As Dew mentioned, in trials that have multiple primary endpoints, the alpha (i.e. p-value) typically gets split amongst them. If 0.05 was used for each individual endpoint, the overall Type I error/false positive rate for the study would be >0.05.
…why would the confidence interval for the HR be over 1 w p<.05
The confidence interval reported for Imfinzi-monotherapy OS is not the usual 95% CI, but rather a 97.54% CI, so the p-value would’ve had to be <0.0246 to be deemed statsig. The non-standard CI is due to the non-hierarchical 4-way splitting of the alpha in this trial.
p.s. If hierarchical endpoints had been used in MYSTIC, the PFS endpoints would've had to have precedence over the OS endpoints, else the PFS analysis would've been indeterminate while waiting for the OS data to mature, which wouldn't make sense from a business standpoint. (This could've made a good quiz question.)
News 
Market Data 
Discover 
