unirec: Survival analysis for univariate recurrent event processes...
In blupsurv: Recurrent events with BLUPs

Description Usage Arguments Details Value Author(s) References See Also Examples

View source: R/unimethods.r

Tools for fitting proportional hazards models to clustered recurrent events data. Nested frailties are modeled by their best linear unbiased predictors under an auxiliary Poisson model. The computations are done using the code for bivrec, but effectively reduce to the method of Ma et al (2001).

1 2	## S3 method for class 'formula' unirec(formula, data = parent.frame(), ...)

`formula`	See details
`data`	See details
`...`	See details

Supported parameters are:

unirec(formula, data = parent.frame(), K1 = 10,
             excludevars1 = NULL, verbose = 1, dispest = "pearson",
             correction = "none", computesd = TRUE, fullS = TRUE, 
             fixzero = NULL, smooth = FALSE, maxiter = 200, 
             convergence = 1e-3, initial = NULL)

formula: a formula object, similar to coxph. The response to the left of the ~ should be a survival object generated by Surv, with three components (start, stop, status). The right side must contain an id(x) term, where x is a variable that takes a unique value for each subject within a cluster. It may also contain a cluster(y) term if y is the variable that indicates cluster membership, and a strata(z) term.
data: a data.frame with columns corresponding to the terms in the formula.
K1: either an integer, a vector of integers, or a value between 0 and 1, to determine the level of discretization. If it is an integer, K1 gives the number of breakpoints in the baseline hazard for the process. If it is a vector of integers, K1 should have length equal to the number of strata, and each value gives the number of breakpoints in the baseline hazard for each stratum. If it is a number between 0 and 1, it gives the ratio of the number of breakpoints relative to the maximum possible. Defaults to 10.
excludevars1: a vector of strings giving the names of variables or interactions that should be excluded from the model.
verbose: an integer from 0 to 3 that determines the quantity of output printed to the screen. Setting verbose=0 is completely silent.
dispest: a string, determining the method used to estimate the dispersion parameters. Possible values are "pearson", "marginal", "ohlsson", defaults to "pearson".
correction: a string describing the degree-of-freedom correction proposed by Ma. It only applies when dispest="pearson". Possible values are "single", "double", "none", defaults to "none"
computesd: a boolean determining whether standard errors should be computed. Defaults to TRUE.
fullS: logical determining whether to use the full sensitivity matrix in computing standard errors, or only the covariate portion. Using fullS=TRUE leads to more accurate results but increases computer time. Defaults to TRUE.
fixzero: a vector of strings, listing any dispersion parameters that should be set to 0. Can contain any subset of "clust1", "subj1", all else is ignored. Defaults to NULL.
smooth: logical, determines whether the baseline hazard should be smoothed at each iteration. Defaults to FALSE.
maxiter: an integer giving the maximum number of iterations permitted.
convergence: double, determines the value of the convergence criterion required at termination.
initial: a list of initial values in the format used by the code internally.

An object of class unirec with the following components:

`call`	the original call to the model-fitting function
`regression`	a list containing results from the regression fit in the last iteration. It has components `coefficients` a vector of regression coefficients for the process `loglik` conditional loglikelihood.
`frailty`	a list containing results from the frailty estimation in the last iteration. It has components `clust` cluster-level frailty estimates `subj` subject-level frailty estimates
`dispersion`	a list containing results from the dispersion parameter estimation in the last iteration. It has components `clust` cluster-level frailty variance `subj` subject-level frailty variance
`hazard`	a list describing the baseline hazard. It has components `breaks` matrix of breakpoints in the hazard for each stratum `hazard` matrix of hazards in each interval for each stratum
`summaries`	a list of summary matrices. It has components `regression` summary for the regression coefficients `dispersion` summary for the dispersion parameters

Emmanuel Sharef ess28@cornell.edu

R. Ma. “Random effects Cox models: A Poisson modelling approach”, Biometrika, 90 (1) 157-169, 2001.

E. Sharef and R. Strawderman. “A nested frailty model for clustered bivariate recurrent events”, in preparation.

summary.unirec, plot.unirec, bivrec

    data(m10Ji5)
    
    fit <- unirec( Surv(start, stop, delta) ~ 
                Z1 + cluster(i) + id(j), data = m10Ji5 )
                
    summary(fit)
    plot(fit)