The notion that high cholesterol (more specifically, high “bad” — LDL — cholesterol) causes heart disease may be as widely accepted as the notion that humans have caused dangerous global warming. It is much easier to test, however. An excellent study published in 2006 compared two groups of people at risk for heart disease: those given a high dose of statins and those given a low dose. The high dose reducd LDL cholesterol levels; as it was meant to; the low dose did not. But there was no effect on coronary heart disease progression. After a year of statins, persons in both groups had increased their coronary artery calcification score by the same amount — about 25%. Totally contradicting the cholesterol hypothesis.
Regular readers of this blog may remember that after a year of eating butter (half a stick per day), my coronary artery calcification score decreased 24%. Because increases of about 25% are the norm, my score was about 50% less than expected. Decreases are very rare, I was told.
Thanks to Hyperlipid. Statin side effects.
Saying that the study “compared two groups of people at risk for heart disease” leaves out a crucial fact. The study compared two groups of people *who responded well to low doses of Lipitor*. That is, patients with high bad cholesterol (<160) were recruited and treated with 10mg Lipitor. If their bad cholesterol fell below 130, they were kept in the study, and only then were they split into the low-dose (10mg) and high-dose (80mg) groups. About half the patients were excluded from the study due to their inadequate response to low-dose Lipitor.
So we can conclude that *among people who responded well to low-dose Lipitor*, high-dose Lipitor provides no additional benefits (at least according to the measure used in this study). To me, it seems foolish to draw more sweeping conclusions than that from this one study.
And of course, none of this changes the fact that other studies have found that Lipitor and other statins reduce the incidence of stroke and heart attack.
I find that the conclusions from these studies are always highly qualified, for example:
“Using high doses of cholesterol-lowering drugs called statins appears to reduce the risk of heart attack, stroke or the need for additional cardiac procedures more than regular doses of statins in people who have had a stroke or suffer from heart disease, two new studies find.”
This applies only to certain high-risk individuals, which doesn’t include Seth, me, or most of the population. So while this is interesting to a small subset of the population, it isn’t accurate to say that “Lipitor and other statins reduce the incidence of stroke and heart attack” without lots of qualification, at least say “in certain people under certain circumstances.”
I agree with what Seth and many “statin skeptics” state or imply, which is that nobody should rush to put themselves on an extremely strong drug for the rest of their lives if they are healthy to begin with. This is just common sense. Where are the studies that compare a sedentary lifestyle and high-carbohydrate diet to the opposite, and then measure the health effects? I don’t think we’ll see any, because there is no money in it.
The best part is the last sentence of the abstract for this study: “The possibility remains that the time window was too short to demonstrate an effect.”
Yes, more testing is definitely needed. Never mind the 6 decades of repeated failures to confirm the Lipid Hypothesis. Never mind that the results of the Framingham study in 1960 should have been enough to falsify the theory. Let’s just keep spending more money and misapply more research resources and beat this dead horse some more.
The benefits of statins — in studies done sparing no expense to maximize those benefits — are small. Here is a reference about this:
https://pharmamkting.blogspot.com/2008/01/statin-lottery-number-needed-to-treat.html
which doesn’t even factor in the cost of the bad effects of statins. It is interesting, to say the least, that one of the most prescribed drugs in the world is barely (if at all) beneficial. The study I describe here suggests why: It is based on wrong ideas.
Statins are “barely (if at all) beneficial?” I think you mean that randomized clinical trials haven’t demonstrated large long-term benefits. For someone who’s so skeptical of clinical trials, it’s odd that you’re implicitly demanding a 20-year randomized trial when the trials already conducted suggest major benefits.
That is, according to the business week table at your link, statins are awesomely effective in preventing a second heart attack (in people who’ve already had a first). Just that, plus the fact that statins seem to work by preventing plaque build-up in the arteries suggests that long-term use by those at moderate risk is likely to be highly beneficial. Add that to the fact that such use by those at moderate risk has some benefits even in the short term (reducing risk of a cardiac event from 1.5% per 3 years to 1.0%) seems like a very strong argument for long-term use to me.
