Much has been discussed about artificial intelligence in our industry. Coming to the core issues in clinical trial planning and execution, such as protocol design optimization, investigator site selection and the business process to put select investigator sites to work [namely site activation], we have yet to see hardly anything in the public space. In fact, our industry has been chronically challenged by these issues, resulting in unnecessarily delayed timelines and contributing to the rising costs of bringing life-saving medicines to patients in need. In an industry-wide analysis on site activation we conducted last year, Site Effectiveness Index (SEI), a key metric measuring site activation, had not changed at all in performance terms since we published a similar analysis 10 years ago.
Investigator site selection is different. We are pleased to see that the life sciences industry recognizes tremendous opportunities to improve the method of selecting investigator sites, which can help us to recruit the right patients in the right time frame. We applaud the efforts of various site selection service organizations who are attempting to tackle this challenging task of helping clinical development organizations to improve investigator site performance.
Despite these and other efforts, however, poor investigator site performance persists. When a trial runs into trouble, what people see is that the sites are not enrolling patients as planned. Often where a site selection organization has been involved, people blame that organization, believing that the investigator sites provided are “not good enough.” Often this is true, but nearly always what is lacking is the ability to do an integrated analysis of the protocol, the sites, and site activation, to be able to pinpoint the issues.
Phesi has engineered and patented a different and fully integrated approach. Through our holistic and data led approach, we do count the number of trials an investigator has conducted, but our AI analytics platform goes much beyond this one factor. We understand the complexity around the performance of an investigator site in a particular trial context, plus a multitude of key variables determining site’s ability to deliver adequate enrollment results within a defined time frame.
One set of such variables is the clinical experience profiles of a pool of investigators for a particular disease indication, or often a clinical trial with a specific design. A machine learning mechanism, powered by a set of sophisticated algorithms/statistical models, tirelessly sucks primary data from over 70,000 sources around the world in real time. The primary data are automatically analyzed, structured, and depicted to be visible to human eyes as shown by the following “heat map”:
To help us understand the dynamically updated “heat map,” the rows are individual investigators in code and the columns are clinical experiences in specific medical terms. From red to green, the color represents the increasing level of clinical experiences for a particular investigator in a particular medical area. Two representative profiles from UK and France, with their names intentionally altered, are shown in the two pie charts to the right.
While it is impossible to provide a comprehensive description of this heat map in a short blog, this artificial intelligence capability in Phesi investigator site selection platform does offer us unprecedented insight into the relevant variables that are important for a site’s performance potential in a particular clinical trial.
In this one example above, of a neurological indication, the map helps us to understand the most essential clinical experiences that are common to all site candidates. It also offers insight in terms of how these investigators are expanding their clinical experience to other areas and how those experiences are related to the core attributes of the indication under study.
This power of machine learning capabilities allows us to locate investigators with similar profiles globally, dynamically, reliably, and consistently. A pool of investigator sites recommended by Phesi are not only disease indication specific, but also quite frequently are specific to a particular protocol design.
Why not embrace AI within clinical trial design and execution? Why waste time and resources and risk trial failure, when patients are eagerly waiting for new and better therapies?