Massive amounts of clinical data are generated every day. For many conditions large databases exist containing the results of observational studies, registries, case studies, and clinical trials. At the same time the statistical methodology for incorporating historical data in prospective clinical trials has also increased, now including methods such as power priors, commensurate priors, hierarchical modeling, and clustering methods.

These methods may provide valuable information in a clinical trial, significantly reducing the uncertainty in point estimates produced by the trial as well as leading to better decisions, often in the form of increased power for a trial. However, when there are systematic differences between the historical data and the data within the trial, any borrowing of information may bias the results in the current trial. This is scientifically problematic, and creates regulatory issues primarily due to the potential inflation of type I error.

We will review a number of current methods for incorporating historical information into a clinical trial. Depending on the historical data available, different methods may be more applicable. For example, a situation with many large clinical trials as historical data should be treated differently than a situation with only one observational study. We will provide suggestions on matching the historical data format to the historical data method, both in terms of statistical and regulatory challenges.