Visit our website: http://bit.ly/2GtXaiw
Quality Transformation Initiative
Yael Gozin, Global Clinical Submissions Quality, Senior Director, Pfizer
Abstract:
The size and complexity of regulatory submissions are consistently increasing. However, though data verification is the key to every quality review, the process associated with the quality checks of health authority reporting has not changed for decades. The process is manual, laborious, time-consuming, error-prone, and has been a challenging task for humans and machines.
By utilizing AI capabilities such as natural language processing (NLP) and machine learning, Pfizer, in partnership with BeaconCure, developed a new general method for structured data verification, which mimics the way a human reviewer verifies the table’s accuracy.
Furthermore, this technology allows remarkable accuracy as demonstrated in the proof of concept wet-test results: F1 = 0.98, precision = 0.99, and recall = 0.97.
“Verify” Version 1.0 was released in April 2019 for the pre-production pilot supporting .rpt format. Following a change in the regulatory requirement and an internal process changes, Pfizer moved away from the .rpt source table format to the .html format. To support the .html format requirements, “Verify” Version 1.2 was released in October, and the results analysis process is ongoing.
Add comment