Background: In studies of oral contraceptive oral contraceptive use and risk of venous thromboembolism, bias related to heightened diagnostic suspicion and preferential referral of oral contraceptive users has been an issue. The aim of our study was to determine the presence and potential impact of diagnostic suspicion and referral bias.
Methods: We conducted a case/non-case study in 21 regional study centers in Germany and, in parallel, a conventional case-control study using the same cases but randomly selected population controls from the same areas. Women with symptoms compatible with venous thromboembolism were included in the study between 1994 and 1999, and classified as cases or non-cases (first reference group) according to the diagnostic work-up (case/non-case study). A second reference group consisted ofpopulation controls (conventional case-control study as an internal comparison for the case/non-case study): 606 cases, 462 non-cases and 2942 population controls aged 15-49 years. Adjusted unconditional regression analyses were performed.
Results: Adjusted odds ratios for venous thromboembolism in oral contraceptive users were systematically higher in the classical case-control study compared to the case/non-case approach (using the same cases) across all subgroup analyses (e.g. for idiopathic cases, the odds ratio was 67% higher in the case-control study: 4.33 (95% confidence interval (CI) 3.27-5.74) versus 2.60 (95% CI 1.75-3.88)). We found a significantly increasing trend of oral contraceptive use in four categories of increasingly sophisticated diagnostic tests that were applied to 1067 women with a suspicion of venous thromboembolism, irrespective of the outcome. Stratified analysis showed the diagnostic level to be a confounder.
Conclusion: In our population-based study with the possibility of internal comparison, we found clear evidence that diagnostic suspicion and referral bias does play an important role in case-control studies of venous thromboembolism risk among oral contraceptive users. This underlines the importance of making an effort to avoid this bias when designing a new study.