Clinical safety data are routinely evaluated using between group p values for every reported adverse event (AE), with multiple testing procedure applied to the p values to adjust for multiplicity. However, the p value generated for each AE is often based on comparing only the AE incidence rate between two randomized groups, regardless of AE severity. To enhance the evaluation of drug safety, for each AE we propose to use AE occurrence and severity as co-primary endpoints and to perform a statistical test of the composite null hypothesis that the incidence rate and severity are equivalent between groups. The p value of the test of the composite null hypothesis is obtained by combining the p values of the Fisher’s exact test for AE incidence and the test for AE severity. The test for AE severity is based on a biased sampling model, which is an extension of the work by Gilbert et al. (2003, Biometrics 59, 531-541) to ordinal response. We conduct simulation studies to investigate the power and type I error rate of the proposed tests of the composite null hypothesis and compare them with the test of equality of AE incidence rate. The simulation results show that, in general, the proposed method performs as well or outperforms the test of equality of AE incidence rate in detecting a safety signal.