Monthly Archives: July 2013

The Case of the Killer Vitamins Revisited

A lot of people have asked about the Atlantic article by Paul Offit, “The Vitamin Myth: Why We Think We Need Supplements.” Offit, a pediatrician, is best known for his defense of childhood immunizations. In arguing against benefits from vitamins, as he did in arguing against harm from vaccines, he takes a strong stand.

Before I take a look at his article, let me mention a competing publication about supplements that became available today, and might be more worthy of your time.’s Supplement Goals Reference Guide

A friend of the blog, Sol Orwell, has been an expert on nutritional supplements for many years and has consulted extensively for vitamin and supplement manufacturers. He and his colleagues have spent years compiling the most extensive database of peer-reviewed literature available, and have compiled the data into a 762 page reference guide that has a comprehensive overview of the supplement literature. It is an encyclopedic resource with background information about almost every supplement that has been studied, including herbal remedies, each graded for quality of evidence, and easily searchable by:

  • supplement, to view the evidence for effects, good and bad, of each supplement; and
  • health goal or biomarker, to see which supplements may help you achieve your health goal (such as, “blood glucose,” “breast tenderness,” “glycemic control,” “canker sores,” “fecal moisture,” “fat oxidation,” “free testosterone,” “food intake,” “memory,” “migraine,” “pain,” “postpartum depression,” and even, I kid you not, “penile girth”).

If this sounds interesting, read more about it here.

The Atlantic Article

OK, so what evidence does Offit compile against supplements?

Central to his case is a paper that we’ve already discussed on this blog (“Around the Web; The Case of the Killer Vitamins,” October 15, 2011): the 2011 analysis of the Iowa Women’s Health Study by Jaakko Mursu and collaborators in Archives of Internal Medicine. [1] Offit highlights it in the very first sentence of his article:

On October 10, 2011, researchers from the University of Minnesota found that women who took supplemental multivitamins died at rates higher than those who didn’t.

Apart from this reference he discusses two issues:

  • Linus Pauling’s argument that vitamin C can prevent colds and cancer has not been supported. A history of Pauling’s romance with vitamin C occupies about two-thirds of the article, and Offit seems to think the whole idea of supplementation was originated by Pauling and persists due to his influence (“What few people realize, however, is that their fascination with vitamins can be traced back to one man,” Pauling.).
  • Multiple studies have shown that supplementing vitamins A and E is often harmful.

This is hardly a comprehensive case against supplementation; it only shows that a few supplements tend to be harmful or lack benefits.

Regarding vitamins A and E, we agree; they are noted in our book as nutrients that should generally not be supplemented, or should be supplemented in low doses from natural sources. For example, we recommend eating a quarter pound of liver per week for vitamin A and other nutrients.

Regarding vitamin C, that it may not prevent colds or cancer does not mean it has no benefits. A few:

Our book has more evidence for benefits of vitamin C supplementation. Many people notice improved skin, hair, nails, gums, and teeth when supplementing with vitamin C, and faster wound healing. Vitamin C needs rise dramatically in sickness and stress.

It would be easy enough to compile further evidence of benefits of supplementation. Indeed, it was not so long ago that pellagra was rampant in the US South, and beriberi in East Asia. Many foods are subject to mandatory micronutrient fortification – a form of supplementation – to prevent iodine, folate, niacin, and thiamin deficiencies. So at least one branch of the government is convinced some supplementation is desirable.

So Offit’s case basically comes back to the Mursu et al paper. [1] Let’s revisit that one.

The Iowa Women’s Health Study

The study followed a large number of women in Iowa, and queried them several times about supplement use. In 1986, the baseline, the women had an average age of 62 (range of 55 to 69) and 66% were taking supplements. By 2004, the surviving women had an average age of 82 and 85% were taking supplements.

Here is the data on overall mortality vs supplement use:

“Cases” are instances of someone dying. “HR” or hazard ratio is the likelihood of dying if you supplement divided by the likelihood of dying if you don’t. Note that all the hazard ratio estimates are “adjusted.”

We can calculate the raw data from the Users and Nonusers columns. In general, supplements had no obvious effect – certainly no statistically significant effect. The fraction of Users and Nonusers who died was essentially identical. If we eliminate copper which only had 229 supplementers, the hazard ratio of supplementers averaged 99.8% that of non-supplementers – i.e., supplementers were very slightly less likely to die.

But – and this is a key point – supplement use increased with age throughout the study. Roughly, 66% of 62 year olds took supplements and 82% of 82 year olds took supplements. But mortality at age 82 is about five times higher than mortality at age 62. So the high-mortality 82 year olds were supplementing more than the low-mortality 62 year olds, but supplementers had the same mortality as non-supplementers! This indicates that with age adjustment, supplementation would have shown a clear benefit.

Did Mursu et al offer an age-adjusted analysis? No, they did not. The next column in the table is age-and-energy-adjusted. “Energy” means calories of daily food intake. But the purpose of eating is to supply our body with nutrients, and supplementing nutrients reduces appetite and energy intake. (This is discussed in Chapter 17 of our book.) Lower energy intake is associated with better health, largely because a high proportion of the elderly are diabetic: 27% of those over age 65 or older are diabetic, and 50% are diabetic or prediabetic; diabetics and prediabetics benefit from lower energy intake. By adjusting for energy, they are removing credit from the supplements for the health improvements due to reduced energy intake.

Nevertheless, after age-and-energy adjustment, we find that supplements generally decreased mortality. Nine of the fifteen supplements decreased mortality, five increased mortality. At the 95% confidence interval, five supplements decreased mortality, only one increased mortality.

Making the Elephant Wiggle His Trunk

The mathematician John von Neumann gave us the insight we need to understand this paper’s analysis:

With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.

Mursu et al used multivariable adjustments with 11 parameters and 16 parameters respectively to obtain their “results.” Using so many parameters lets the investigators generate whatever results they want.

I don’t think it’s a coincidence that both multivariable adjustments substantially increased the hazard ratio of every single one of the 15 supplements. The 11-variable adjustment increased hazard ratios by an average of 7%, the 16-variable adjustment by an average of 8.2%.

Rest assured, it would have been easy enough to find multivariable adjustments that would have decreased hazard ratios for every single one of the 15 supplements by 7 or 8 percent.

I may as well quote my earlier analysis:

I believe it verged on the unethical that the variables chosen include dangerous health conditions: diabetes, high blood pressure, and obesity. These three health conditions just happen to be conditions that are improved by supplementation.

Anyone familiar with how regression analyses work will immediately recognize the problem. The adjustment variables serve as competing explanations for changes in mortality. If supplementation decreases diabetes, high blood pressure, and obesity, and through these changes decreases mortality, the supplements will not get credit for the mortality reduction; rather the decreased diabetes, blood pressure, and obesity will get the credit.

It’s appropriate to ask: if it’s proper to include health conditions like diabetes as variables in the regression, why not include other health conditions like cancer? The likely answer: Supplementation does not generally help conditions such as cancer, so including such conditions as adjustment factors would not have made the supplements seem more harmful. Rather, by giving greater weight to diabetes and obesity – conditions supplementation benefits – it would have made supplements look more beneficial.

It is impossible to take seriously studies that provide 11- and 16-variable adjustments, with arbitrarily chosen adjustment factors and no sensitivity analysis showing how alternative choices of adjustment factors would have altered the results.


The great economist Ronald Coase (in his essay “How should economists choose?”) said, “If you torture the data enough, nature will always confess.”

The Mursu paper was an exercise in torturing data until it declared, “Supplements are harmful!”

The Offit piece is a polemical exercise pretending that an unsettled part of biology – our nutrient needs, and the circumstances in which food fails to meet them – is a settled subject with a simple answer.

Now, it’s quite difficult to establish the healthfulness of supplementation in general, because you can always get too much of a nutrient, nutrient needs differ among persons depending on their health and age, and whether a person will benefit from a nutrient depends on whether the rest of the diet is deficient in that nutrient. So any given supplement is going to be harmful in some circumstances, beneficial in others.

A proper scientific approach would be to try to determine the circumstances under which supplements (or dietary changes eliminating the deficiency) would be beneficial.

Offit’s piece doesn’t attempt that. Our book does, and would make a much better resource to those considering supplementation. So would the Examine supplement goals reference guide.


[1] Mursu J et al. Dietary supplements and mortality rate in older women: the Iowa Women’s Health Study. Arch Intern Med. 2011 Oct 10;171(18):1625-33.

Live to 110 Podcast

A new podcast is up: the Live to 110 Show on Blogtalkradio with Wendy Myers. Wendy is a health and nutrition coach who was motivated to become a healer after her father died of cancer; her research eventually led her to the Paleo movement. If you like podcasts, check out our conversation!

Latest in Paleo Episode 75

A new podcast is out: Angelo Coppola and I talk about traditional versus modern diets, weekly versus daily variability in nutrition, the importance of circadian rhythms, the new Perfect Health Retreat in Austin, TX, the real-world results of people applying the Perfect Health Diet approach, and more.

Check it out at Latest in Paleo Episode 75: Perfect Health Diet with Paul Jaminet.

Omega-3 Fats and Cancer

On Wednesday a new paper reported that higher levels of long-chain omega-3 fats (EPA, DPA, and DHA) in blood are associated with a 43% increased risk of prostate cancer and a 71% increased risk of aggressive prostate cancer. [1] This built on earlier work by the same group. [2] In a press release, the authors stated:

“We’ve shown once again that use of nutritional supplements may be harmful,” said Alan Kristal, Dr.P.H., the paper’s senior author …

“[W]e have confirmed that marine omega-3 fatty acids play a role in prostate cancer occurrence,” said corresponding author Theodore Brasky, Ph.D.

They sound confident! Is there anything to it, and should it affect our dietary advice?

Mechanisms Linking Omega-3 Fats to Cancer

In our book and on this blog, we’ve already discussed two mechanisms linking excessive omega-3 intake to cancer risk.

First, there is the issue of lipid peroxidation. Of all fatty acids, long-chain omega-3 fats are the most readily peroxidized:

PUFA relative peroxidizability

Peroxidation of PUFA generates highly toxic compounds, such as aldehydes, which mutate DNA and turn proteins into advanced lipoxidation end products (ALEs). [3, 4] These lipid peroxidation products have been implicated as causal factors in cancer. [5]

Second, oxidation products of DHA promote angiogenesis – the creation of new blood vessels to feed tumors. These products make cancers grow rapidly. I’ve blogged about this (DHA and Angiogenesis: The Bottom Line, May 4, 2011; Omega-3s, Angiogenesis and Cancer: Part II, April 29, 2011; Omega-3 Fats, Angiogenesis, and Cancer: Part I, April 26, 2011).

So there are known mechanisms by which the long omega-3s in fish oil may promote cancer.

The Brasky et al Papers

The new study by Brasky et al measured omega-3 fat levels in plasma phospholipids. Thus, it doesn’t measure any omega-3s in cells, only omega-3s in serum particles like LDL, HDL, and VLDL; and even in those particles it excludes omega-3 fats found in triglycerides.

This is a very different biomarker than the Omega-3 Index of William Harris, which looks at the omega-3 phospholipids in red blood cell membranes. [6] This biomarker might behave quite differently than the Omega-3 Index.

The study measured plasma phospholipid omega-3s in a group of people, then followed them for 6 years or so to see who developed cancer. Here are the group averages [1]:

Brasky 2013 Table 2

Statistically the most reliable data is in the no cancer vs total cancer comparison. There we find that subjects who went on to develop prostate cancer averaged 3% more DHA and 4% more EPA+DPA+DHA in plasma phospholipids than those who didn’t develop cancer.

Does This Variation Reflect Dietary Intake?

Chris Kresser kindly sent a link to an analysis of the study published at LecturePad by William Harris: “Omega-3 Fatty Acids and Risk for Prostate Cancer.” Harris tells us how to translate the plasma numbers to the corresponding Omega-3 Index numbers:

Based on experiments in our lab, the lowest quartile would correspond to an HS-Omega-3 Index of <3.16% and the highest to an Index of >4.77%).

Even the top quartile of the Brasky et al subjects had quite low omega-3 levels:

In Framingham, the mean Omega-3 Index of participants who were not taking fish oil supplements was 5.2% and for those taking supplements, it was 7.5% [7]. Both of these numbers are considerably higher than the values reported by Braskey et al., even in their highest quartile.

The trial asked its participants not to take supplements, and it looks like they drew a study population whose fish intake was much lower than that of Framingham, Massachusetts, residents.

