Bayesian Inference for Causal Mechanisms with Application to a Randomized Study for Postoperative Pain Control

• Activity Number: 115
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Alessandra Mattei*+ and Michela Baccini and Fabrizia Mealli
• Companies: University of Florence and University of Florence and University of Florence
• Keywords: Bayesian inference ; Causal mechanisms ; Mediation analysis ; Potential outcomes ; Principal stratification ; Randomized Experiments

Abstract

Using a prospective, randomized, double-blind study to evaluate the effect of preoperative oral morphine sulphate on postoperative pain relief as a motivating example, we investigate the causal mechanisms underlying the positive overall effect of a better postoperative pain control. Specifically, we investigate to what extent this effect is mediated by postoperative opioid administration of IntraVenous Patient Controlled Analgesia.

Using the Bayesian approach for inference, we estimate natural and controlled direct effect under the Baron-Kenny framework, as well as more advanced associative and dissociative principal strata effects. These analyses are conducted under different sets of identifying assumptions.

Analyses conducted under the principal stratification framework allow to assess the heterogeneity of effects across sub-groups of patients characterized by a different use of postoperative self-administrated analgesia, and thus different pain perception. We were also able to investigate effect heterogeneity with respect to relevant pretreatment patient characteristics, such as gender.