A new trend has emerged over the last decade that has changed the way many clinical trials are conducted. Unlike placebo-controlled randomized control trials, single-arm trials forgo the use of a placebo or standard-of-care as a control and establish clinical benefit by demonstrating the effects of a new therapy or treatment via comparison with a synthetic control arm (SCA) derived from external sources. This alternative approach leverages real-world data from various sources or evaluations of historical trial data for the sake of comparison.
In the development of an SCA, researchers may find that the historical data involves homogenous or heterogenous patient populations to those enrolled in the new trial: that is, the population sample may have a similar demographic, clinical endpoint, or other similar variables; or it may be different, involving different demographics like age, sex, or race, or different clinical endpoints. While homogenous patient population data are easier to combine, when important heterogeneity exists, the historical data may need to be weighted using a Bayesian hierarchical model, which can help account for variability within and between data sources thus making it better suited to study certain trial designs.
Historical data is particularly useful when using a synthetic control arm, rather than a traditional two-arm trial. It enables sponsors to build a control arm based entirely on data already collected, thus reducing patient samples and salvaging resources. When done correctly, it also reduces the timelines of the trial by making the need for extensive enrollment unnecessary. All in all, it is a useful tool for rare disease trials and others with small sample sizes.
However, for some SCAs, such historical data might be combined with data from newly enrolled patients. But what is the ideal ratio for new to old patients when constructing an SCA? To learn more about synthetic control arms and how Bayesian dynamic borrowing of external patient-level control data can be used to optimize previously collected data and minimize the number of new patients enrolled into the control group, including an illustrative case study, download our recent publication:
Many thanks to Paul Arora for his help with this post.
Read more about Bayesian dynamic borrowing:
