Can we really use retrospective subset analyses and surveillance, epidemiology, and end results data to drive clinical practice?