Mediation and measurement error

Kimberley Goldsmith, Ph.D.
King's College London

ABSTRACT:
Clinical trials require many resources and should be used to obtain as much information about treatments as possible.  Mediation analysis can be used to test theoretical models of pathways through which treatments work.  This can lead to refinements of models and treatments, and to a greater understanding of the reasons for negative trials.  Interest in mediation analysis has increased recently, and this has led to methodological advances.  The recent literature has generally focused on the issue of unmeasured confounding, however, measurement error in the mediator could be another cause of bias in mediation effect estimates.  This talk will examine some approaches to address measurement error in mediation analysis, and apply these models to data from PACT (Preschool Autism Communication Trial), and to simulated data patterned on the PACE trial (adaptive Pacing, graded Activity and Cognitive behaviour therapy; a randomised Evaluation).