A response by the Linus Pauling Institute to an article published in the February 2009 issue of the Archives of Internal Medicine.
A study published recently in the medical journal Archives of Internal Medicine (Arch Intern Med. 2009;169(3):294-304), which followed 161,808 women from the Women's Health Initiative over eight years, claimed to provide "convincing evidence that multivitamin use has little or no influence on the risk of common cancers, cardiovascular disease, or total mortality in postmenopausal women." This message was immediately sent around the world by the news media, leading people everywhere to believe that taking a daily multivitamin does no good and is a waste of money. Is it, really? Actually, nothing could be further from the truth.
The study was an observational study, not a randomized controlled trial. Both types of studies are called "epidemiological" or population-based studies, but there is a fundamental difference between the two of them. As its name implies, an observational study "observes" what people do, what they eat, what dietary supplements they take, how they live, and what kind of diseases they develop. Randomized controlled trials take a group of subjects and randomly assign half of them to get a specific treatment, for example, a certain pharmaceutical drug or vitamin, and the other half gets a dummy pill, or placebo. After several years, researchers assess whether those who got the actual treatment develop less disease than those who got the placebo.
Every epidemiologist will tell you that observational studies cannot establish cause-and-effect relationships; they only can observe associations. For example, a study may find that intake of a certain vitamin was associated with a lower incidence of a specific disease. Whether that vitamin was the cause for the decreased disease risk cannot be answered by an observational study. In order to answer that question, a randomized controlled trial is necessary. In other words, every epidemiologist knows that observational studies are only good enough to generate a new hypothesis, like "multivitamins might not lower risk of heart disease", but this hypothesis needs to be tested in randomized controlled trials to either prove it, establishing a cause-and-effect relationship, or refute it. Unless and until such trials have been conducted, one cannot draw any conclusions regarding causality, let alone make recommendations for the public.
Observational studies are only hypothesis-generating because they are notoriously difficult to evaluate and interpret. For example, the data are based on information collected from the participants, which is often selective and inaccurate (called "recall bias"). Behavior can change appreciably over eight years of observation. Multivitamin formulations vary considerably, and participants may have changed brands during the study. Most importantly, people who volunteer to be part of these studies are generally healthier than the average person - they are more health conscious, have a healthier diet, exercise more, etc., which can significantly affect the outcome of the study (called "healthy enrollee effect").
In the study, 41.5% of the participating women took multivitamins, and these multivitamin users were healthier than the non-users. Multivitamin users were more likely than non-users to be Caucasian, live in the Western U.S., drink moderate amounts of alcohol, smoke less, have a lower body mass index and a higher level of education, and report being physically more active and eat more fruits and vegetables and less fat. Each of these factors can strongly influence the multivitamin users' risk of disease, which makes it very difficult, if not impossible, to tease out the role of multivitamins alone. Epidemiologists use statistical models, in this case the "proportional hazards model," that they claim allows "adjusting" their data for all of these factors, but they often do not acknowledge that these statistical models are based on many assumptions and are imperfect, and are applied to incomplete and inaccurate data.
Here is the dizzying list of factors for which adjustments were made to the data in the study, quoting directly from the paper: "age; race/ethnicity; years since menopause (<5, 5-10, 10-15, and >15 years); body mass index; education; alcohol use; smoking; general health; history of bilateral oophorectomy [or ovariectomy, the surgical removal of one or both ovaries]; geographic region; physical activity; duration of prior postmenopausal estrogen therapy use (0, <5, 5-10, 10-15, and >15 years); duration of prior postmenopausal estrogen plus progesterone use (0, <5, 5-10, 10-15, and >15 years); fruit and vegetable intake; percentage of energy from fat; single supplements of vitamin C, E, or calcium and any other single supplement use and stratified according to age (5-year groups), and hormone therapy trial randomization assignment or study enrollment."
Because all of these adjustments were made using imperfect data and an imperfect statistical model, they are very unlikely to reveal the true effect of multivitamins. Furthermore, despite the statisticians' best efforts to take all of these "confounding" factors into consideration, there are numerous additional factors left that haven't been discovered yet or were not measured in the study. This phenomenon is called "residual confounding" and is a major reason why observational studies can only generate hypotheses. In contrast, in randomized controlled trials subjects are randomly assigned to treatment or placebo, so all confounding factors, even the unknown ones, should be distributed equally between the two groups. That's why well-designed, randomized controlled trials are superior to observational studies and, in contrast to observational studies, can establish cause-and-effect relationships.
Given these considerations, it appears inappropriate for the authors of the study to conclude that it provides "convincing evidence" for multivitamins having little or no effect on cancer or cardiovascular disease risk. The evidence is far from convincing; it is suggestive at best. In addition, while endpoints like cancer, heart disease, and death are important, it is possible, for example, that a daily multivitamin helps protect against other diseases, improves immune or brain function, or promotes general health. Also, eight years of multivitamin supplementation in women over 50 years of age, as assessed in the study, may be too little too late to have a significant effect. Obviously, the data do not apply to men, because they have a different risk profile for cardiovascular diseases and hormone-dependent cancers, among many other reasons.
The reality is that most people in the U.S. have a poor diet and don't come close to consuming the recommended nine servings of fruit and vegetables every day. As a consequence, high percentages of the U.S. population do not meet the recommended dietary allowances — set by the U.S. Institute of Medicine — for many vitamins and essential minerals, including vitamins A, C, E, and K, folic acid, zinc, magnesium, and calcium. For example, data from the National Health and Nutrition Examination Survey indicate that over 90% of the population doesn't meet the recommended dietary intake for vitamin E, over 40% for vitamin A, 30% for vitamin C, and 50% for magnesium. And evidence is accumulating that most people in the U.S. are vitamin D deficient.
Given the reality that people will not improve their diet and often cannot afford to buy more fruits and vegetables, the next best thing and most cost-effective solution is to take a multivitamin. Despite the cynics' assertion that "popping vitamins is a waste of money," taking a daily multivitamin costs less than 10 cents a day. Even Dr. JoAnn Manson, a principal investigator of the Women's Health Initiative and co-author of the study, acknowledges that "the research doesn't mean multivitamins are useless. Multivitamins may still be useful as a form of [health] insurance for people with poor eating habits." And let's not fool ourselves, that's the large majority of the people in this country!