On Choosing a Measure of Explained Variation for Cox Models

There have been several attempts to extend the notion of explained variation to models for analyzing lifetime data, Cox models in particular . Most of the authors have concentrated on overcoming the difficulty of using the censored observations in the calculations, but did not pay enough attention to other, equally important issues . We present a list of properties that a measure of explained variation for the Cox model should have, and discuss these properties with reference to the measures by . Kent and O'Quigley and by Korn and Simon . We show that only the approach by Kent and O'Quigley gives good results . We also stress the need for the use of such measures in survival studies .