Against that very strong evidence, you cite (unspecified) “wrong ideas,” (implicitly) cite the lack of a very long term clinical trial, and (reasonably) point to side effects. I think I’ll keep taking statins, thank you very much, and stop if side effects develop.
I meant my critical remarks to apply to the use of statins for primary prevention — that is, given to people who haven’t yet had a heart attack. That’s how most statins are used. Here is a paper about the overselling of statins:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1079612/
for primary prevention. Here is a quote from that article:
>>> …statins are awesomely effective in preventing a second heart attack (in people who’ve already had a first). Just that, plus the fact that statins seem to work by preventing plaque build-up in the arteries suggests that long-term use by those at moderate risk is likely to be highly beneficial. <<<
I think it's the leap from your first sentence to your second one that many find objectionable. There are studies that show benefit for certain high-risk individuals. But none that show benefit for healthy people, and many that show harm. Assuming that something that makes sense for high-risk people also makes sense for low-risk people just doesn't logically follow.
Some studies have found that moderate smoking can help prevent Alzheimer's in certain groups of people. Should we extrapolate from this that the general population should begin to smoke moderately in order to prevent Alzheimer's? Obviously not.
in seven years’ time there is a one chance in about 120 that your death will have been prevented
Yes, if you only examine *direct evidence from randomized clinical trials*, you can only show that statins have this (pretty important) benefit for those with some cardiovascular risk factors. But why do we have to limit ourselves to this tiny slice of the evidence?
In fact, there is every reason to believe that statins are even more beneficial when taken for decades. Many heart attacks and strokes are caused when blood clots lodge in arteries that have been narrowed by plaque buildup. Many studies have shown that statins reduce this plaque buildup (for example). Applying the slightest bit of deduction to these facts, we can conclude that statins, taken over many decades by those moderately at risk, are likely to reduce the incidence of heart attack and strokes.
Thanks for the reference. You write: “many studies have shown that statins reduce this plaque buildup”. If it’s so clear and repeatable, why did the study I discuss fail to find a difference in plaque build up between a high dose and a low dose? As far as I can tell, the authors expected a difference — agreeing with you — but they didn’t find one.
Some studies have found that moderate smoking can help prevent Alzheimer’s in certain groups of people.
We know little about the causes of Alzheimer’s and I doubt we have much idea about how smoking might prevent it, so of course we shouldn’t be convinced by one study.
But we know tons about the causes of heart attacks (e.g., blood clots lodging in plaque-narrowed arteries), and we know tons about how statins suppress one cause of heart attacks (statins suppress plaque build-up). That really ought to be enough to convince us of the benefits: studies that show that statins actually do prevent heart attacks are just icing on the cake.
why did the study I discuss fail to find a difference in plaque build up between a high dose and a low dose?
I could explain why I think their power analysis was incorrect, and hence the sample size was much too small to detect the expected effects. But I doubt you mean this as a serious question, so I’ll just give a general answer. The main reason is that they were testing high-dose vs. low-dose statins, rather than statins vs. a placebo, so one can expect smaller effects that are harder to detect with a small sample. Of course, they designed the study that way because the benefits of statins have become so widely accepted in the medical community that it would now be unethical to test statins vs. placebo.
I did mean it as a serious question. What’s the answer?
“The sample size was much too small to detect the expected effects.” The sample size was about 500 people. You seem to be saying that the effect was so small that an experiment — not a survey, an experiment — of thousands of subjects would have been necessary to detect it. Experimental psychologists routinely find reliable effects with n = 8. The beneficial effect of lime juice on scurvy was noticed with n = about 10. The idea that such an effect — one that requires a year-long experiment with thousands of subjects to detect — is large is new to me.
“The main reason is they were testing high-dose versus low-dose statins”. The amount of progression they measured was what you’d expect from people taking no statins. Suggesting that neither the high nor the low dose had an effect.
The amount of progression they measured was what you’d expect from people taking no statins.
