Accelerating phase 2 clinical development with real-world data (RWD): An external control arm (ECA) pilot in HER2-positive (HER2+) metastatic breast cancer (mBC).

Abstract

11172 Background: Accelerating clinical development of promising therapies is essential to improving patient (pt) outcomes. While ECAs have provided valuable supplementary evidence in Phase (Ph) 3 trials to support regulatory decision-making, ECAs in the Ph 2 setting, particularly for single-arm trials, can offer critical context for evaluating efficacy with rapidly available insights. To explore this potential, we piloted the development of a contemporaneous ECA for a newly launched Ph 2 trial of Tucatinib and Doxil in HER2+ mBC (NCT0578834). Methods: RW structured data, supplemented by unstructured data collected via chart abstraction, were sourced from iKnowMed electronic health records. The ECA cohort was designed for maximal comparability to the trial population by (1) emulating and applying trial eligibility criteria to RWD and (2) using propensity score matching (PSM) to balance covariate distributions. To demonstrate ECA feasibility in this interim analysis, a simulated dataset was created via multiple imputation based on the distribution of observed baseline characteristics in the 8 pts enrolled on the trial, expanding the trial cohort to 40 pts. Propensity scores (PSs) were estimated using logistic regression with covariates selected based on literature and expert input (age at index, number of prior treatments, prior fam-trastuzumab deruxtecan-nxki (Enhertu), and prior tucatinib). PSM was performed using greedy nearest-neighbor matching (caliper ≤0.25 standard deviation of logit of PSs) without replacement at a 1:1 ratio. Results: The simulated trial cohort included 40 pts, and the ECA cohort included 77 pts. After PSM, 82% of the trial cohort were successfully matched, with 18% excluded due to non-overlapping PSs, resulting in matched cohorts of 33 pts each. Following PSM, most variables in the PS model achieved a standardized difference (SD) ≤0.10, indicating adequate balance between trial and ECA cohorts. The mean age (61 vs. 60 years in the trial and ECA cohorts, respectively; SD = 0.03), distribution of number of prior treatments (e.g., 21% vs. 27% with 1 prior treatment; SD = 0.09), and prior tucatinib exposure (48% in both cohorts; SD = 0) were well balanced between cohorts. However, prior Enhertu exposure remained imbalanced, with 45% of trial patients receiving prior Enhertu versus 33% in the ECA cohort (SD = 0.29). Conclusions: This pilot demonstrates that RWD can be quickly and efficiently assembled in pace with an ongoing Ph 2 clinical trial, offering promise for accelerated decision-making in clinical drug development. When small sample sizes in Ph 2 trials pose challenges, this approach shows that an exchangeable ECA can be achieved as trial accrual progresses, providing a framework for ECA evidence generation.