Bayesian Analysis of Time-Series Data Under Case-Crossover Designs

• Activity Number: 575
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Shi Li*+ and Bhramar Mukherjee
• Companies: University of Michigan and University of Michigan
• Keywords: case-crossover ; Markov Chain Monte Carlo ; estimating equation ; posterior inference ; proportional hazard ; matched case-control

Abstract

We compared case-crossover designs with matched case-control study and time-series analysis under the same time frame and underlying risk model. We discussed the similarity and difference between case-crossover design and matched case-control study, and proposed a full likelihood approach to avoid overlap bias for general referent window selections. We extended the frequentist equivalence results in terms of estimating equations between case-crossover design and time-series analysis shown in Lu (2007) to a full likelihood framework. We established Bayesian equivalence results of the two analysis in terms of posterior distribution of the log odds ratio parameter. Case-crossover designs, regardless of window selection schemes, can be readily fitted using a random effects model. We proposed a Dirichlet process prior on these random intercept parameters under a full likelihood model. The proposed methods were utilized to the Detroit Asthma Morbidity, Air Quality and Traffic study.