A Translational Approach to Study Drug-Drug Interaction with Hormonal Contraceptives: From Pharmacoepidemiology to Mechanistic Modelling

Background: Oral combined hormonal contraceptives (CHCs) are commonly used worldwide, but their efficacy may diminish due to concomitant CYP3A4 inducers. Different progestins in CHCs vary regarding fraction metabolized by CYP3A4 (fmCYP3A4) and may have different efficacy reductions by such potential drug-drug interactions (DDI).

Objectives: To compare the risk of CHC failure, i.e., unintended pregnancy, in concomitant use of CYP3A4 inducer antiepileptic drugs with CHCs containing levonorgestrel (LNG, low fmCYP3A4)) or drospirenone (DRSP, high fmCYP3A4) via real-world evidence (RWE) and physiologically-based pharmacokinetic (PBPK) modeling approaches.

Methods: Our RWE study used a retrospective cohort design to identify females of childbearing age (18-48) in MarketScan® Research Databases (2005-2018) who were diagnosed with epilepsy, personality, or depressive mood disorder. Patients entered the cohort if we observed concomitancy of CHCs with CYP3A4 inducer antiepileptic drugs (carbamazepine, oxcarbazepine, phenytoin, or topiramate). We excluded patients with infertility or hormonal dysfunction diagnoses. Concomitant use was determined based on information available from pharmacy dispensing data. During follow-up, we operationalized contraception failure with our team's previously developed algorithm that uses pregnancy-related diagnoses and procedures to characterize pregnancy start date, duration, and end date. We used a generalized linear regression model to estimate the rate ratio for contraception failure, comparing DRSP and LNG cohorts. We used treatment propensity score and standardized mortality ratio weighting to adjust for measured confounders. Lastly, we used PBPK simulations of DRSP and LNG administered alone or with carbamazepine to obtain drug exposure changes in scenarios of DDI. Ratios of drug exposure under the concentration-time curve (AUC) were used to create ratios of DDI vs non-DDI AUC (AUCR) to compare PK changes of the two drugs to our RWE analysis results.

Results: We identified 17,039 subjects in the LNG cohort and 15,254 subjects in the DRSP cohort. Incidence rate ratios were 1.70 [1.09, 2.67] and 1.53 [0.94, 2.43] for the unadjusted and adjusted analyses, respectively. PBPK simulations resulted in AUCRs of 0.41 and 0.47 for LNG and DRSP, respectively, resulting in an exposure ratio of 0.87 between the two drugs in DDI scenarios.

Conclusions: Our translational approach showed consistent directions in the modeled PK mechanism and real-world contraception failure. While the magnitude of relative differences between progestins was not similar, a linear ‎relationship between reduced AUC and contraception failure is unlikely and requires further ‎investigation.‎

*The first two authors contributed equally.