 26. Cornflakes with skim milk and a banana have a glycemic load of 37 (cornflakes (21) plus banana (12) plus skim milk (4)) whereas two eggs, bacon and a slice of toast have a glycemic index of 10. Eggs and bacon, of course, don't contain carbohydrates. Well, OK, egg yolks have a little carbohydrate.

The glycemic load paper found that people who consumed a diet having an average daily glycemic load of 206 were about twice as likely (multivariate RR = 1.98) to have a heart attack as those whose diet had an average glycemic load of 117, all other things being equal. Another way of saying this is that an extra four servings of potatoes a day doubles your risk of a heart attack. People whose diets had a higher glycemic load ate more potatoes, dark bread, white bread, orange juice, cold breakfast cereal and bananas. So I think Dr. Willett's data certainly supports his point that "high intake of carbohydrate, particularly if it is refined, has adverse effects".

Willett openly admits to finding no correlation between butter consumption and heart disease, but says they found a "small increase in risk with saturated fat compared to carbohydrate". That small increase in risk was from the paper Dietary fat intake and the risk of coronary heart disease in women. This paper shows a very small and statistically unconvincing risk for "replacing carbohydrate intake with saturated fat". In 2005, Willett's group published Dietary Fat Intake and Risk of Coronary Heart Disease in Women: 20 Years of Follow-up of the Nurses Heart Study. This paper superseded the previous one and dropped all claims that saturated fat consumption is a risk factor for heart disease. The paper concludes "Polyunsaturated fat intake was inversely associated with CHD risk (multivariate relative risk (RR) for the highest vs. the lowest quintile = 0.75, 95% confidence interval (CI): 0.60, 0.92; p(trend) = 0.004), whereas trans-fat intake was associated with an elevated risk of CHD (RR = 1.33, 95% CI: 1.07, 1.66; p(trend) = 0.01)". In layman's terms, this means that eating lots of polyunsaturated fat reduced the nurses risk of heart attacks by twenty five percent whereas eating lots of trans fat increased their risk of heart attacks by thirty three percent. The paper showed that there was no correlation whatsoever between saturated fat consumption and heart attack risk (multivariate RR = 0.97). Throw out your margarine, though.

I'm going to repeat that in case you missed it. In bold this time. The Nurses Heart Study, one of the largest and longest running health studies ever done, following more than 80,000 nurses for twenty years, found no correlation whatsoever between saturated fat consumption and heart disease risk. Conversely, the same study showed that eating lots of starchy and sugary food roughly doubled heart disease risk.

Which brings us quite nicely to the second part of Willett's statement. "However, high intake of saturated fat is not part of an optimal diet because risk of cardiovascular disease and diabetes will be reduced if saturated fat (or carbohydrate) is replaced with either polyunsaturated or monounsaturated fats. Thus, Atkins has mainly compared two unhealthy forms of calories [saturated fat and refined carbohydrates]. The answer is to emphasize healthy fats."

I disagree strongly with the suggestion that saturated fat and refined carbohydrates are equally bad with regards to heart health. As we have seen from Dr. Willett's own study, a diet high in refined carbohydrates doubled the risk of heart attack whereas a high saturated fat diet showed no risk whatsoever. Enough said.

Then he argues that we should replace saturated fat with mono or polyunsaturated fat, certainly implying that we should avoid pork chops. That's where his argument really starts to break down and this is where we'll clarify the difference between "saturated fat" and "animal fat".

There are no foods made of saturated fat and there are no foods made of polyunsaturated fat. All naturally derived fats are mixtures of saturated, monounsaturated and polyunsaturated fats. Perhaps the mostly widely misunderstood fat is lard (porkfat). Lard, even from factory raised, cornfed hogs, is only 39% saturated according to the USDA. 45% of lard is monounsaturated and 11% is polyunsaturated. (You'll have to ask the USDA why the numbers don't add up to 100%.) So when you eat that big rack of fatty pork ribs, you're consuming some saturated fat along with a larger amount of "healthful" fats. This is part of the reason why a tablespoon of olive oil has about 75% more saturated fat than a pan-fried strip of bacon. Look it up. It may also be why some of Dr. Willett's papers have reported very small, largely insignificant risks for "saturated fat" consumption, but no risk at all for "animal fat" consumption. Animal fat is not saturated fat.

