Abstract:
|
Clinical trials of experimental treatments require control arms. However, a randomized control arm may be difficult or impossible for reasons including ethical concerns about assigning a placebo and patient unwillingness to be randomized, possibly to a placebo. Historical control groups from one or a few previous clinical trials have often been used, but this approach introduces biases due to differences in baseline covariates, sites, and other factors. We minimize these problems by constructing a synthetic control arm (SCA) from Medidata's archive of >3000 trials with data rights for anonymized aggregated analyses. For a specified single-arm trial, we create an SCA containing patients from recent trials with similar eligibility criteria. We use several approaches to select patients for the SCA to match the patients in the trial, including all available patients and patient-level matching on key baseline covariates. The SCA can provide a superior alternative to using a single arm or historical controls from literature, where covariates cannot be matched.
|