Overall Survival (OS) is the gold standard end point for oncology trials. Ultimately we need to show a drug works in term of extending life (OS) (or in slightly less degree, improving quality of life). Since it’s often not practical to demonstrate OS benefit directly in trials due to lengthy duration and high cost, we have many widely used surrogate end points (e.g. TTP, PFS). These surrogates work well in some types of cancer but not so well for others, in term of predicting OS benefit. The interest of our discussion will focus on the following scenario: The primary end point was pre-specified using a surrogate (e.g. TTP) and barely failed to show statistical significance (say p=0.06). However, once final survival data were available, analysis revealed that OS was highly significant in the ITT population. This OS analysis was not designed as the primary end point (maybe a secondary end