Ziheng Hu 1
1 Merck & Co., Inc. (Rahway, USA)
Introduction
PK bridging studies are routinely conducted to extrapolate efficacy from one population in which efficacy has been established (reference population, e.g. adults) to another population of interest (target population, e.g. pediatrics) via exposure matching. Doses for the population of interest are typically selected by accounting for various assumptions about the demographics and/or PK characteristics of the population of interest. The selected doses and these assumptions are then evaluated in a PK study. In cases where high uncertainties exist around these assumptions, an interim analysis (IA) based on a small “sentinel” cohort can be useful for early detection of clinically relevant deviations between observed and predicted PK, and if warranted, for triggering re-evaluation of PK and dose selection.
Objectives
The goal of this research is to establish a methodological framework for determining the optimal IA criteria to allow detection of clinically relevant deviations between the observed exposure in the sentinel cohort (at IA in a PK bridging study) and the target exposure, and to optimize the operating characteristics of the IA criteria.
Methods
During the IA, the observed exposure in the sentinel cohort is compared to the target exposure by calculating the geometric mean ratio (GMR) of observed versus target exposure and its 90% confidence interval (CI). The IA criteria is typically defined in the form of lower and upper bounds of the GMR (also referred to as IA bounds). If the 90% CI of the GMR falls within the IA bounds, the study will continue as planned. Otherwise, re-evaluation of PK and the selected dose may be triggered.
Monte Carlo simulations were conducted to assess the operating characteristics of IA criteria. Different scenarios were considered by varying the factors that potentially impact the operating characteristics, including the maximum acceptable true difference (MATD) from the target exposure, magnitude of PK variability, and IA sample size. The MATD is defined as the largest deviation that is not considered clinically relevant and should not trigger PK re-evaluation (e.g. if MATD = 20%, then up to 20% difference in the true underlying geometric mean exposure is acceptable). To simulate observed PK exposure at the IA, individual exposures were randomly sampled from a lognormal distribution with an assumed geometric mean accounting for the MATD and an assumed variance. Under each scenario, the simulation was repeated 1000 times, and outcome of the IA for each replicate was determined as described above (“trigger” vs “not trigger”). The specificity (1 – probability of falsely triggering the IA criteria) under each scenario (where true difference is
For a smaller MATD of 10% (CV% = 50%, IA sample size = 10), a narrower IA bound of (0.58, 1.72) is required to maintain a specificity > 90%. Based on this criterion, if the true underlying difference is 40%, the sensitivity will be 91%.
Conclusion
This work presents a practical framework for designing IA in PK bridging studies to facilitate early detection of clinically relevant deviations between observed and target exposure. The operating characteristic analysis revealed that the sensitivity and specificity of the IA criteria are impacted by the IA sample size, the MATD, the selected IA bound, and the CV%. This framework enables rational selection of IA decision criteria by balancing specificity and sensitivity. The utility of this methodology may be generalized to other clinical trials with PK as the primary endpoint.
Reference: PAGE 34 (2026) Abstr 12034 [www.page-meeting.org/?abstract=12034]
Poster: Methodology - Study Design