Using Double Negative Controls to Adjust for Healthy User Bias in a Recombinant Zoster Vaccine Safety Study

Abstract

Bias due to unmeasured confounding haunts the field of pharmacosurveillance. We applied a novel causal inference method utilizing negative control variables to correct for unmeasured confounding bias in a safety study of Recombinant Zoster Vaccine. This ‘double negative control’ method uses a confounding bridge to estimate the magnitude and scale of bias, relying on weaker assumptions than common causal inference methods. A principled approach was developed to determine negative control outcome and exposure pairs capable of correcting for healthy user bias. Relative risk estimates and confidence intervals calculated using double negative control methodology were compared to those calculated using a flexible propensity score regression method. The double negative control methodology was feasible to implement in an electronic health record setting, adjusted estimates in the expected direction (e.g., attenuated protective effects that were potentially due to healthy user bias), and more accurately quantified uncertainty due to unmeasured confounding.