# Drug Interactions With Tamoxifen and Treatment Effectiveness in Premenopausal Breast Cancer Patients: A Bayesian Joint Modeling Approach

## TL;DR

This study examines how drug interactions with tamoxifen affect breast cancer recurrence in premenopausal women.

## Contribution

The novel use of Bayesian joint modeling to assess drug interactions with tamoxifen in breast cancer treatment.

## Key findings

- Bayesian joint models showed a slight increase in recurrence risk with CYP2D6 inhibitors.
- No significant recurrence risk was found with CYP2C19 or CYP3A4/5 inhibitors using Bayesian modeling.
- Traditional Cox regression models gave less plausible results compared to Bayesian joint models.

## Abstract

Tamoxifen is guideline treatment for premenopausal women with estrogen receptor‐positive (ER+) breast cancer. Therapeutic efficacy relies partly on tamoxifen biotransformation by CYP2D6, CYP2C19, and CYP3A4 enzymes. We conducted a cohort study to evaluate whether concomitant prescription of drugs that inhibit these enzymes impacted breast cancer recurrence.

We enrolled 4493 premenopausal women with stage I–III ER+ breast cancer (2002–2011) treated with tamoxifen. We defined time‐varying CYP‐inhibiting drug exposures as the proportion of overlapping days during the tamoxifen treatment period. We estimated associations of concomitant medication use with recurrence using: (1) Bayesian joint modeling (hazard ratio [HR] and 95% credible intervals [95% CrI]), (2) traditional Cox regression (HR and 95% confidence intervals [95% CI]).

During tamoxifen therapy, 13% of the cohort used strong CYP2D6 inhibitors, 31% weak CYP2D6 inhibitors, 37% CYP2C19 inhibitors, and 12% CYP3A4/5 inhibitors. Bayesian joint models showed that women with ≥ 50% overlap between tamoxifen and CYP2D6 inhibitors had increased recurrence risk compared with 0% overlap (HR: 1.24, 95% CrI: 0.96, 1.58). No recurrence association was seen for CYP2C19 inhibitors (≥ 50% vs. 0%, HR = 1.0, 95% CrI: 0.69, 1.40), but traditional Cox models yielded positive associations for CYP2C19 overlap (≥ 50% vs. 0%, HR = 1.45, 95% CI: 1.07, 1.96). With Bayesian joint models, we observed no association between ≥ 50% versus 0% overlap with CYP3A4/5 inhibitors (HR: 0.84, 95% CrI: 0.32, 1.93).

With Bayesian joint modeling, we saw a slight increase in recurrence among CYP2D6‐inhibitor users, but no increase among CYP2C19‐ or CYP3A4‐inhibitor users. Results from Cox regression models were less plausible.