Many medical studies end up concluding that two groups have different health outcomes -- death rates, heart attack rates, cholesterol levels and so forth. This difference is typically expressed as a relative change , as in the statement: "The treatment group had 50 percent fewer cases of eye cancer than the control group." The problem with this comparison is that it provides no information about how common eye cancer is in either group.
Thinking about relative changes in risk is like deciding when to use a coupon at a store. Imagine you have a coupon that says "50 percent off any one purchase." You go to the store to buy a pack of gum for 50 cents and a large Thanksgiving turkey for $35. Will you use the coupon for the gum or the turkey? Most people would use it for the turkey.
Why? Because paring half the price off $35 reaps a bigger savings --$17.50 --than cutting half off 50 cents -- or $0.25.
The analogy in health is that "50 percent fewer cases" is a very different number when applied to eye cancer -- a rare problem accounting for about 2,000 new cases in the U.S. each year -- than when applied to heart attacks -- a common problem accounting for about 800,000 new cases annually.
To really understand how big a difference is, you need to find out the starting and ending points -- sometimes called " absolute risks ." In the coupon example, the start and end points are the regular and the sales price. In a study about medical treatment, the start and end points are the chances of something happening in the untreated and treated groups.
Presenting the starting and ending point requires a few more words than presenting relative changes. For example, "In a year, two of 100,000 untreated people developed eye cancer; in contrast, one of 100,000 treated people developed eye cancer." For the price of a few more words you gain perspective: The chance of developing eye cancer is small.
Cause or Association?
Many important insights into human health come from observational studies -- studies in which the researcher simply records what happens to people in different situations, without intervening. Such studies first linked cigarette smoking to lung cancer and high cholesterol to heart disease. But not all observed associations represent cause and effect. And problems can occur when this key point is overlooked.
An example may help make the distinction clear. A man thought his rooster made the sun rise. Why? Because each morning when he woke up while it was still dark, he would hear his rooster crow as the sun rose. He confused association with causation until the day his rooster died, when the sun rose without any help.
A more serious example involves the long-held belief that most women should take estrogen after menopause. That idea, only recently discredited, also came from observational studies. The observation -- shown in more than 40 studies involving hundreds of thousands women -- was that women who took estrogen supplements also had less heart disease. But it turned out that estrogen was not the reason why this was the case. Instead, women taking estrogen tended to be healthier and wealthier. Their health and wealth -- not their estrogen supplements -- were responsible for the lower risk of heart disease.
The only way to reliably distinguish a cause from an association is to conduct a true experiment -- a randomized trial . In this type of study, patients are assigned randomly --that is, by chance--to receive a therapy or not receive it. This study design is the best way to construct two groups that are similar in every way except one -- whether they get the therapy being studied. That means any differences observed afterward must be caused by the therapy. In the case of estrogen and heart disease, such a study showed that the long-held beliefs were wrong.
Unfortunately, it is not always possible to do a randomized trial. For example, it is extremely unlikely that we could get people to agree to be randomly assigned to either eating only fast food or only organic food every day for a year (and that they would actually adhere to the diet if they did agree to be randomized). In such cases, scientists have to rely on observational studies. But when new tests or treatments are proposed, randomized trials ought to be conducted prior to their widespread use. Doctors prescribed estrogen to millions of women for many years until the randomized trial showed that intuition and dozens of observational studies were wrong.
-- Lisa M. Schwartz, Steven Woloshin and H. Gilbert Welch
