Seeking Value: GSK’s Drug-Safety AI Strategy Connects Data With Outcomes

British pharmaceutical company GSK is seeking to improve its drug-safety reporting processes through the use of artificial intelligence (AI). Currently, GSK's drug-safety workflows involve multiple steps, including receiving safety reports, assessing their priority, and directing them to the appropriate specialists. This process can take hours and involves manual reentry of information, leading to inefficiencies.

GSK and Cognizant, a global consulting firm, have been working together to automate these workflows. The goal is to preserve information as it moves through various activities, such as processing, evaluation, regulatory reporting, and inspections, while reducing the manual steps required for preparation. GSK's existing safety systems will remain the source of truth for much of the company's information, but the new approach aims to make the workflows more efficient without replacing the underlying systems.

GSK uses a combination of AI-based and conventional automation to classify incoming safety reports, establish priorities, and direct information to the appropriate specialists. Structuring data as it arrives helps preserve the context needed for subsequent checks and reviews. GSK and Cognizant envision a coordinated system of specialized AI agents to support various tasks across the drug-safety process, including intake, coding, checking information, and preparing regulatory information.

According to Mohammad Haque, Cognizant's global head of intelligent enterprise platforms and large deals, the proposed model could eventually include approximately 40 specialized agents, with some tasks running in parallel to reduce waiting times between team activities. While human judgment remains crucial for drug-safety decisions, AI may help prepare information that requires complex assessment.

GSK and Cognizant are also reevaluating how they measure the value of their technology investments. Instead of solely relying on conventional automation measures, such as the percentage of activities handled by software, they aim to measure outcomes, such as reducing case processing time and the effort required to prepare aggregate safety reports.

They target a 50% reduction in case processing time, a 40% reduction in the time between identifying a potential safety signal and making a decision, and greater than 90% consistency in quality-control activities.