Abstract
One of objectives in epidemiologic studies is to detect the amount of change caused by a specific risk factor. Risk ratio is one of the most useful measurements in epidemiology. When we perform the inference for this measurement with rare events, the standard approach based on the normal approximation may fail, in particular when there are no disease cases observed. In this paper, we discuss and evaluate several existing methods for constructing a confidence interval of risk ratio through simulation when the disease of interest is a rare event. The results in this paper provide guidance with how to construct interval estimates for risk difference and risk ratio when there are no disease cases observed.