Statistical Methods for the Analysis of Case Series Data
Date
2014-08-14
Authors
Ling, Shen
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This paper compares the three methods -conditional Poisson method, unconditional Poisson method and self-controlled case series (SCCS) method, based on the retrospective cohort study with full cohort and case series sampling designs, with particular emphasis on their assumptions, power, MSE, relative efficiency, and handling of confounding. The performance of the three methods is contrasted in a study investigating the causality between vaccination and rare adverse events seizures. And we extend these methods to random effect model.
Description
The thesis is motivated by a study investigating the causality between
vaccination and seizures, or more general adverse events. The
assumption is that seizure events occur according to a homogeneous
Poisson process, and that the rate may change after vaccination. The
observation period consists of a baseline or low risk interval, and a
high risk interval after vaccination. Three methods for estimating
the relative incidence rate are considered - an unconditional method
with allows for different rates before and after vaccination, a
conditional method which conditions on the total number of events in
the two intervals, and the self-controlled case series (SCCS) method,
which conditions on a subject having at least one event. The
performance of the methods, with particular emphasis on their MSE,
power, and efficiency, is considered, under both full cohort and case
series sampling designs.