I think you’re wrong to suggest that a 25% progression per year is some kind of universal law. For example, in the study the researchers cited in their power analysis (Callister et al), untreated patients experienced a 50% progression per year.
Anyway, here’s the mistake the authors make in their power analysis (used to determine the needed sample size). The authors write: “With regard to the sample size, a reduction of LDL cholesterol levels by 20% was expected in the 80-mg atorvastatin group compared with the 10-mg atorvastatin group. On the basis of results presented by Callister et al16 with a regression coefficient of 0.6, a difference of 10% between treatment groups was expected for progression of the CAC volume score.” Unfortunately for the authors, if you look at the regression in Callister et al, it’s a regression of the change in plaque on the *level* of LDL cholesterol. But what the authors need for their power analysis is a regression on the *change* in LDL, a very different thing, so their power analysis is wrong and they used too small a sample size.
I take statins, various blood pressure medications, and now a drug for diabetes, plus baby aspirin, plus an antidepressant. When I asked my doctor, who I respect, whether we really know that these are helping prolong my life, he was honest. He said we don’t know, but that something is helping at-risk people like myself to live longer with fewer heart attacks and strokes; it could be reducing stress or better aerobic exercise, or people changing their diets, more than the medications. There is simply no way to disentangle what is causing the positive effects, but no one is denying that people at risk who take these drugs are living longer, with less morbidity… Correlational data is dubious and you can’t do studies where you deprive people of exercise and rest and take it from there. And longitudinal follow-up studies have serious problems as well…
So I take them… My intuition is that getting plenty of rest and plenty of exercise are probably the 2 most important factors, but it’s only an intuition, probably biased by the fact that I get plenty of both…
The beneficial effect of lime juice on scurvy was noticed with n = about 10.
Presumably — unlike the study you cited — they compared lime juice to placebo! Imagine if ethical considerations forced them to compare 1 oz. to 2 ozs. lime juice, or lime juice to lemon juice!
A graph that shows plaque change per year as a function of LDL concentration certainly allows one to estimate how much plaque change per year will go down (or up) if LDL concentration is altered. So I am afraid I don’t follow your criticism of the power analysis.
So I am afraid I don’t follow your criticism of the power analysis.
It’s wise of you to focus on this distinctly secondary question. I’ll answer it, but first, let me ask you about the main question: how you can possibly rely on this study to back up your original claim about statins?
To recap: numerous studies of statins against placebo have shown that statins reduce plaque build-up and reduce the incidence of hear attacks and stroke. Now scientists can no longer ethically or legally conduct such studies, since it would be immoral to withhold drugs the medical community has determined are beneficial, so scientists have moved on to secondary questions. In the study you cite, the secondary question is whether high-dose statins have additional benefits in patients already responding well to low-dose statins. Even if the answer to this secondary question is no, how can you possibly conclude that statins have no benefit?
On the question of what precisely is wrong with the power analysis of the particular study you cite: surely you don’t deny that using change in LDL would be better for the analysis than the level of LDL, since changing LDL is exactly what the authors plan to do in their experiment.
More generally, I see no special reason to think that these two regressions will have the same coefficient b:
(1) d Plaque = a + b LDL and
(2) d Plaque = a + b d LDL,
where d is the 1-year change.
In fact, there’s every reason to think that the b coefficients will be different. I think that LDL in regression 1 is a proxy for the previous lifetime path of LDL, which is what really mattters. By contrast, in regression 2, past values of LDL have been differenced out. In short, what you’re really missing is that atherosclerosis (plaque buildup) is a cumulative process that occurs over many years.
A graph that shows plaque change per year as a function of LDL concentration certainly allows one to estimate how much plaque change per year will go down (or up) if LDL concentration is altered.
It occurs to me that a simpler answer is: no it doesn’t. A graph that shows the level of plaque as a function of the level of LDL might arguably allow one to estimate the change in plaque as function of the change in LDL. However, the graph you mention (changes vs. levels) simply doesn’t.
How can I say statins have no benefit? I don’t. I say the benefit is surprisingly small, given how much money is spent on them.
