The rule of events per independent variable (EPV) has provided a useful indication for the performance of Cox model and sample size estimation in survival analysis. However, this indicator only reflects the number of events and variables required, but not the characteristic of variables included in a regression model. The rule of EPV says that at least 10 to 20 EPV are required for Cox model for unbiased estimation of the ‘true’ effect of a variable on the survival outcome. However, this is not easily achieved for most studies in rare diseases. This study investigates the performance of the Cox coefficient estimate of a treatment variable that exhibits strong association to survival outcomes at low EPV condition, manifested by statistically significant outcome with pvalue less than 0.01 in both log-rank test and multivariable Cox model at low EPV condition for both overall survival (EPV = 6.5) and disease free survival (EPV = 6.75). The Cox estimate of the treatment variable is associated with highest level of bias and disparity below 4 EPV only. At 5 to 6 EPV, its performance starts to acquire stability and consistency. Above 6 EPV, increasing the number of events is less likely to improve its overall performance. Therefore, strong variable association reduces the EPV required, rendering trustworthy Cox coefficient estimate possible at low EPV (

survival analysis, Cox model, events per independent variable, proportional hazards model, acute myeloid leukemia.