"There is a tendency for the least substantiated findings to be the ones that come out in the popular press," says epidemiologist Walter C. Willett of the Hardvard School of Public Health. "A lot of suspected associations are not ready for the public to take action or even worry about."

   The science of risk-factor epidemiology has become one of the most newsworthy disciplines in science. Each week seems to bring media coverage of a new study and its findings on factors of environment, lifestyle, and diet that affect our health either positively or negatively. While many of these studies are considered preliminary findings to be taken with a healthy dose of skepticism, others tend to carry more authority, thanks to their longevity and sturdiness of design. For example, the Harvard Medical School and School of Public Health have three huge ongoing studies—the Nurses' Health Study, the Health Professionals Follow-up Study, and the Nurses' Health Study 2—which together encompass 300,000 participants and which have provided some of the most reliable and definitive findings on health and diet. The principal investigator for the latter two studies is Walter C. Willett of the Harvard School of Public Health, who is now considered one of the premiere epidemiologists in the world.

   Science Watch's recent Top Ten lists in medicine have provided ample evidence of the impact of the work by Willett and his colleagues. One example is a hot paper from the Health Professionals Follow-up Study,"Vitamin E consumption and the risk of coronary heart disease in men," (E.B. Rimm, et al., New Engl. J. Med., 328[20]:1450-6, 20 May 1993), which first appeared last October at #5 on the Hot Papers chart in medicine and quickly rose to the #1 spot. In fact, in the previous issue of Science Watch (6[5]:5; 8, May 1995), Willett and colleagues posted the #1 and #2 papers in the medicine Top Ten: the above paper on vitamin E and heart disease in men, and a companion paper investigating the same phenomenon in women (M.J. Stampfer, et al., New Engl. J. Med.,328[20]:1444-9, 20 May 1993). In the Top Ten list currently being prepared for the next issue, these two will be joined by another Willett paper on trans fatty acids and coronary heart disease in women (see The Lancet, 341[8845]:581-5, 6 March 1993). Thus, in Science Watch's most recent tally, Willett and colleagues account for three of the Top Ten papers in medicine.

   Willett, 49, studied food science at Michigan State University from 1963 to 1966. He went on to study medicine at Harvard Medical School, receiving his M.D. degree in 1970. At the Harvard School of Public Health, he obtained his Master's and doctoral degrees in public health, the latter in 1980. Since 1987 he has been professor of epidemiology and nutrition at the School of Public Health. In 1992 he became a professor of medicine at Harvard Medical School. From his office at Harvard, Willett spoke to Science Watch correspondent Gary Taubes.

            What is the history of these three prospective studies, and what do you hope to learn from them?

   Willett: The first Nurses' Health Study began in 1976—led by Frank Speizer—as a study to look at the long-term consequences of oral contraceptive use, particularly in relation to breast cancer. I came aboard in 1977 and started the development of methods for assessing dietary intake within that population. It's an ideal study population in which to look at the long-term effects of diet.
   We first assessed the dietary intake of the nurses in the study in 1980. We had about 95,000 women return dietary questionnaires, and we've been following them ever since. We've also continued to work on refining and improving the methods of dietary assessment, and we've been keeping track of changes in diet as time goes along. We've had repeated dietary assessments in 1984, 1986, 1990, and 1994. Clearly, this is largest long-term, detailed study of dietary intake in any population at this point in time.

            What about the Health Professionals Follow-up, and the Nurses' 2?

   Willett: The Health Professionals Follow-Up began in 1986, basically because we didn't have any men in the Nurses' Health Study. This population consists primarily of dentists, veterinarians, optometrists, and osteopathic physicians. We've collected details on dietary intake data in 1986, and updated that in 1990 and 1994. We're looking at a wide variety of cancers and cardiovascular diseases as well.
   The second Nurses' study began in 1989. Our main justification was not diet, but the same as the original Nurses' Health Study—specifically to look at oral contraceptives and breast cancer. This was a population that in 1989 was 25 to 42 years of age and made maximal use of oral contraceptives—starting with their teenage years and continuing for a long period of time. If there are any associated problems with breast cancer, it should be showing up in this population. There have been several case-control studies that have suggested positive associations. We felt this was so important that we really had to get the best possible prospective data.

            For the past ten years, the newspapers have been full of the findings from the Nurses' 1 and the Health Professionals Follow-up. Can you summarize what the most convincing findings have been so far?

   Willett: We've mainly discovered that many of our preconceptions haven't held up, and that many things we didn't expect have emerged as potentially important. For instance, one of the major factors that has shown up for breast cancer has been alcohol consumption. That now seems to be a very reproducible finding related to higher risk of breast cancer.
   With fat intake, which was a primary hypothesis that got us going, we just haven't seen anything emerge in relationship to breast cancer. For colon cancer, it's been different. We have seen an association with animal fat and specifically with high red-meat consumption. We've also looked at heart disease and found a possible association with trans fatty acid consumption.

            What exactly are trans fatty acids?

   Willett: They're a group of artificial fats that are produced in the process of partial hydrogenation. Manufacturers take liquid vegetable oil, like corn oil or soybean oil, and process it in the presence of high-temperature hydrogen and a metal catalyst. That changes the fats from liquid to solid so that they can be used in margarine and shortening, and it alters the shape of the molecules so that they function differently in terms of metabolism.

            Have you reported any findings yet from the second Nurses' study?

   Willett: Not much so far. We will probably have a report out later this year on oral contraceptives and breast cancer. But the whole intent of the Nurses' 2 population was to start off with very young, very healthy people. They don't have much cancer and heart disease. That's by design. So the major findings for that study are down the road a few years. It's really a long-term investment.

