Presented byDanh V. Nguyen, Ph.D.
September 12, 2013
11 – 12 PM
Sprague Hall Room 205
The self-controlled case series (SCCS) method is useful for estimating the relative incidence of acute events, such as adverse events during a specified risk window (e.g., two weeks) following an exposure, e.g., after vaccination. However, in practice, the true or “optimal” risk period length in the SCCS method is unknown and must be specified. To date, only two approaches are available to guide in the specification of the optimal risk window length, specifically a graphical approach and a testing approach based on a scan likelihood ratio test statistics. We characterize, for the first time, the bias of SCCS estimate of the relative incidence under model misspecification of the optimal risk period. Based on the form of this misspecification bias, we propose a more principled approach to estimate the optimal risk length and associated relative incidence of adverse events based on the functional form of the bias, which have a linear and nonlinear components. Efficacy and improvement of the proposed approach will be discussed. Standard inferential procedures will also discussed in this talk and the proposed method will be illustrated with simulation studies and applications to vaccine safety data.