in other reviews (Schlesselman et al., 1989; Thomas, 1989), the results of case-control and cohort studies in middle-aged women have been reassuringly negative. Some commentators, however, have suggested that a new phenomenon may be emerging (McPherson et al., 1986; Lund, 1989).
Interpreting the positive findings in several recent case-control studies is difficult for two main reasons. The first problem is one of shifting goalposts: the positive subgroups keep changing. For example, Pike's group in Los Angeles reported in 1981 that women who used the pill before their first full-term pregnancy had an increased risk of breast cancer at a young age (Pike et al., 1981). Oral contraceptive use after the first pregnancy was not associated with any change in risk. In 1983, the same group reported on an expanded analysis (Pike et al., 1983). There was a strong association with use of the pill before age 25: the relative risk was estimated to be 2.0 for four to six years of use and 4.9 for more than six years of use before age 25. The previous association with use before the first pregnancy was now attributed to a positive correlation between this variable and use before age 25. Attention was subsequently focused again on use before the first pregnancy by the results of a case-control study conducted in Britain by McPherson and colleagues (McPherson et al., 1983, 1987). They found a strong association with use before the first pregnancy but no effect of use after it—indeed, there was a suggestion that use occurring only after the first pregnancy might be protective.
Pike, McPherson, and Vessey were then instrumental in setting up the U.K. National Case-Control Study, with 755 cases under age 36 and an equal number of controls (U.K. National Case-Control Study Group, 1989). The results published in 1989 showed a highly significant trend in risk of breast cancer with total duration of pill use. But not only were the relative risk estimates much closer to 1.0 than in the previous studies; there was now no greater effect of use before the first pregnancy than after the first pregnancy. The results were presented clearly, but many readers apparently did not appreciate that the authors' previous hypothesis (about a risk confined to use at a specific time in early reproductive life) had been rejected.
Clearly, use before the first pregnancy, use before age 25, and total use will all be correlated, and many studies will have insufficient power to distinguish their effects with confidence. There is a danger, however, in focusing attention on the subgroup in each study that happens to give the highest relative risk estimate. This source of bias is illustrated by the summary table in a recent British review (Chilvers and Deacon, 1990), which shows only the subgroup in each study that gave the most positive results (ignoring the fact that negative results were obtained in some subgroups listed for other studies).