Guilt by Association

Guilt by Association

Part of a series of occasional columns on making use of new health information.

News items:

“Eat more bananas, oranges… -and you may cut your risk of stroke.”

“Chemical in broccoli fights ulcers, stomach cancer.”

“Too much vitamin A may increase the risk of hip fractures in older women…”

“Eating fish cuts stroke risk.”

“Beans, beans, they’re good for your heart.”

“Researchers say regular drinking lowers men’s risk of heart attacks.”

An ad suggests that lycopenes in tomato products may reduce the risk of prostate and other cancers.

What should you believe? What should you do?

As mentioned in an earlier column, randomized controlled trials can provide proof of a causal relationship, while observational studies can only reveal associations. However, because observational studies are more practical they account for the majority of information we get about nutritional factors.

Unfortunately, “associations” are subject to confounding and therefore may be misleading. This may sound like a hypothetical problem, but it is actually one of the most important problems in observational studies. It is real and it may account for some of the inconsistent reports you hear.

What is “confounding?” Suppose an observational study finds that alcohol consumption is associated with increased death from cancer. But suppose that smoking, lower consumption of fruit and decreased exercise each actually causes increased death from cancer. And suppose that increasing alcohol intake is associated with more smoking, lower fruit intake and less exercise, but does not cause any of the three. The association observed between alcohol consumption and cancer death is confounded by the increased smoking or by decreased exercise or fruit intake.

Unless you can separate out the effects of smoking, fruit and exercise, you can’t tell whether alcohol consumption is actually causing increased cancer deaths or simply appears “guilty” by being associated with “real” causes like smoking.

Investigators often try to adjust for known confounders like gender, age, smoking, caloric intake, family history, physical activity. However, adjustments are only approximations and can only be made for known confounders.

Unknown confounders are an important problem. Maybe the association observed recently between vitamin A intake and hip fractures is confounded by something else that persons who consume high vitamin A also tend to eat more often—or to avoid. That unknown something else could be a real cause of increased hip fractures, while vitamin A might not be, but looks like it could be because of its association with the unknown factor.

In observational research, people make their own decisions about what they do or don’t ingest, and they are observed, forwards or backwards in time, to see what their health outcomes are. The investigators often must rely on information gathered from the participants through dietary questionnaires.

This information is, at best, an approximation of their dietary habits. People have to remember and summarize their general habits of the past, perhaps the past year—or more. An important unknown is what time period is actually relevant? If the study is looking at the development of cancer over the next several years, is intake last year, or 5 or 15 years ago important? So there are problems remembering, problems reducing varying habits to one description and problems with timing—not knowing what period in a person’s dietary history to consider.

Randomized controlled trials, on the other hand, should be free of confounding (when conducted correctly) and don’t rely on the participants remembering or summarizing their intakes. Unfortunately, such trials are uncommon in nutritional research because they are generally impractical. At times, they may even be unethical, for example, if some participants would have to ingest specified amounts of something thought to have some adverse effects (like alcohol or trans-fats). Randomized controlled trials tend to be small and relatively short, and if a study ends without finding an effect, it’s possible that something was missed because it takes longer to occur.

Consequently, results from nutritional studies have often yielded inconsistent, and therefore, inconclusive results. For example, questions such as does fruit and vegetable consumption or overall fiber intake affect the risk of colorectal cancer remain unanswered because of inconsistent findings.

The issues are immensely complex, often controversial and generally only partially worked out. Yet we can make changes in what we ingest at will. Be wary. The latest study is not necessarily the best. Avoid rushing to change your dietary habits based on the morning news.

Mark H. Zweig, M.D., previously a staff physician at NIH, has more than 35 years of experience in medical research and patient care, with particular interest in medical decision-making and preventive medicine.