WS-0003 - Increasing The Credibility Of Real World Evidence: Identifying And Reducing Bias Using Negative Controls

Alexander Breskin, Matthew Maciejewski, Leah McGrath, Vera Ehrenstein, Brian Bradbury, Gianmario Candore

Live Session with Q & A
October 20, 2020
10:00 am EDT
Confirm Attendance

Background: Common approaches for real world studies of comparative effectiveness and safety include covariate adjustment (e.g., with propensity score weighting, regression, or matching) and instrumental variable methods. Recently, negative control methods have been increasingly used as an approach to both detect and reduce bias when estimating causal effects. Negative control variables share the same unobserved confounding structure as the exposure-outcome association of interest. Unlike the exposure and outcome of interest, however, negative exposures do not have a causal effect on the outcome of interest, and negative control outcomes are not causally affected by the exposure of interest. Notably, the causal conditions underlying negative control methods are unique and different from those needed for covariate adjustment or instrumental variables. Thus, negative controls may also be useful in situations where other methods are infeasible.

Objectives: 1. To provide an introduction to the theory and concepts underlying negative control methods. 2. To demonstrate approaches to identifying negative control variables. 3. To describe analytical methods to detect and mitigate uncontrolled confounding using negative control variables. 4. To gain perspectives from industry and regulatory leaders on the potential for negative control methods to complement real world data analyses. 5. To benefit researchers involved in real world data analysis for comparative effectiveness and safety studies.

Description: The symposium will begin with Dr. Maciejewski introducing the theory and concepts underlying negative control methods. This will be followed with a case study by Dr. McGrath on selecting negative control outcomes, including the identification of novel negative controls using real world data. Next, Dr. Breskin will describe analytic methods for detecting uncontrolled confounding and calibrating estimates with negative controls, with an emphasis on the conditions required for their validity. Drs. Bradbury and Candore will then present industry and regulatory perspectives, respectively, on how negative control approaches can be used to complement and support comparative effectiveness and safety research. The symposium will conclude with reflections and an audience discussion guided by Dr. Ehrenstein regarding the potential uses of negative control methods in real world data analysis. Each presenter will be allotted 15 minutes.