Epidemiologists who are not entirely newcomers to the discipline will remember the debate G Taubes activated by his Science paper in 1995 ‘epidemiology faces its limits’. Now he is back with a paper written to a much larger audience called ‘do we really know what makes us unhealthy’, published in The New York Times, September 16, 2007.
The history this time is about hormone use and menopause, sometimes misleadingly called hormonal replacement therapy (HRT), as if a reduction in estrogen level after menopause is a mistake made by nature that has to be replaced (corrected). Taubes is right that the general wisdom among many clinicians was that these drugs were beneficial for older women in reducing their risk of heart diseases and osteoporosis. He fails to tell that many epidemiologists, however, remained sceptical. Most epidemiologists recalled from their basic training the discussion from the 1970s between epidemiologists from Yale and Harvard on how much use of estrogen increased the risk of endometrial cancer. Estimates from the Yale group were much lower but based upon a flawed design, failing to estimate the exposure distribution in the source population that gave rise to the cases.
Taubes blames the epidemiologists for leading the public astray when they published the data from the large cohort study in 1985 (nurses’ health study) showing that women who used HRT had only one third of the heart attacks found in women who never used HRT. This wisdom was seriously questioned when a randomized trial called HERS found the opposite and these results were further substantiated when the ‘Women’s Health Initiative Trial’ in 2002 also found that use of HRT was a cardiovascular health risk for postmenopausal women.
At the time that the apparent protective effect of HRT was being promoted, other epidemiologists drew attention to the fact that taking HRT also appeared to protect women from suicide and other external causes of death such as homicide. This lack of specificity of effect was a clear signal that the cardioprotective effect of HRT was likely due to confounding. Unfortunately this information was ignored.
Taubes apparently holds the belief that epidemiology is a purely observational science: epidemiologists generated a problem using observational methods that may have killed thousands of women and the problem was corrected by non-epidemiologists. But epidemiology is not defined by its methods but by its research aims. The randomized trial is also one of the methods epidemiologists use to test causal hypotheses and those who did the randomized trials performed epidemiological studies. The problem may have been generated by epidemiologists and was corrected by epidemiologists – making mistakes are not unusual in any scientific discipline. Taubes is right in saying that the health consequences of our flawed research findings can be serious.
Let’s hope for the time being that the trial got it right but it is wise to remember that they have shortcomings as well. Randomization only works well on average in large trials and compliance to the protocol is not a random process making it difficult to estimate effect sizes in an unbiased way. The trial may work well in testing the null hypothesis but we need much more information in health planning and a trial of long duration may be seriously biased by non compliance.
The epidemiologists are usually well aware of the shortcomings of their studies and they were not the strongest advocates of HRT. Epidemiologists know that the evidence they produce may change over time and they know better than others that one should be careful when treating people without a disease. The HRT story is important and should be among the lessons we teach our students.
Jorn Olsen, Shah Ebrahim, Chitr Sitthi-amorn