Improving the Efficiency of Outcome Validation in the Sentinel System

May 17, 2018 - 9:00 am

1201 Pennsylvania Ave, NW Suite 500

Washington, DC 20004

Description

The Sentinel System, authorized in 2007 by The Food and Drug Administration Amendments Act (FDAAA), is an active and fully functioning post market surveillance system that can rapidly scale distributed analyses on data collected by a diverse range of Sentinel Data Partners. In close partnership with key stakeholders, FDA has accomplished numerous milestones designing, building, and using Sentinel’s data infrastructure to inform regulatory decisions. A key component of Sentinel, the Active Post-Market Risk Identification System (ARIA), represents a set of querying tools combined with electronic health care data in the Sentinel common data model to conduct safety assessments. FDA is routinely using ARIA to inform a variety of regulatory actions including label changes, Advisory Committee deliberations, and other important safety assessment decisions.

By law, before using ARIA, the FDA must first determine whether the data and methods under ARIA are “sufficient” to answer regulatory questions of interest. The FDA defines sufficient as the availability of adequate data (e.g. the drug or biologic of interest, comparators, confounders, and covariates) and appropriate tools to provide a satisfactory level of precision to answer questions. The FDA has determined ARIA to be sufficient to inform some regulatory actions, however, there are instances when the infrastructure is deemed insufficient. Preliminary agency analyses have identified outcome validation as a major contributing factor driving ARIA insufficiency.

Under a cooperative agreement with FDA, the Duke-Robert J. Margolis, MD, Center for Health Policy convened an expert workshop to consider potential opportunities to build a stronger foundation for ARIA sufficiency through improved outcome validation. Workshop discussion explored how innovative technologies such as natural language processing, machine learning, and electronic phenotyping could support more streamlined and automated validation processes. Input solicited from diverse stakeholders at the workshop will inform subsequent decisions between the FDA, Sentinel Data Partners, and Sentinel Operating Center to identify and develop high priority pilots for future implementation.