Feasibility of constructing an external control arm for early efficacy assessment in phase 1a: Methodological framework from the INVOKE trial.

Abstract

11521 Background: Phase 1a dose-escalation trials are primarily designed for safety but often lack sufficient visibility into efficacy signals due to small sample sizes and the absence of a comparator arm. This uncertainty can delay decision-making and increase late-stage failure rates. We hypothesized that a synthetic External Control Arm (ECA) aggregating real-world data (RWD) and historical clinical trials could be established to contextualize efficacy readouts from the ongoing OKN4395 Phase 1a study (INVOKE; NCT06789172; Trial in Progress abstract submitted to this meeting). Methods: We curated a retrospective dataset of 298 solid tumor patients derived from two complementary sources, RWD and recent clinical trials (inclusion date ≥ 2016). To ensure comparability with the INVOKE population, patient selection adhered strictly to the study’s inclusion and exclusion criteria, with only individuals who had exhausted all standard-of-care (SoC) options being retained. We further refined the cohort to ensure exact overlap, retaining only external patients who shared an identical baseline covariate profile with at least one patient in the INVOKE cohort. Finally, we applied statistical adjustment methods, including Inverse Probability of Treatment Weighting (IPTW) and G-Computation, to align the remaining cohort on major prognostic variables (e.g. ECOG status, prior lines of therapy, inter alia). Results: A total of 13 INVOKE patients with available endpoint data were eligible for ECA matching. The selection process yielded a final robust ECA cohort of 40 patients. In this refined population, we achieved robust covariate alignment between the ECA and the INVOKE cohort, with absolute Standardized Mean Difference (SMD) < 0.13 for all major prognostic variables. Despite the methodological differences between IPTW and G-computation, both statistical approaches yielded highly consistent estimates. Data from INVOKE, a study in early stages, is yet to publish data (TIP presented at this meeting), comparative efficacy data will be disclosed at a later date. Conclusions: We demonstrated that constructing a synthetic ECA generating outcomes directly comparable to those of the population of an active Phase 1a dose-escalation study is methodologically feasible. The establishment of this matched external cohort and first application to the INVOKE Phase 1a study provides a proof of concept framework for early efficacy assessment. This "de-risking" tool is now positioned to provide rapid, concurrent contextualization of emerging data from the INVOKE trial upon data maturity.