047: How Much PK Data Do We Really Need in Phase 3? A Milvexian Case Study

Statement of Purpose, Innovation or Hypothesis: Milvexian is an oral Factor XIa inhibitor being evaluated in patients at risk for thrombosis due to atrial fibrillation, stroke, or acute coronary syndrome. The milvexian Phase 3 program includes three large clinical trials. The current analysis was performed to assess the proportion of study participants from which pharmacokinetic (PK) samples should be collected to allow for robust characterization of the relationship between milvexian exposure and event outcomes (safety or efficacy).

Description of Methods & Materials: Simulations were performed for milvexian Phase 3 studies based on the proposed trial designs of ~15,000 patients with 1:1 randomization between milvexian (active treatment) and placebo/standard of care. Milvexian Phase 2 results and event rate outcomes reported for comparable compounds, including rivaroxaban and apixaban, informed the model parameters used for the simulations. Two key scenarios were evaluated with a presence or absence of a relationship between milvexian exposure and event outcomes. In the first scenario, safety/efficacy events occurred stochastically irrespective of milvexian systemic exposure levels. In the second scenario, a logistic regression-based approach was employed to characterize the relationship between milvexian exposure and safety/efficacy event outcomes. Sensitivity analyses were performed to assess the impact of different event rates (1 to 5% per year), expected between-subject variability in milvexian exposures (30 to 50%), steepness of the exposure-response slope, and trial simulation replicates (2,000 to 10,000). For each scenario, the proportion of study participants (25, 33, 50 or 75%) that required PK sampling to adequately predict event rate and characterize any underlying exposure-response relationship were evaluated; simulations were compared to when PK was collected in 100% of participants (reference). Results were evaluated using graphical assessment (e.g., boxplots) and statistical analyses to compare any differences in exposure between patients with and without an event and logistic regression to characterize any exposure-response relationship. All modeling activities were performed using R (www.r-project.org).

Data & Results: Modeling results demonstrate that PK sampling in only a subset (33% or 5,000 of 15,000) of the study population provides sufficient data to adequately predict event rates ≥1%, detect the absence or presence of the exposure-response relationship, and characterize the underlying exposure-response relationship (based on the logistic regression slope and difference in exposure between the two groups). The power to characterize the exposure-response relationship improves as the event rate increases.

Interpretation, Conclusion or Significance: The modeling and simulation framework presented herein can also be applied across other Phase 3 programs to guide the PK sampling strategy. In large clinical trials with conservative event rates, sampling in a subset of the study population can significantly reduce patient burden, improve study feasibility, and conserve resources while generating data that adequately characterizes the drug’s exposure-response relationship.

Disclosures: JLF, WZ, and MNS are employees and shareholders of Janssen Research & Development, LLC, a Johnson & Johnson Company; EB, YS, JPR, and NG are employees of Janssen Research & Development, LLC, a Johnson & Johnson Company.

Citations/References:

Additional Information/Authors: N/A