Re the power analysis. Your two regressions are:
(1) d Plaque = a + b LDL and
(2) d Plaque = a + b d LDL,
where d is the 1-year change.
With regression 1 you would look at the graph and compute a difference: d Plaque at one level of LDL minus d Plaque at another level of LDL. That would give you the difference to expect between groups in your experiment. The scatterplot behind the second regression is mysterious (how could it be made? what does it mean?) and could not tell you what difference in d Plaque to expect between groups. One way to make a plot with d LDL on the x axis would be to compute differences between groups. But then the y axis would be d d Plaque, not d Plaque.
[Regression 1] would give you the difference to expect between groups in your experiment.
Regression 1 gives you the difference to expect between people with initially low LDL and initially higher LDL, but these aren’t analogous to the treatment and control groups in the experiment. In fact, because the experiment is a randomized trial, both groups will have the _same_ initial level of LDL, on average. So regression 1, used by the authors of the paper you cited, tells you nothing relevant to the power analysis.
More properly, the treatment group consists of people whose LDL will go down more during the course of the experiment compared to the control group. So you make regression 2 by looking at an earlier experiment. For each person in the earlier experiment, you know (a) how much their plaque changed over the course of the experiment and (b) how much their cholesterol changed. You plot a against b. What’s so hard about that? (aside: you might want to use only the treatment group in your plot). So Regression 2 would have been the relevant calculation to perform for the power analysis.
I say the benefit is surprisingly small, given how much money is spent on them.
Then you should look at experiments that examine the benefits of statins (vs. placebo), not whether more statins are better than less statins!
You bring up a good point. I am indeed assuming that the statin changed LDL quickly — in a week, say. If the statin changed LDL more slowly — say, took 1 year to have its full effect — then your comments make sense. I think the statins change LDL quickly — so a person who starts taking a statin January 1 will have much lower LDL (if the statin dose is large enough) for almost the whole year. But I did not make that assumption clear. I think it is correct, however.
Now I understand your Regression 2, thanks. It makes an assumption of linearity that the power analysis done by the authors does not. (For example, it assumes that a change from 80 to 40 will produce the same effect as a change from 60 to 20.) Their power analysis however assumes fast-acting effects of statins, as I said. I don’t agree with you that your assumption (linearity) is plainly more plausible than theirs (fast-acting effects), but at least I now understand what you are saying.
I prefer dosage-variation experiments to placebo-vs.-treatment experiments because they are far less confounded.
I’m afraid that’s the last comment I am going to make about this. Thank you, Ragout, for your comments, which have helped me understand this better.
Prof. Roberts,
Your last comment does not accurately describe my position (I think statins are fast-acting *on LDL* and I never said or implied otherwise) nor do I think it accurately describe the study authors’ (entirely linear) calculations but I certainly agree that I’ve learned something from the discussion and that we’ve beaten this topic to death.
A friend of mine is an ER nurse with over 10 years experience. When I introduced her to your blog and ideas, this was one of the first posts she read. She was appalled. She pointed out that the machines measuring your cholesterol were calibrated a year apart by 2 different techs and wondered if one of the readings might be an outlier.
Bryan, I’m afraid I don’t understand your friend’s comments. Why was she “appalled”? I would have thought someone in health care would be happy that I managed to reduce my heart disease risk.
By “the machines measuring your cholesterol” I assume you mean the machine measuring my heart disease progression. My two heart disease progression measurements were done on the same machine. I strongly doubt your friend knows details of the machine’s calibration, such as who calibrated it (“two different techs”), when it was calibrated (“a year apart”), and the accuracy of the calibration.
Neither of my measurements is unusual, that is, neither is an “outlier”. It is when they are compared that the results are unusual.
My fault for not being more specific: she was appalled that you were eating so much butter b/c it totally contradicts everything she knows about nutrition.
Let me just offer some appreciation to Ragout for his critical analysis, insights, and responses here. His comments suggest how difficult some of these phenomena are, and how important it is to be wary of cut and dry inferences from complex studies.