So with this in mind, let's think about Dr. Willett's statement that, "risk of cardiovascular disease and diabetes will be reduced if saturated fat (or carbohydrate) is replaced with either polyunsaturated or monounsaturated fats". Now visualize yourself sitting in front of a platter stacked high with perfectly grilled, fatty pork chops on one side and mashed potatoes on the other side. You have the opportunity to "substitute" protein and pork fat for starch or vice versa, depending on how much you take of each. Ignoring the protein for now, let's say you take only pork chops and only eat the fat rind around the edge. Forty percent of your calories are now from saturated fat (supposedly bad) and sixty percent are from unsaturated fat (good). But if you just take the mashed potatoes (without adding butter), you're getting ninety percent of your calories from a high glycemic load starch (definitely bad). So by eating the pork fat aren't you taking Dr. Willett's advice of replacing refined carbs with heart friendly fats?

Of course, Dr. Willett could argue that we'd be better off getting our calories from a salad topped with vegetable oil because then we'd get even more polyunsaturated fat. And what goes better with pork chops than a salad?

Further Reading

If you're interested in digging into this stuff further, here's a little primer on the statistics used in the Willett group's papers. They're actually pretty readable and I recommend it if you're into this sort of thing. Other interesting things that they've found include the fact that coffee drinkers are at significantly less risk for diabetes. Just go here and search for "willett nurses heart disease". Or "willett nurses coffee". Or whatever.

The Willett group's papers are largely concerned with the concept of "relative risk", often referred to as RR. The way relative risk is calculated is to split a group of people up based on some statistic, like how much fat they eat. Then you watch them for ten or twenty years and count how many heart attacks occur in the different groups (or whatever health problem you're interested in - it could be diabetes or gout). The group that eats the least fat is said to be the "baseline" group and is given a RR score of 1. Let's say that after ten years the people in this group have had 100 heart attacks. If the people in the group that eats the most fat have also had 100 heart attacks, their relative risk score would also be 1, and we would say that total fat consumption has no effect on heart disease risk. If the people in the highest fat group have had 200 heart attacks, their RR would be 2 and we say that people in the group that eats the most fat were 100% more likely to have a heart attack compared with those in the group that ate the least. Or if the people in the highest fat group had only 50 heart attacks, the RR score for high fat consumption is 0.5 and we would say that people in the group that ate the most fat were 50% less likely to have a heart attack compared with people who ate the least fat.

Generally, they like to split people into five groups based on their fat consumption (or whatever else is being studied). So people in the group who ate the least fat are said to be in "the lowest quintile" and the group who eats the most fat is said to be in "the highest quintile".

Another thing you need to understand before you can interpret their papers is what a "multivariate RR" is. The problem with studying people is that they're complicated and comparing the raw numbers numbers between two of our "quintiles" of people ends up being an apples to oranges comparison. The reason for this is that, for instance, the people in the highest quintile of fat consumption might also smoke more, get less exercise, drink less alcohol and be less likely to take a multivitamin, etc. (Incidentally, all of those things are true.) What we want to do is compare, for instance, an active forty year old non-smoker in the first (lowest) quintile of fat consumption to an active, forty year old non-smoker in the fifth (highest) quintile of fat consumption. This tell us the effect of the fat consumption itself. Without taking into account factors like exercise and smoking levels into the calculation, you don't know if the people in the highest quintile of fat consumption have more heart attacks because they eat more fat or because they smoke. Incidentally, the "multivariate RR" score is calculated by a statistical model on a computer. The downshot is that anytime you want to make an apples to apples comparison for any risk factor, you should look at the "multivariate RR" score. Usually, the relevant one in a Willett paper is the last one listed for any given category.

Lastly, all statistical scores have a P score, which tells you whether a given correlation is "significant" (real). A non-signifcant correlation is likely due to chance or other factors rather than the risk factor being looked at. A significant correlation should have a low P score. Generally, P scores of less that 0.05 are considered significant and scores less than 0.01 are considered very significant. Incidentally, the P score in the paper that Willett is referring to as his "small
increase in risk with saturated fat compared to carbohydrate" was 0.10, which I would say is statistically insignificant.