Keywords: Composite likelihood, entropy balancing, RWD, uncontrolled, RCT
Randomized control trials (RCTs) remain the primary evidentiary standard for demonstrating efficacy and safety of an investigational treatment. However, it is sometimes unfeasible because of their size, duration, cost, or ethical concerns in many trials including pediatric and rare disease populations. One solution to potentially reducing sample size requirements is to leverage data from pooled external controls or real world data (RWD). Usually, propensity score matching methods are used to create defined groups of patients that are controlled for confounding based on a set of measured covariates. However, the balance check on all confounding covariates and tweaks to the propensity score model are necessary to achieve the joint balance across all covariates. In some cases, this balancing approach may not lead to desired results for which entropy balancing does automatically and outperforms the propensity score matching in terms of assuring the balance of covariates in the pre-specified moments. For this reason, we will use this entropy balance approach to create a composite likelihood of the data from the treated and the weighted control observations. We will discuss how this method can be applied in the context of uncontrolled and asymmetrically allocated randomized controlled trials.