If dietary omega-3 intake was low in all subjects and varied only slightly among participants, how do we know that this biomarker is related in any way to dietary intake? There could be other factors – genetics, oxidative environment, omega-6 fat intake, antioxidant intake, changes in the proportions of VLDL, LDL, and HDL, to name a few – that affect this biomarker.

Does High Dietary Intake Lead to More Cancer?

If high dietary intake of omega-3s caused more cancer, we would expect cultures that consume lots of fish oil to have higher prostate cancer rates. But epidemiological studies have found that high omega-3 intakes seem to be associated with low cancer rates. For instance, the Japanese eat eight times more omega-3 fatty acids than Americans and their blood levels are twice as high, yet the prostate cancer rates are only one-sixth the American rate.

Of course, there are many confounders in epidemiological studies. Harris helpfully provides a summary of clinical trials in which fish oil was provided as part of the study and cancer outcomes measured:

Harris reply to Brasky 2013

Although none of these studies produced a statistically significant link between omega-3 intake and cancer, incidence of cancer diagnosis or death was increased in every one of the clinical trials except the GISSI-Heart Failure study and perhaps the Origin study. A meta-analysis might find a small cancer promoting effect of omega-3s.

UPDATE: Vladimir Heiskanen points me to an interesting paper in which the effect of dietary fatty acids on cancer metastasis was examined. Colon carcinoma cells were injected into the portal vein (which leads from intestine to liver) of rats and 3 weeks later rats were sacrificed and their livers were examined for metastases. The rats were on three diets — low-fat, high omega-6 (safflower oil), high omega-3 (fish oil). The results:

At 3 weeks after tumor transplantation, the fish oil diet and the safflower oil diet had induced, respectively, 10- and 4-fold more metastases (number) and over 1000- and 500-fold more metastases (size) than were found in the livers of rats on the low-fat diet. [7]

I wish they’d used a saturated fat or monounsaturated fat diet, rather than a low-fat diet, as the control, as this would have clarified that polyunsaturates specifically promote metastasis; in the study the rats’ food was mixed with fish oil or safflower oil, greatly increasing the fat fraction and decreasing the carbohydrate, protein, and micronutrient fractions, so the control diet deviates in many respects from the high-PUFA diets. However, the results are consistent with the idea that fish oil is more cancer-promoting than the less peroxidizable safflower oil, perhaps because of the unique pro-angiogenic effects of DHA products.


There might be biological contexts in which omega-3 fats promote cancer.

This doesn’t mean we should refrain from eating omega-3 fats. Cardiovascular disease causes more deaths than cancer, and omega-3 fats are protective against CVD.

However, I think these studies support the PHD advice:

  • Eat enough oily marine fish to achieve omega-6 and omega-3 balance;
  • Minimize omega-6 intake so that omega-6 and omega-3 balance is achieved at the lowest possible intake of polyunsaturated fats.

All nutrients can be eaten in excess, and omega-3 fats surely fall into this category. The right amount of oily fish is probably about one to two meals per week.


[1] Brasky TM et al. Plasma Phospholipid Fatty Acids and Prostate Cancer Risk in the SELECT Trial. J Natl Cancer Inst. 2013 Jul 10. [Epub ahead of print]

[2] Brasky TM et al. Serum phospholipid fatty acids and prostate cancer risk: results from the prostate cancer prevention trial. Am J Epidemiol. 2011 Jun 15;173(12):1429-39.

[3] Hulbert AJ et al. Life and death: metabolic rate, membrane composition, and life span of animals. Physiological Reviews 2007 Oct;87(4):1175–213,

[4] Hulbert AJ. Metabolism and longevity: is there a role for membrane fatty acids? Integretive and Comparative Biology 2010 Nov;50(5):808–17,

[5] Nair U, Bartsch H, Nair J. Lipid peroxidation-induced DNA damage in cancer-prone inflammatory diseases: a review of published adduct types and levels in humans. Free Radic Biol Med. 2007 Oct 15;43(8):1109-20.

[6] Harris WS, Von Schacky C. The Omega-3 Index: a new risk factor for death from coronary heart disease? Prev Med. 2004 Jul;39(1):212-20.

[7] Griffini P et al. Dietary omega-3 polyunsaturated fatty acids promote colon carcinoma metastasis in rat liver. Cancer Res. 1998 Aug 1;58(15):3312-9.