Abstract: Logrank test and Cox model were usually applied to compare multiple survival rates. A mistake maybe takes when data are tied or/and all data are censor. How to analyze these data is a problem. Based on a clinical trial example, Discrete, Breslow, Efron and Exact methods were discussed with tied data. Discrete method is best when time is discrete. Otherwise, Exact is best. In addition, data analysis with heavy censoring was explored. When one group is total censor Wald test is unreliable. However, LR test and Score test or logrank test is acceptable. Chi-square test is another choice to analyze survival rates with heavy censoring.

Key words: Logrank test, Cox model, Ties, Heavy censoring