116: A Multi-layer AI Algorithm that Transcribes SDTM Data into Full-length CSR

Statement of Purpose: We have developed an AI platform that directly transcribes the SDTM datasets of a phase I/II study into a full-length clinical study report (CSR). The underlying algorithm addresses the challenging nature of clinical study data where every single observation point specified per protocol exhibits a distribution across subject population. The platform has been implemented for hundreds of NDA/BLA submissions and over 1,500 clinical trials.

Description of Methods & Materials: The underlying AI algorithm mimics the subject matter expert’s (SME) thinking process, consisting of multiple layers of data pattern recognition. The algorithm addresses the challenging nature of clinical study data where every single observation point specified per protocol exhibits a distribution across subject population. The first basic layer of the algorithm identifies, from the inter-subject distribution of the observation time, the specific set of per-protocol nominal time points of each observation variable. The second layer identifies the time courses and value-time shape of each primary clinical observation and determines the start and end points of each observation-time shape. The third layer determines the start and end time of each protocol-specified study period based on the sets of start/end times from all observation-time shapes. The fourth layer constructs the study structure and scheme (e.g., parallel, sequential, crossover, multi-cohort, etc) by assigning subjects to relevant arms and examining the patterns of treatment regimen and available observation data in each arm and period. The fifth layer identifies outlier subjects or data points within the study structure determined in the previous layers. At this layer, the entire study design and study conduct have been deciphered by the AI algorithm. In the sixth layer, the proper clinical and statistical methods are applied to generate all routine and sophisticated analyses and produce the full-length CSR with supporting appendix.

Data & Results: The algorithm is implemented to develop the AI Analyst, which is a full autonomous platform that reads data from the CDER Electronic Data Room (EDR) and generates study analysis reports following the standard review templates of BA/BE and biosimilarity studies (BMER). The AI platform has been tested and utilized for the analyses of hundreds of NDA/BLA submissions and over 1,500 clinical trials.

Interpretation, Conclusion or Significance: The Office of Clinical Pharmacology (OCP) recently initiated the RealTime Analysis Depot (RAD) project that aims to efficiently apply the AI platform to complete all clinical phase I/II study review reports within a few weeks after the submission date, covering NME, 505b2 and 351K submissions.

Disclosures: