Data Driven Strategies for Improved Site Activation | Page 13

When forecasting enrollment, it is important to keep in mind that the process is not linear, with slower overall enrollment during start-up when few
sites are activated and peak enrollment once all sites are active, some months later.
During forecasting, which initially occurs very early in the process, the specific sites may not yet be identified. Educated estimates regarding the
mixture of sites assist with appropriate modeling, and a statistical distribution of the time from the final protocol approval to site activation can
be estimated. As a starting point, we have found that using a Gaussian or Gamma distribution closely aligns with historic information about site
activations. For example, in oncology studies, the Gaussian (normal) distribution closely models actual site activations, with an average start-up
length of 206 days that can vary by a standard deviation of 64 days. The Gamma distribution, which is another logical choice to model wait times,
does a better job of accounting for a “long tail” of site activations, which can help to model unexpected events that are not typically part of the initial
plan, such as the addition of back-up sites, replacement sites, IRB rejection of t