Search This Site!
Nutrition Past and Future

24 Cholesterol Confusion 7 The Measurement Problem 

Cholesterol research, especially the older cholesterol research that Gary Taubes selectively used in his book, has produced seemingly inconsistent findings. This is a consequence of its inherent complexity. Some very important considerations for this type of research were not understood early on. Even today, key nuances are often missed or ignored, and this has led to unnecessary confusion about cholesterol. With this short video, I’ll lay out for you a few of the typical sources of measurement error which have been some of the wildcards of cholesterol research, accounting for much confusion. The first problem I’ll show you is the most important one, in my opinion. Some of the other ideas I’ll talk about in this video derive from this one. I’m talking about intra-individual variation. This is a big term for a simple concept. A person does not maintain perfectly steady cholesterol levels all the time. There are many factors which can cause your cholesterol levels to vary. Altogether, they can cause variations of up to 20 points on your cholesterol score.

Ancel Keys knew about intra-individual variation and this is why he stated that the proper evaluation of the response to a dietary change requires the taking of multiple measurements. He knew that a single measurement of a person’s cholesterol might not be representative because that single measurement might catch that person’s cholesterol at a high-point or a low-point.

The next quirk of data collection which has muddled some cholesterol research is a statistical artifact called regression dilution bias. This is another intimidating term for a concept that isn’t too hard to understand. Here you see what I believe is the first paper to describe this phenomenon, but on this occasion these authors were discussing it in reference to blood pressure. This passage is basically about how intra-individual variation can distort the data gleaned from a group of people. The idea here is that because some parameter that can be measured in a person, like blood pressure or cholesterol, is unstable, that means that some people will be recorded at a given moment as having a lower value than they really have on a normal basis. Just the same, some people will be recorded as having an especially high value even though that isn’t their normal value. When some people wind up at the low end in your data who really aren’t all that low normally at the same time that you have other people recorded at the high end even though they aren’t usually that high, that means whatever effect you are looking for, like say the effect of high blood pressure on heart disease, will not stand out as much. Your data has the scores from people who don’t normally have high blood pressure mixed in with the scores from people who really do normally have high blood pressure. That mixing up of numbers will make it less obvious in your data that high blood pressure is harmful. That’s why it’s called “dilution.” The real relationship you are trying to see just got diluted, so it will be harder for you to figure out what is really happening. Does that make sense? If not, don’t worry. Just read this slide a few times and you’ll get it.

Regression dilution bias is now understood to be an often overlooked factor which corrupted a lot of old research. When you see a small association in a study looking at saturated fat or cholesterol, it may well be that the association would have been larger had the researchers taken multiple measurements to minimize the effect of this problem. This review of this issue found that regression dilution bias can introduce huge distortions of observed effects. You can reasonably assume that most if not all of the oddities in my Anomaly Hunter videos were affected by regression dilution bias to some degree.

Regression dilution confused the results from Framingham, for example, especially for the older people they studied. Do you think the cholesterol confusionists who talk about Framingham are even aware of this problem?

Today, legitimate researchers know and acknowledge this. Fat-promoting bloggers and authors don’t.

Another important issue affecting cholesterol research is something you surely understand already. We are all different. Not everyone reacts the same way to the same foods. When we look at examples of genetic mutations that raise or lower cholesterol, those mutations are powerful examples of this fact. For example, the people I showed you who had those amazingly low LDL numbers in my Cholesterol Is Necessary for Life video were probably consuming high-fat diets in most cases. For them, genes were a more important factor than diet. Here is this issue visualized. This chart shows you cholesterol measurements taken from a group of students who had all been eating essentially the same diet. As you can see, they did not all have the same cholesterol scores. That’s because while their diets were the same, their genes were not. There was a distribution of cholesterol values among them resembling the bell curve, more or less. This is the pattern we would expect to see in any similar experiment with a mixed population.

This same effect was at work in the Framingham study and in other studies of populations where everyone was eating similar foods. The old research in Framingham was conducted using a population among whom 93% were eating high-cholesterol diets. Use some common sense, you silly confusionists. There wouldn’t have been many low-fat, whole food vegans in that bunch. Against this backdrop, both genetic variations and individual health problems would have wound up being the dominant generators of their participants’ variations in cholesterol. Diet couldn’t have been the explanation because diet didn’t vary all that much.

For all these reasons, cross sectional studies that examine homogeneous populations are not of great value. The problems with that type of research have been demonstrated in a detailed mathematical fashion in this paper. This is why the small-scale cross-sectional studies that Gary Taubes dumped into his book are of very little value. They were likely to have systematically underestimated the risks associated with dietary saturated fat and cholesterol. This is yet another irony we see with Taubes. The man who is always condescending to others about scientific rigor is the most invested in poorly controlled and interpreted studies. It’s as if his capacity for critical thought is switched off when he sees a result that superficially appears to contradict diet-heart.

Taubes has no excuse for this. He knows very well that cholesterol levels will vary in the same person. This provides me with yet another opportunity to show you how much difficulty he has with logical consistency.

Here is Taubes refusing to submit to a cholesterol test in public. Even though he doesn’t use the phrase, intra-individual variation is his excuse.

MEHMET OZ: You’ve been on this diet for how many years?

TAUBES: Oh, ten years.

OZ: And you look good. You look good. What’s your cholesterol?

TAUBES: Mmmm, I don’t know.

OZ: So, I understand, I acknowledge this. May I say ... we asked Gary if he’d check his cholesterol, and you didn’t want to. And I’m not trying to set you up. I would never do that on the show.

TAUBES: Well, here’s what we said, is they said, there are certain tests you can do: total cholesterol, HDL, LDL the bad cholesterol, and triglycerides. I’ve read the literature. Total cholesterol is meaningless. Absolutely meaningless. And actually LDL cholesterol is meaningless.

OZ: OK, would you allow us to check your cholesterol any way you want to check it.

TAUBES: What I’ve said is for the past three months, because my book’s come out and I have two young children in my mid-fifties, I’ve been getting about five hours of sleep, um, and I don’t want a single measurement done to reflect – I have no idea what my num.. The problem with finding out the stuff I’ve learned is you don’t want the normal tests to be done ...

That was pathetic to watch, wasn’t it?

Taubes is aware of old research like this, which showed that stressed-out medical students had elevated cholesterol. It looks like psychological stress causes an unhealthy effect not unlike what results from saturated fat consumption – both will raise your cholesterol. I am sympathetic to Taubes. I’d be very stressed as well if my career was entirely based on spreading ignorance and misinformation.

While I am discussing the general topic of problems with the measurement of cholesterol, permit me an aside. Very often cholesterol screening is done using finger pricks and portable machines. Unfortunately, these machines do not always produce scores that are as accurate as when a sample is taken with a blood draw from the arm and sent to a proper pathology lab.

This study was one of several to assert that a particular finger prick cholesterol screening device underestimated cholesterol, especially at the low end of the range. It is worse for a test to produce scores that are deceptively low. People may walk away from such a screening believing that they don’t need any follow-up with their doctors and that they don’t need to watch what they eat.

There are portable finger prick measurement devices in use now that are fairly accurate, but generally a blood draw from your arm at a pathology lab will give you the most reliable result.

For someone to get a good reading from just a finger prick, he or she needs to have good equipment and proper training.

One of the more recent sources of confusion is the issue of LDL subclass. All your LDL particles are not the same size. This is yet another added layer of complexity in the cholesterol story that is exploited by the confusionists in their efforts to take advantage of you and lead you to harm. I’ll explain why we shouldn’t exaggerate the importance of LDL subclass in the next video.