November 1987

Epidemiologic evidence is particularly prone to this problem. Epidemiology, the study of the causes and patterns of disease in human populations, is misunderstood so often precisely because it appears to be so accessible. The epidemiologist's job can be likened to that of a detective--he cracks the code of the disease by uncovering its mode of operation. To do this, an epidemiologist can do several things. One is to follow a large group of people for a long period of time, see what diseases they develop, and then correlate the diseases with other factors in their lives. An example of such a "cohort study" is the Framingham heart study, in which a sample of the adult population of Framingham, Massachusetts, has been followed since 1948, in order to tease out factors that influence their risk of heart disease. Another epidemiologic method, called the case-control approach, involves comparing victims of a particular disease or condition with people who don't have it, and looking for a significant difference between the groups in some variable like diet, living conditions, age, or ethnic background. An example would be a study published in 1979 in The Lancet, for which the researchers had looked at oral-contraceptive use by a group of women who had suffered heart attacks and by another group who hadn't.

The purpose of both approaches is to piece together a pattern that will point to the underlying cause, or causes, of a particular disease or syndrome. An individual study provides clues but is not designed to provide proof that a particular disease is caused or prevented by a particular agent. Since epidemiologic research involves people rather than laboratory animals, it is especially tempting to extrapolate from the sample in a particular study to the public. This is generally a mistake.

THE recent front-page reports of a link between breast cancer and moderate alcohol consumption offer an excellent example of how misleading such extrapolations can be. The studies prompting the headlines, published last May in The New England Journal of Medicine, were done by Dr. Arthur Schatzkin, a senior staff fellow at the National Cancer Institute, and Dr. Walter Willett, an associate professor of epidemiology at the Harvard School of Public Health. Both were cohort studies. Both found an association between moderate alcohol consumption--as few as three drinks of beer, wine, or liquor a week--and an increased risk of breast cancer, an increase of as much as 50 percent. These papers were accompanied by an editorial by Saxon Graham, an epidemiologist at the State University of New York at Buffalo. Graham recommended that women at high risk of breast cancer "curtail their alcohol ingestion." Such women, he said, include those who have a family history of the disease, who are obese, who were older than twenty-five when they had their first child, or who have few or no children.

This, clearly, is no small group, and it would seem that millions of women are risking decades of life for the sake of an occasional quick one. Such is not, however, the case. Understanding why requires a grounding in statistics.

Breast cancer is an extremely common disease. One in eleven women, or about nine percent of all women, will get it at some point. It seems logical to infer from this fact, in combination with the studies' estimates of increased risk, that drinking half a bottle of wine a week increases one's risk of breast cancer from nine to nearly 14 percent--a sobering prospect. Several reports made just this inference, but it is incorrect. As the Harvard Medical School Health Letter pointed out in a follow-up of the Schatzkin and Willett studies, published last July, the figure of nine percent includes all women, while the estimates of a 50 percent increased risk given by both studies are based on a comparison between drinkers and non-drinkers. Also, the studies suggest that the women whose risk of breast cancer is most increased by drinking are women otherwise at low risk for the disease--young women of normal weight with no family history of breast cancer. Very few people in this category get breast cancer, so it is easy for any variable to appear to the lay person to modify greatly their risk. A statistician's evaluation of the data, undertaken at the request of the Harvard newsletter, concludes that if these studies accurately reflect incidence in the population as a whole, moderate drinking is responsible for only one or two percent of all breast-cancer cases.

Does this mean that one to two percent of all women who get breast cancer bring it upon themselves by enjoying an occasional beer? Not necessarily. There are factors that confound even this relatively modest conclusion--factors that have to do with the way the researchers chose to frame their studies. For example, neither study took the subjects' early diet into consideration. What a woman ate as a child and teenager is suspected of having a significant effect on her risk of breast cancer. But since the diet-cancer connection has not been proved, and since it is almost impossible to get information on early diet (how many fifty-year-olds remember in detail what they ate as teenagers?), early diet was not taken into consideration in either the Willett or the Schatzkin study.

Nor was exercise considered. Research done at the Harvard School of Public Health and the State University of New York at Buffalo suggests that exercise, especially during childhood and adolescence, decreases the risk of breast cancer. But, again, this research is not conclusive, so exercise was not included as a variable in either study.

