(C-199) A Framework for Combining Diverse and Complementary Datasets to Create External Comparator Arms

Background: External control arms (ECAs) have been leveraged for regulatory and Health Technology Assessment submissions in oncology and rare disease; however, identifying ECA data for studies, particularly in rare diseases or subtypes, often requires combining real-world data for the same or different patients, from disparate sources, geographies, or time periods. Limited guidance exists on the framework (why, when, and how), use cases, and considerations for combining data sources to generate fit-for-purpose ECAs for single-arm trials (SATs) in clinical development programs.

Objectives: To describe a conceptual framework and use cases for combining datasets to generate robust ECAs for SATs.

Methods: We describe various motivations for combining data for an ECA, including situations where complementary data is needed. Building on existing guidance from literature, we present a conceptual framework and propose minimum criteria for combining data for an ECA under different motivations.

Results: The proposed framework includes (1) Developing a detailed data-agnostic study design with a priori specification of the research question, (2) Assessing feasibility of available data sources, (3) Identifying the optimal data combination approach, and (4) Assessing fit-for-purpose of the combined dataset. Feasibility assessment, including methodological (data reliability, relevance, harmonizability) and practical (ethical, regulatory, governance) considerations, informs whether combining data is required or feasible to meet study objectives. Motivations for combining data may include expanding study scope (e.g. to include relevant patient subgroups/disease subtypes) as well as improving internal validity (e.g. to increase longitudinal follow up and/or look-back period), external validity, or precision of estimates (by increasing sample size). The decision to pool, and the optimal approach for pooling disparate data as ECAs depends on the extent to which candidate data sources fulfill use case-specific methodological and practical considerations. Lastly, it is crucial to use the combined dataset’s metadata to assess its fitness for addressing the specified research question. Validation studies and sensitivity analyses that test pooling assumptions may be necessary for bias mitigation.

Conclusions: Combining datasets to generate ECAs can provide a viable solution for addressing limitations to obtaining fit-for-purpose data required to contextualize SATs. The proposed framework offers guidance for different use cases, including situations in which complementary data sources are desired. This approach advances the use of integrated real-world and clinical trial data to support robust evidence generation for clinical development programs.