(P-565) Recent Use of External Controls in Clinical Trials: A Pragmatic Literature Review

Background: The use of external control arms (ECAs) to augment clinical trial data has been receiving increasing interest from different stakeholders, not the least from regulatory authorities. Benefits of using ECAs include allocating more or all patients to the active treatment, which overcomes ethical constraints and speeds up clinical development.

Objectives: To assess the recent literature about the use of external controls in clinical trials.

Methods: Medline (PubMed) was searched from 01-Jan-2018 through 05-Feb-2021 for clinical trials of investigational pharmacological therapies that reported the use of external controls. Ten exact search phrases for external controls were used. Retrieved citations were screened for relevant studies. Data were abstracted on the indication, trial design, and type and build of ECA, among others.

Results: Seventy-seven of 157 identified citations were publications of trials of pharmacological interventions that reported the use of external controls. The majority were early phase studies in oncology (n=52, 68%) and rare diseases (n=9, 12%). In 48 of the 77 studies (62%), results on efficacy or safety in the trial were compared to “historical controls” that consisted of merely a historic benchmark. In contrast, 26 studies (34%) used patient-level data to build an ECA, of which 21 had full text freely accessible for further assessment. Among these 21 studies, 15 (71%) were single-arm trials. The ECA was built from historic clinical trial data in 8 studies (38%) and from real-world data in 12 studies (57%); the ECA build could not be determined for 1 study. Eleven studies (52%) applied trial eligibility criteria to the external controls without balancing covariate distributions between the trial patients and external controls. Seven studies (33%) balanced covariates by applying methods such as group-matching (1 study); matching on the propensity score (1 study); inverse probability treatment weighting using the propensity score (3 studies); stratification on the propensity score (1 study); or multiple methods based on propensity score (1 study). The method by which the ECA was built was not described in 3 studies (14%).

Conclusions: The use of patient-level data to generate ECAs is still in an early stage. The overwhelming majority of trials in this sample relied solely on a historical benchmark as the ECA. Although most of these studies were small phase II trials, this approach may be vulnerable to bias. Of the few trials that used patient-level data to build ECAs, most applied trial eligibility criteria to the external controls without covariate balancing, which for some studies may be explained by small sample size.