Confidence Interval Estimation for the Ratio of Simple and Standardized Rates in Cohort Studies

Abstract

Computer simulation has been used to compare four methods for calculating confidence intervals for simple rate ratios estimated from cohort studies. The method proposed by Cornfield (1956. In Proceedings of the Third Berkeley Symposium on Mathematical Statistics and Probability. Vol. IV, 135-148) for interval estimation for the odds ratio in a 2 X 2 table (adapted to the ratio of two Poisson variables) gives reasonable and conservative results, and is recommended in view of its relative computational simplicity. Three methods are compared for interval estimation for the ratio of standardized rates, such as the standardized morbidity or mortality ratio, when the data are stratified by some confounding factor. The adapted Cornfield method again is the recommended procedure. An examination of the use of a summary chi square test with and without continuity correction indicates that the test size closest to the nominal is given by excluding such a correction when the test statistic is computed.