Bias versus causality: interpreting recent evidence of oral contraceptive studies

Late in 1995 and early 1996, 4 epidemiologic studies were published that resulted in a crude mean weighted relative risk of approximately 2 when third-generation oral contraceptives were compared with second-generation oral contraceptives as risk factors for venous thromboembolism. This article reviews empirical evidence on bias or systematic error that may have influenced the estimates of association. The Bradford-Hill criteria to distinguish causality from an observed association were used to consider whether third-generation oral contraceptives cause an apparent excess in the occurrence of venous thromboembolism. Bias is more likely than a causal relationship to explain the associations observed for venous thromboembolism. For myocardial infarction, bias may mask the full benefit of third-generation oral contraceptives. For stroke, the question of causality is moot because statistically significant differences between third- and second-generation products have not been detected. The clinical importance and the public health significance of any differences among the various products with respect to adverse cardiovascular outcomes are trivial and undetectable because of the extremely low incidence of those disorders among users of oral contraceptives. The oral contraceptive pill is 99.9% effective when used correctly. All oral contraceptives on the market are safe and getting safer.