|Biases in Epidemiological Studies||To Epidemiology theme page
To EBM theme page
1. Core Knowledge:
All too often, the results of earlier studies are overthrown by more recent ones. And sometimes, the findings flip-flop back and forth, leaving patients confused and clinicians irritated.
A classic example is the use of hormone replacement therapy (HRT). In part due to a 1966 book, Feminine Forever by gynecologist Dr. Robert Wilson, estrogen therapy evolved into a long-term remedy for the chronic ills of aging. He argued that menopause was an illness, akin to diabetes or kidney failure, that could be treated by taking estrogen to replace the hormones that a woman’s ovaries secrete in ever diminishing amounts as she ages. Literally millions of women were prescribed HRT.
In 1985, the Harvard Nurses’ Health Study (a huge cohort study of 122,000 women) reported that women taking estrogen had only one third as many heart attacks as women who had never taken the drug.
By the mid-1990s, the American Heart Association, the American College of Physicians and the American College of Obstetricians and Gynecologists had all concluded that the beneficial effects of HRT were sufficiently well established that it could be recommended to older women to ward off heart disease and osteoporosis.
But a warning bell sounded in 1998, when a clinical trial called HERS, for Heart and Estrogen-progestin Replacement Study, was undertaken on 2,700 women who already had heart disease. It concluded that estrogen therapy increased, not decreased, the risk of heart attack.
Then, in the summer of 2002, the Women’s Health Initiative trial, involving 16,500 women, concluded that HRT caused more harm than good. Millions of women abruptly ceased taking estrogen therapy.
What had happened?
While difficult to undertake, observational studies are nonetheless cheaper and easier than experiments, but observational studies frequently show a greater effect than is seen in a randomized trial. Why may this be?
An epidemiologist named Gary Taubes lists the following:
The Bias of healthy users, and the Compliance bias. A study of nurses will involve health-conscious, informed consumers who are more likely than average women to take the hormone as prescribed. This sort of person also tends to be thinner, to exercise more, to have fewer risk factors for heart disease and to be more educated and wealthier. Because these factors are associated with reduced risk of heart disease, the study may have given an unduly positive picture.
The Prescriber effect and the Eager patient effect. The interaction between certain doctors and some of their patients can exert a strong placebo effect. Doctors involved in research tend to transmit their interest and enthusiasm; the eager patient tends to ask for the latest product that the average patient would not have heard of, so will tend to get onto the medication being studied. The eager patient also differs in other ways (more compliant, etc) that may have an independent effect on the outcomes.
Errors in classification: it is often difficult in observational studies to know precisely what patients are taking. And not just to know the medication of interest, but a whole range of other nutrients and products. In addition, in very large studies there will be difficulty in accurately measuring outcomes such as menopausal symptoms.
Updated August 21, 2014