Is bottled water better?

Sir-In a French case-control study of breast cancer (Le et al., 1984), we interviewed as controls (nonmalignant disease or no disease) 403 women aged 25 to 45 years in eight French hospitals and clinics during the period 1981-1984. All these women already had an experience of sexual intercourse. The year of birth was strongly associated with the main characteristics of oral contraceptive (OC) use and with the age at the first sexual intercourse (FSI). The percentages of OC users before age 25 or before the first full-term pregnancy were much greater for the women born in the 1950s than for women born earlier (Table I). The women born in the 1950s started OC use at a younger age than the women born earlier. They also had FSI at a younger age and started OC use sooner after FSI than the women born in the 1940s (Table I). These results indicate that very deep modifications in the lifestyles of young women occurred during the two last decades, probably in part owing to OC availability. As these results reflect a phenomenon which seems to be shared by many Western countries (Population Reports, 1979), we wonder whether this important cohort effect might be a source of bias in some of the case-control studies which reported a positive association between the duration of-early OC use and the risk of breast cancer (Harris et al. (i) Case-control studies where the controls are matched with the cases on the year of birth: in Pike's studies (Pike et al., 1981, 1983) for instance, each control was matched with the case on the year of birth within 5 years, but the control had to be at least as old at interview as the matched case was at diagnosis. Thus, on the average, the controls were born 9 months earlier (the standard deviation was not reported), and the difference between the year of birth of the cases and of their matched controls was >2 years for 35% of the pairs (Pike et al., 1981). We wonder whether this small but systematic difference in year of birth is not sufficient to explain a lower percentage of OC users among the controls than among the cases. (ii) Case-control studies where the controls are matched on the age at interview: for these studies, the year of birth is well controlled for only if, on the average, the difference between the date of …