Cox regression

["Chaffin, Mark J. (HSC)" <Mark-Chaffin@xxxxxxxxx>]

Bill,

How to estimate effect sizes in Cox regression? I might try one of these
two approaches.   Approach 1:  Look at (or calculate) r-square or effect
size values and confidence intervals.  You can calculate these using the
model betas and beta confidence intervals which most programs output.
The formulas for calculating are in any multivariate stats book.  I
would double check with a stats expert, but I suspect the formulas used
for regular regression weights would yield decent estimates with Cox
regression as well.  Lazy man's approach:  If you are interested in a
grouping variable (or don't mind carving up a continuous variable into
groups purely for graphing purposes), just look at the lines on the
survival plot, paying attention to any y-axis scaling problems which may
make effects look larger or smaller than they actually are.  

MC