MSC Stats Project Presentation "Interval Estimation for Standardized Event Ratios" by Dan Ni Yang

DAN NI (JENNY) YANG

ABSTRACT:

Standardized event ratios (SERs) are used to compare the observed to the expected numbers of events for a population, where the expected number of events is often calculated using event rates of a reference/standard population. Often it is of interest to explore stratum specific SERs in situations where the overall population involves multiple strata. Seven methods of interval estimation for standardized event ratios, including 5 traditional methods, the percentile bootstrap method, and an adaptation of an asymptotic Normal method, are compared. These methods are illustrated using two datasets, where the events of interest are kidney transplant and mortality, respectively. The traditional methods tend to have nonsymmetric confidence intervals when compared to the percentile bootstrap and asymptotic Normal approaches. Typically, the percentile bootstrap approach produces narrower confidence intervals. However, to identify a confidence interval method that provides the best coverage, further work is needed such as a comprehensive simulation study.