pif.kernel: Kernel-based estimate of Potential Impact Fraction

In INSP-RH/pif: Potential Impact Fraction and Population Attributable Fraction for Cross-Sectional Data

Description

Function for estimating the Potential Impact Fraction pif from a cross-sectional sample of the exposure X with a known Relative Risk function rr with parameters theta using kernels.

Usage

pif.kernel(X, thetahat, rr, cft = NA, weights = rep(1/nrow(as.matrix(X)),
  nrow(as.matrix(X))), adjust = 1, n = 512, ktype = c("gaussian",
  "epanechnikov", "rectangular", "triangular", "biweight", "cosine",
  "optcosine"), bw = c("SJ", "nrd0", "nrd", "ucv", "bcv"),
  check_exposure = TRUE, check_rr = TRUE, check_integrals = TRUE,
  is_paf = FALSE)

Arguments

X	Random sample (data.frame) which includes exposure and covariates. or sample mean if approximate method is selected.
thetahat	Estimator (vector) of theta for the Relative Risk function.
rr	Function for Relative Risk which uses parameter theta. The order of the parameters shound be rr(X, theta). **Optional**
cft	Function cft(X) for counterfactual. Leave empty for the Population Attributable Fraction paf where counterfactual is 0 exposure.
weights	Normalized survey weights for the sample X.
adjust	Adjust bandwith parameter from density from density.
n	Number of equally spaced points at which the density is to be estimated (see density).
ktype	kernel type: "gaussian", "epanechnikov", "rectangular", "triangular", "biweight", "cosine", "optcosine". Additional information on kernels in density
bw	Smoothing bandwith parameter from density from density. Default "SJ"
check_exposure	Check that exposure X is positive and numeric.
check_rr	Check that Relative Risk function rr equals 1 when evaluated at 0.
check_integrals	Check that counterfactual and relative risk's expected values are well defined for this scenario.
is_paf	Boolean forcing evaluation of paf.

Value

pif Estimate of Potential Impact Fraction

Note

In practice pif.empirical should be prefered as convergence is faster than pif.kernel for most functions. In addition, the scope of pif.kernel is limited as it does not work with multivariate exposure data X.

Author(s)

Rodrigo Zepeda-Tello [e-mail]

Dalia Camacho-García-Formentí [e-mail]

See Also

pif which is a wrapper for all pif methods (pif.empirical, pif.approximate, pif.kernel).

For estimation of the Population Attributable Fraction see paf.

For more information on kernels see density

Examples

#Example 1: Relative risk given by exponential
#--------------------------------------------
set.seed(18427)
X        <- data.frame(rnorm(100,3,.5))
thetahat <- 0.12
rr       <- function(X, theta){exp(theta*X)}
pif.kernel(X, thetahat, rr, cft = function(X){ 0.5*X })

#Choose a different kernel
pif.kernel(X, thetahat, rr, cft = function(X){ 0.5*X }, ktype = "gaussian")

#Specify kernel options
pif.kernel(X, thetahat, rr, cft = function(X){ 0.5*X }, ktype = "gaussian", 
bw = "nrd", adjust = 0.5, n = 1100)

#Without counterfactual estimates PAF
pif.kernel(X, thetahat, rr) 
 
#Example 2: Linear relative risk
#--------------------------------------------
pif.kernel(X, thetahat, rr = function(X, theta){theta*X + 1}, 
               cft = function(X){ 0.5*X })

#Example 3: More complex counterfactual
#--------------------------------------------
set.seed(18427)
X       <- data.frame(rnorm(100,4,1))
thetahat <- c(0.12, 0.03)
rr       <- function(X, theta){1 + theta[1]*X + theta[2]*X^2}

#Creating a counterfactual. As rr requires a bivariate input, cft should 
#return a two-column matrix
cft  <- function(X){
   X[which(X > 4)] <- 1
   return(X)
}
pif.kernel(X, thetahat, rr, cft) 

INSP-RH/pif documentation built on May 7, 2019, 6:01 a.m.