Function: Compose an S Function to Compute X beta from a Fit

Description

Function is a class of functions for creating other S functions. Function.rms is the method for creating S functions to compute X beta, based on a model fitted with rms in effect. Like latexrms, Function.rms simplifies restricted cubic spline functions and factors out terms in second-order interactions. Function.rms will not work for models that have third-order interactions involving restricted cubic splines. Function.cph is a particular method for handling fits from cph, for which an intercept (the negative of the centering constant) is added to the model. sascode is a function that takes an S function such as one created by Function and does most of the editing to turn the function definition into a fragment of SAS code for computing X beta from the fitted model, along with assignment statements that initialize predictors to reference values. perlcode similarly creates Perl code to evaluate a fitted regression model.

Usage

# S3 method for rms
Function(object, intercept=NULL,
 digits=max(8, .Options$digits), posterior.summary=c('mean', 'median', 'mode'), ...)
# S3 method for cph
Function(object, intercept=-object$center, ...)
# Use result as fun(predictor1=value1, predictor2=value2, \dots)
sascode(object, file='', append=FALSE)
perlcode(object)

Value

Function returns an S-Plus function that can be invoked in any usual context. The function has one argument per predictor variable, and the default values of the predictors are set to adjust-to values (see datadist). Multiple predicted X beta values may be calculated by specifying vectors as arguments to the created function. All non-scalar argument values must have the same length. perlcode returns a character string with embedded newline characters.

Arguments

object: a fit created with rms in effect
intercept: an intercept value to use (not allowed to be specified to Function.cph). The intercept is usually retrieved from the regression coefficients automatically.
digits: number of significant digits to use for coefficients and knot locations
posterior.summary: if using a Bayesian model fit such as from blrm, specifies whether to use posterior mode/mean/median parameter estimates in generating the function
file: name of a file in which to write the SAS code. Default is to write to standard output.
append: set to TRUE to have sascode append code to an existing file named file.
...: arguments to pass to Function.rms from Function.cph

Author

Frank Harrell, Jeremy Stephens, and Thomas Dupont
Department of Biostatistics
Vanderbilt University
fh@fharrell.com

Examples

Run this code

suppressWarnings(RNGversion("3.5.0"))
set.seed(1331)
x1 <- exp(rnorm(100))
x2 <- factor(sample(c('a','b'),100,rep=TRUE))
dd <- datadist(x1, x2)
options(datadist='dd')
y  <- log(x1)^2+log(x1)*(x2=='b')+rnorm(100)/4
f  <- ols(y ~ pol(log(x1),2)*x2)
f$coef
g  <- Function(f, digits=5)
g
sascode(g)
cat(perlcode(g), '\n')
g()
g(x1=c(2,3), x2='b')   #could omit x2 since b is default category
predict(f, expand.grid(x1=c(2,3),x2='b'))
g8 <- Function(f)   # default is 8 sig. digits
g8(x1=c(2,3), x2='b')
options(datadist=NULL)


if (FALSE) {
require(survival)
# Make self-contained functions for computing survival probabilities
# using a log-normal regression
f <- psm(Surv(d.time, death) ~ rcs(age,4)*sex, dist='gaussian')
g <- Function(f)
surv <- Survival(f)
# Compute 2 and 5-year survival estimates for 50 year old male
surv(c(2,5), g(age=50, sex='male'))
}