[R] (no subject)

Terry Therneau therneau at mayo.edu
Wed Mar 31 16:26:23 CEST 2010

The answer depends on what your goals are.

If you are simply trying to put all the hazard ratios on a similar
scale, then I often use the HR comparing risk of a subject at the 25th
percentile to one at the 75th percentile.  This has already been
suggested by David W.

But this is not perfect.  The key issue is that a definition of
"important" is slippery.  Is a risk factor that increases your death
rate 3-fold but only appears in 10% of the subjects more or less
important than one that confers a 50% increase of risk but in 1/2 the
subjects?  (Smoking confers a major increase in the risk of a rare
disease -lung cancer- and a more modest increase in the risk of heart
disease.  The second generates more excess deaths than the first.  Which
to highlight is an old debate in public health.)

Frank's answer went more to the issue of statistical importance, which
is a yet another dimension.

Terry T.

More information about the R-help mailing list