You can obtain any odds ratios of interest easily, while keeping continuous variables
continuous. For example, to estimate the odds ratio after increasing saturated fat
from its 0.1 to its 0.9 quantile you just ask for two predicted values and take the
difference in these. This handles nonlinearity quite easily. You have to go to a bit
more trouble to get the confidence limits. The S-Plus and R Design library
(http://hesweb1.med.virginia.edu/biostat/s/Design.html) handles this easily:
# Regression spline for fat, 4 knots, interacting with age (linear)
f <- lrm(death ~ rcs(fat,4)*age+sex)
nomogram(f) # draw nomogram depicting entire fitted equation
fat.quantiles <- quantile(fat, c(.1,.9))
contrast(f, llist(age=40,fat=fat.quantiles[2]),
llist(age=40,fat=fat.quantiles[1])
# Log-odds for 40 year old at 0.9 quantile minus
# log-odds for 40 year old at 0.1 quantile of fat consumption
# Fat effect depends on age setting since interaction allowed
# Doesn't depend on sex setting so let sex=default reference group
Frank Harrell
On 21 Apr 2002 13:51:18 -0700
[EMAIL PROTECTED] (wuzzy) wrote:
> > I also see this sort of thing done sometimes to make results more easily
> > interpretable (and applicable?) for doctors and other clinicians. Some of
> > them struggle with interpretation of coefficients for continuous
> > variables, and have a clear preference for categorical predictor
> > variables.
> >
>
>
> This is a good reason, for instance, dietary advice often involves
> dramatic reduction in things like saturated fat, so you would compare
> 90th percentile to 10th percentile, (say in a logistic regression for
> risk of heart disease) not caring about the Beta, of the full spectrum
> of saturated fat.
> An Odds Ratio of 1.5 looks better than an odds ratio of 1.00001
--
Frank E Harrell Jr Prof. of Biostatistics & Statistics
Div. of Biostatistics & Epidem. Dept. of Health Evaluation Sciences
U. Virginia School of Medicine http://hesweb1.med.virginia.edu/biostat
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
