Search

An improved statistical method for assessing the efficacy of intrauterine contraception is presented. This method, called the log-rank method, is based on a daily life table, which yields the best estimates of the cumulative net termination probabilities. It provides a sensitive chi-square test statistic to detect differences among these probabilities for two or more contraceptive devices. When three or more devices are compared, the chi-square statistic may be partitioned into additive components representing comparisons among groups of devices. The log-rank method also permits adjusting for differences among subgroups by controlling for factors such as parity or age. Although calculations may be performed on a desk calculator, a FORTRAN-IV computer program is made available to interested researchers.

It is now well established in case-control studies with individual matching that the statistical analysis must take the fact of individual matching into account. The problems of so doing have been thoroughly dealt with for the situation of the factor of interest being a two-level factor (all-or-none response) and with one or more matched controls per patient. These results are extended in this paper to the situation where the factor can assume multiple levels in studies where each patient is matched with a single control.