Background: External control arms (ECAs) provide an alternative source of comparative evidence when randomised controlled trials (RCTs) are infeasible, such as in rare diseases or small subpopulations. Determining appropriate sample sizes is challenging, as small ECAs contend with statistical power, balancing confounders, amongst other factors all vital for generating robust evidence for regulatory or health technology assessments (HTA). Power simulations require assumptions on the key confounders and patient characteristics, their correlation structure, and extent of heterogeneity amongst patients. These all introduce uncertainty into the estimation of sample sizes required given a desired power level. A better understanding of how factors affect ECA study design is crucial and could inform the design of prospective studies or identifying additional patient sources.
Objectives: This study will outline approaches for estimating sample size requirements for ECAs in small populations for survival outcomes under varying scenarios covering different patient synthesis methods, correlation structures and adjustment methods. Additionally, we propose a way of estimating the additional patients needed to achieve a pre-specified power and covariate balance. The calculations’ outputs can inform decisions regarding feasibility and data sourcing for an ECA study.
Methods: The simulation study will focus on survival outcomes and assume proportional-hazards. Synthetic patient data will be generated under several scenarios for the joint distributions of covariates in the target cohorts. Evaluation of how different correlation structures affect sample size calculations will be included.
ECAs will be adjusted using methods inverse probability of treatment weighting (IPTW) to account for confounding. Relevant metrics will include achieving covariate balance with sufficient effective sample size, power, and bias. Additionally, we will discuss approaches, including Bayesian borrowing, for leveraging additional data sources to achieve desired power.
Results: We will present power simulation results showing how sample size requirements vary based on covariate imbalance scenarios.
Conclusions: This study supports feasibility assessments for ECAs by proposing an initial framework guiding sample size requirements and associated uncertainties in study characteristics. Beyond informing needs for additional patients for ECAs, outputs can also potentially inform prospective study planning. Understanding ECA study characteristics is essential for sponsors and decision-makers planning evidence generation strategies for rare diseases and small subpopulations.