This does not mean that the studies are incorrect or misleading. In fact, there have been at least a dozen other reports linking breast cancer with alcohol consumption. One recent study was published in the April issue of the Journal of the National Cancer Institute. Elizabeth Harvey, an epidemiologist who was also a co-author on the Schatzkin study, concluded in the NCI Journal report that

drinking even moderate amounts of beer, wine, or liquor was, indeed, associated with increased breast-cancer risk, but only for women who began drinking when they were under thirty years old. What women drank after they hit thirty seems to have had no significant effect on their risk. Harvey told me that this paper received little publicity, and it was not acknowledged in the New England Journal of Medicine editorial. But what concerns her is that all the press and television reports have called for a change in women's drinking habits because of the other studies. She considers particularly misleading the comments of a network-television medical reporter, who suggested that viewers already at high risk of breast cancer should abstain entirely. However, such comments are certainly understandable in light of Saxon Graham's NEJM editorial. Graham recommended that women at high risk cut back on their drinking not because the science necessarily dictated it but because he believed it to be prudent. Journalists simply took his recommendations one step further.

The NEJM published the Schatzkin and Willett studies back to back, along with the Graham editorial, because the journal's editors realized that they were provocative, not only to the informed academic community but also to the public. Health and medical reporters commonly subscribe to The New England Journal of Medicine--it is the most comprehensive, easily understood, and highly publicized journal in its field--and so the Schatzkin and Willett papers received much more attention than they would have if they had appeared in some more obscure or specialized publication. It is clear that the authors wanted as broad an audience as possible--an audience that would include journalists and other lay people.

The alcohol-and-breast-cancer studies are valid and important pieces of work, but, as the authors insisted, they are preliminary. No one can assume a cause-and-effect relationship between drinking moderate amounts of alcohol and contracting breast cancer. It is quite possible that something that goes along with drinking, such as eating cocktail snacks or sitting on bar stools, is what increases the risk of cancer. It could also be that women who drink, on average, exercise less or eat more fat or consume fewer vitamins than women who do not drink. Drinking might thus be a mere artifact, the fall guy for some other, as yet unknown culprit.

EPIDEMIOLOGIC studies are particularly vulnerable to premature interpretation by both the press and scientists eager to stimulate public interest. They are naturally intriguing. But they cannot, by definition, provide proof of the cause of a problem. The associations they uncover can seem overwhelming, and when enough studies over a long period have pointed to the same conclusion, scientists will agree to accept an agent or combination of agents as the "cause" of a disease. The majority of smokers, for example, do not get lung cancer, and scientists cannot say with absolute certainty that some other, predisposing factor is not involved. Still, the association between lung cancer and smoking is so strong--smokers have a risk of developing lung cancer ten to twenty times greater than that of nonsmokers--that smoking is generally accepted to be an important cause of lung cancer. Tobacco interests continue to dispute this assumption. The statistical link between breast cancer and moderate alcohol consumption is by comparison extremely weak: light drinking appears not even to double a woman's risk of getting cancer. Such a weak association does not allow conclusions to be drawn. It merely suggests a relationship that deserves further exploration.

Epidemiologic studies are prone to misinterpretations for other reasons, too. For example, it is almost impossible to get a truly random sample of the population to study, and the choice of study participants can markedly bias results. A rather extreme example of this is a study done several years ago that showed an association between pancreatic cancer and coffee consumption. It was a retroactive study, in which patients hospitalized for the treatment of pancreatic cancer were compared with a control group of patients, over 30 percent of whom were hospitalized for other digestive disorders. The authors of the study found that the cancer patients drank significantly more coffee than the other patients, which caused them to suspect a link between coffee drinking and cancer. Their methods were later challenged by other experts, who pointed out that patients with digestive disorders are likely to restrict their coffee intake. Clearly, the control group selected by the research team was entirely inappropriate. Some subsequent studies have suggested a more tentative association between coffee and pancreatic cancer, but the public still links the two strongly because of the highly publicized study.

Another problem is that epidemiologic research is often heavily dependent on subjects' reports of their own behavior. Memory is notoriously selective, and tends to screen out what it would prefer had never happened. For example, the continuing confusion over whether obese people truly eat more than people of normal weight is due in part to the fact that it is nearly impossible to get good dietary histories from anyone, let alone from the obese. Thus scientific reports suggesting that obese people eat no more than people of normal weight share shelf space with reports claiming quite the opposite.

The inherent limitations of the epidemiologic approach do not render it invalid. But like most scientific research, population studies provide information that is meaningless when considered in isolation. There are very few, if any, genuine "breakthroughs" in science or medicine. Yet too often researchers are pressed into implying, and journalists into reporting, that breakthroughs have been made--as one researcher told me, scientists, like everyone else, are driven by ego. But as a rule it is the gradual accumulation of evidence that leads to scientific understanding. It is up to the public to stop expecting miracles and to maintain an open but critical mind.