Oral contraceptive use and venous thromboembolism: a consideration of the impact of bias and confounding factors on epidemiological studies

Abstract

Recent epidemiological studies have reported an approximately two-fold, significant, increased odds ratio for venous thromboembolism in users of third-generation oral contraceptives (OCs) compared to users of second-generation OCs. However, in each study, this association is of borderline statistical significance, and the studies do not indicate an increase in the absolute risk of venous thromboembolism for users of third-generation OCs compared with the data from prior studies primarily involving second-generation products. The data derived from these recent studies show a lower reported incidence of venous thromboembolism with the use of second-generation OCs over time compared to the prior studies, while the incidence of venous thromboembolism for use of third-generation OCs is approximately equal to that previously reported for second-generation products. Generally, evidence has shown that the risk of venous thromboembolism and the impact on hemostatic parameters are reduced with declining estrogen dose. In addition, there is no evidence of a clinically significant effect of the OC progestogen doses on hemostatic parameters. These inconsistencies point to factors other than a causal relationship to explain the higher risk of venous thromboembolism in users of third-generation OCs. An examination of factors not included in the recent studies has identified at least three potential biases that should be considered: prescription bias, a 'healthy-user' effect and referral bias. Available data on prescription bias come from marketing research surveys, databases of prescribing patterns, epidemiological databases and the recent studies. These data indicate that the recently introduced third-generation OCs have been more extensively used by younger women, for shorter periods of time, and by women with risk factors for venous thromboembolism that have the second-generation OCs. There is also evidence for a healthy-user effect, whereby women who were most susceptible for venous thromboembolism have left the cohort of users of second-generation OCs as a result of a venous thromboembolism during pregnancy or early OC use, and thus were possibly not included in these recent studies. The women remaining in the cohort, who are included in the recent studies, are less susceptible to venous thromboembolic events than are those making up the cohort of women using third-generation OCs. In addition, there are indications that referral bias has occurred because women with risk factors for venous thromboembolism (who disproportionately receive third-generation OCs) are more likely to be referred to a hospital for investigation of possible symptoms of venous thromboembolism, and thus users of third-generation OCs have the potential to be over-represented among cases. These biases act spuriously to increase the observed odds ratio for the first-time occurrence of venous thromboembolism amongst users of third-generation OCs, when compared to users of second-generation OCs. None of the recent epidemiological studies have been able to adjust for the three major biases. These biases are likely to contribute to, or even totally account, for the small increased risk of venous thromboembolism observed for users of third-generation OCs when compared to that for users of second-generation OCs.