Use of Fixed-Effects Models to Analyze Self-Controlled Case Series Data in Vaccine Safety Studies

• Activity Number: 659
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Stanley Xu*+ and Chan Zeng and Sophia Newcomer and Jennifer Nelson and Jason Glanz
• Companies: Kaiser Permanente Colorado and Kaiser Permanente Colorado and Kaiser Permanente Colorado and Group Health Research Institute and Kaiser Permanente Colorado
• Keywords: self-controlled case series ; adverse events after immunization ; fixed effects model ; conditional Poisson model ; longitudinal data

Abstract

Vaccine safety data from self-controlled case series (SCCS) design are often analyzed using conditional Poisson (CP) models. But, CP models maximize a partial conditional likelihood and can not be used for likelihood-based hypothesis tests. In addition, it is not straightforward to fit complex models such as a semi-parametric model for SCCS data. In this paper, we demonstrate theoretically and empirically that fixed effects (FE) models generate the same estimates and their standard errors for time-varying variables as CP models. There are a few advantages to use FE models to analyze SCCS data: 1) likelihood-based statistical tests may be employed; 2) statistical analysts may find fitting FE models more intuitive than fitting CP models because they are already familiar with procedures for fitting Poisson models; 3) fitting complex semi-parametric models is achieved by simply treating event days as indicator variables in preparing summary data for person time and number of events and later including them in the FE model. We also show that another traditional longitudinal data analysis method, generalized estimating equations (GEE), produce different results than CP or FE models.