PCDSpline (version 1.0)

PCDReg.nf: Regression analysis of panel count data under the non-homogeneous Poisson process model

Description

Fits the nonhomogeneous Poisson process model to panel count data using EM algorithm.

Usage

PCDReg.nf(DATA, order, placement, nknot, myknots, binit, ginit, t.seq, tol)

Arguments

DATA

use specified data structure.

order

the order of basis functions.

placement

logical, if TRUE knots are placed evenly across the observed intervals based on the input data set; if FALSE knots should be specified by the user. see myknots.

nknot

the number of knots to be used.

myknots

a sequence of increasing points whose length is nknot.

binit

initial estimate of regression coefficients.

ginit

initial estimate of spline coefficients whose length should be (order+nknot-2) or (order+length(myknots)-2).

t.seq

an increasing sequence of points at which the baseline mean function is evaluated.

tol

the convergence criterion of the EM algorithm.

Value

beta

estimates of regression coefficients.

gamma

estimates of spline coefficients.

var.b

the variance covariance matrix of regression coefficients.

Hess

Hessian matrix.

knots

the knots used in the analysis.

bmf

estimated baseline mean function evaluated at the points t.seq; use pmf to plot the baseline mean fuction.

AIC

the Akaike information criterion.

BIC

the Bayesian information/Schwarz criterion.

flag

the indicator whether the Hessian matrix is non-singular. When flag="TRUE",the variance estimate may not be accurate.

Details

The above function fits the non-homogeneous Poisson process model to panel count data via EM algorithm.

References

Yao, B., Wang, L., and He, X. (2014+). Semiparametric regression analysis of panel count data allowing for within-subject correlation.

See Also

PCDReg.wf

Examples

##Simulated Data

n=13; #number of subjects

##generate the number of observations for each subject
k=rpois(n,6)+1; K=max(k);
  
##generate random time gaps for each subject
y=matrix(,n,K);
for (i in 1:n){y[i,1:k[i]]=rexp(k[i],1)} 

##get observation time points for each subject
t=matrix(,n,K);
for (i in 1:n){
  for (j in 2:K){
    t[i,1] = y[i,1]
    t[i,j] = y[i,j]+t[i,j-1]
  }
}

##covariate x1 and x2 generated from Normal(0,0.5^2) and Bernoulli(0.5) respectively
x1=rnorm(n,0,0.5); x2=rbinom(n,1,0.5); x=cbind(x1,x2)

##true regression parameters and frailty variance parameter
beta1=1; beta2=-1; nu=0.5; 
parms=c(beta1,beta2)
phi=rgamma(n,nu,nu) 

##true baseline mean function
mu=function(t){2*t^(0.5)} 

##get the number of events between time intervals
z=matrix(,n,K);
xparms<-c();for (s in 1:nrow(x)){xparms[s]<-sum(x[s,]*parms)}
for (i in 1:n){
 z[i,1]<-rpois(1,mu(t[i,1])*exp(xparms[i])*phi[i]) 
 if (k[i]>1){
 z[i,2:k[i]]<-rpois(k[i]-1,(mu(t[i,2:k[i]])-mu(t[i,1:(k[i]-1)]))*exp(xparms[i])*phi[i])
 }
}

TestD<-list(t=t, x=x, z=z, k=k, K=K)

fit<-PCDReg.nf(DATA = TestD, order = 1, placement = TRUE, nknot=3,
               myknots, binit = c(-0.5,0.5) , ginit = seq(0.1,2),
               t.seq = seq(0,15,0.2), tol=10^(-3))