(009) Breast cancer recurrence due to avoidable drug-drug interactions? Developing a protocol for a target trial in women with breast cancer who concurrently use tamoxifen and antidepressants

Background: Evidence-based guidelines recommend caution with concurrent use of tamoxifen and CYP2D6 inhibiting antidepressants (AD) due to potential pharmacological interactions that may reduce tamoxifen’s effectiveness. However, randomized controlled trials are ethically not feasible and previous observational studies are inconclusive and showed methodological deficits.

Objectives: To develop a protocol for a target trial to study the risk of BC recurrence in women concomitantly using tamoxifen and CYP2D6 inhibiting ADs while avoiding common sources of bias in observational analyses.

Methods: We developed a protocol for the ideal (“target”) trial that would answer our research question by defining eligibility criteria, treatment strategies and assignment, follow-up, outcome, causal contrast of interest and the statistical analysis. Unlike many existing target trial protocols, special attention needed to be given to how the drug-drug interaction would be investigated in a hypothetical randomized study. Further, we investigated how the target trial could be emulated based on data of the German Pharmacoepidemiological Research Database (GePaRD; claims data of ~20% of the German population).

Results: Our ideal trial aimed to estimate the effect of newly using CYP2D6 inhibiting AD for at least one year (strategy A) vs using non-inhibiting AD for at least one year (strategy B) in tamoxifen users on the risk of BC recurrence in the same breast, diagnosed within 5 years of baseline. We explored alternative formulations, but these were of less practical relevance. The emulation required several adaptations according to the constraints giving by the available data: database-related adaptations of the eligibility criteria (e.g. at least three years of cancer-free database history before BC diagnosis) as well as adaptations to obtain larger sample sizes (e.g. extension of the enrolment period). Even though GePaRD includes ~50,000 women with incident BC, only ~2,000 met our inclusion criteria out of which less than 350 fully adhered to either strategy A or B. The per-protocol effect may be best estimated using the parametric g-formula as inverse probability weighting is less stable in case of positivity violations.

Conclusions: Explicitly devising a target trial is especially helpful in the context of complex exposures. The formulation of a target trial was helpful to specify and clarify the research question. Emulating such a trial still requires some compromising, but is useful to improve interpretation and relevance of results and reduces the risk of avoidable biases.