            The science of risk-factor epidemiology is controversial these days because of what people call the "carcinogen-" or "anxiety-of-the-week syndrome." It seems that every week the newspapers carry a new and usually contradictory study telling us what we should or should not eat. Is this our imagination, or is there really a problem?

   Willett: It's true; there is a problem. Part of it is this very direct link between ongoing work and what comes out in The New York Times. The natural course of science is that people do studies and report finding something, but nobody believes it too much—and, hopefully, neither do the investigators—until it's reproduced by other researchers. But in the meantime, it's on the front page of the newspaper. So there is this tendency for the least substantiated findings to be the ones coming out in the popular press, when in fact this is simply part of the scientific process, and a lot of suspected associations are not ready for the public to take action or even worry about.

            You have always preferred to work with these huge prospective studies in which one takes a population of healthy people and follows them for decades. What is the advantage of that over retrospective case-control studies in which one starts with a group of people who have a disease and examines how their lives differed from controls?

   Willett: Prospective studies have some major advantages, which is why I've been willing to make these long-term investments. We avoid the most important pitfalls of case-control studies. The first problem in a case-control study is that, yes, you can identify a large number of people who have a disease—say, cancer—but then, who are the other people you should compare them to? It's often not even theoretically clear who the right comparison group is. And sometimes, even if you can design the study so that you have the theoretically correct comparison group, usually not everyone will participate. And the people who do participate in the study may be different from the people who don't, often in health-related ways. It leaves you with an uncertainty about the right comparison.
   Then there's the added problem of what we call "recall bias." People may recall their past intake of fat differently if they have just been diagnosed with breast cancer than if you pluck them out of a random sample, phone them up out of the blue, and ask them what their past diet was.

            How can this affect the findings of a study?

   Willett: We did some prospective analyses, for example, and then did a case-control study in the same Nurses' Health Study population looking at the association of past dietary fat intake with breast cancer. Depending on whether we asked about diet before or after the diagnosis of breast cancer, we obtained a different answer. Had we relied solely on the case-control data, we would have concluded that there's a positive association between fat intake and breast cancer. But in the prospective analysis, there was absolutely no relationship. That shows that this combination of selection bias—in other words, who participates in the controls—plus the recall bias can produce some bias. What we saw was a modest bias, but we're looking for a modest effect, so it was enough to seriously distort the data.

            How modest an effect can an epidemiological study—even a big prospective study—detect reliably?

   Willett: I don't think there's any strict rule about that. Some people have said that you can't believe any relative risk under two. I don't agree with that. However, it's clear that as you're looking for smaller and smaller relative risks, there is a greater and greater possibility that they're due to distortion by other factors. But the problem is that the associations that we're looking for realistically will have relative risks of well under two—1.5 or 1.3 even. Even though they're modest, they're potentially very important. If it's something like diet, it may apply to the whole population, so it could account for huge numbers of deaths.
   We need to be able to try to look at modest relative risks of that order, and I think we can. I think the association between alcohol and breast cancer is a pretty good one, even though the relative risks are well under two. But it's been seen in so many different studies and different populations that I don't think anyone doubts that there is an association. Some might not be confident that it's a cause-and-effect relationship, but evidence is now accruing that indeed alcohol does affect estrogen metabolism in a way that would likely increase the risk of breast cancer.

            How important is it to have biological underpinnings before believing an epidemiological result of this magnitude?

   Willett: It's not absolutely necessary, because our biological understanding is very incomplete. If we had to wait for the perfect biological understanding, then there would be no use for epidemiology in the first place. And if we had just the biological effect and didn't have the epidemiology, we would say, "well, is this really relevant to breast cancer?" Add those two pieces of evidence together, however, and it really creates a much stronger body of data.

            What key questions would you like to see epidemiology answer in the next decade?

   Willett: In the area of diet and health, which is a very important part of the equation, it seems as though vegetables and fruits are beneficial, but how much and what kinds and what about them? It appears that there's something important going on in that area. We need a lot more refinement of that information. With cardiovascular disease, it seems there's something beneficial about whole grains, fruits, and vegetables, but we don't know what. This needs to be much more clearly mapped out. During the next ten years, there'll be literally a ten-fold increase in the amount of information on those issues. And then there's going to be more interest in diet early in life, but those answers are probably not going to be available in ten years. It's part of the life cycle we have not looked at very well—the early effects of diet in relationship to long-term health effects. This even includes the effects of diet during pregnancy, which, surprisingly, have been examined to a very limited degree. There are no large ongoing studies on diet during this period of life. It seems to me an obvious subject that should be given more attention.

            Last question: These large randomized epidemiological trials, such as the recent one on beta-carotene and lung cancer, have also been roundly criticized. The studies cost tens of millions of dollars, and then critics say they didn't look at the right variable. Is this a problem?

   Willett: It's a good example of how it's neither cheap nor fast to come upon good information—which is a very sobering realization. It's not clear whether we can do a randomized study for long enough to evaluate the effects of diet on cancer risk. You mention the criticism that maybe the Finnish investigators picked the wrong agent, but the beta-carotene study was not really long enough to know if there was no benefit. Particularly disturbing, of course, was that there was actually an increased risk of lung cancer in the high beta-carotene group. Given that we know cancer develops over a period of decades, it's conceivable that it might almost be impossible to answer some of these questions in randomized trials.
   It's also not automatic that we must insist on having a randomized trial in order to make a reasonable judgment about cause or protection regarding cancer. Remember, virtually everything we know about cancer prevention up to this point is based on observational epidemiology, not on randomized trials. Whether it's cigarette smoking, sunlight and melanoma, certain drugs, ionizing radiation, exercise—you go right down the list. None of that is based on randomized trials. Having said that, it puts a special onus upon us to do the best possible epidemiology that we can.