Evaluating Interval Estimates for Comparing Two Proportions with Rare Events

Abstract

Epidemiologic studies frequently try to estimate the impact of a specific risk factor. The risk difference and the risk ratio are generally useful measurements for this purpose. When using such measurements for rare events, the standard approaches based on the normal approximation may fail, in particular when no events are observed. In this paper, we discuss and evaluate several existing methods to construct confidence intervals around risk differences and risk ratios using Monte-Carlo simulations when the disease of interest is rare. The results in this paper provide guidance how to construct interval estimates of the risk differences and the risk ratios when no events are detected.