Causal Inference for Electronic Medical Records Data: Application to Prostate Cancer

• Activity Number: 631
• Type: Topic Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Rebecca Yates Coley* and Scott L. Zeger
• Companies: The Johns Hopkins University and The Johns Hopkins University
• Keywords: causal inference ; latent variables ; Bayesian analysis

Abstract

It is difficult to evaluate the effectiveness of prostate cancer treatments because, first, treatments are not randomly assigned, and, second, treatment effectiveness varies depending on the underlying and often incompletely observed characteristics of an individual's cancer. To address these limitations, we have developed a Bayesian hierarchical latent variable model for estimating the average effectiveness of treatment within subgroups defined by possibly misclassified disease state. The model is applied to data from the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, a randomized trial of the effect of screening on cancer-related mortality. Despite screening randomization, there were high rates of contamination in the control group and treatment decisions upon screening were not randomized. Results of this analysis are presented and implementation with EMDR are discussed.