PDUFA VII Negative Controls

Funder: United States Food and Drug Administration (FDA)

PI: Sengwee Darren Toh

Year Funded: 2008

Project Summary

Real-world evidence (RWE) generated from real-world data (RWD) is increasingly being utilized to address scientific and regulatory questions at the U.S Food and Drug Administration (FDA), including both product safety and effectiveness. Evaluating RWE for regulatory use, however, depends on a robust causal inference framework, and there is growing stakeholder interest to understand how methodological advances with negative controls, auxiliary variables not causally associated with the treatment or outcome of interest, can improve causal inference.

Project Details

With Real World Data (RWD), confounding arising from non-randomized treatment choices remains a fundamental challenge for extracting valid evidence to help guide treatment and regulatory decisions. A powerful tool increasingly recognized to detect, quantify, and correct bias due to unmeasured confounding is the negative control. Negative controls are variables associated with the unmeasured confounders (i.e., share the same potential source of bias) for an exposure-outcome relationship of interest but that are not causally related to the treatment or outcome variables of interest. Typically, the identification of appropriate negative controls is labor intensive and requires extensive background knowledge and subject matter expertise. Further, studies that incorporate negative controls have relied on the assumption that the identified variables that serve as negative controls are appropriate, relying on the subject matter expert determination, as no validation test existed.

A disconnected negative control is a negative control that is not causally related to the treatment nor to the outcome (while standard negative controls could be causally related to either the treatment or outcome). An automated approach to find disconnected negative controls involves the “Data-driven Automated Negative Control Estimation (DANCE)” algorithm.

The PDUFA VII Negative Controls project will extend, test, and adapt the DANCE algorithm to large-scale healthcare data that reflects the Sentinel data environments. A parallel companion IC PDUFA VII project will coordinate the development of this negative control tool, and a separate companion project will convert this tool into the Sentinel System’s portfolio of tools.