multiRec: Fit the multi-type recurrent event model

In multiRec: Analysis of Multi-Type Recurrent Events

Fit the multi-type recurrent event model

Description

multi-type events, may be recurrent or terminating

Usage

multiRec(
  ...,
  data,
  eventVar = "event",
  startTimeVar = "tstart",
  stopTimeVar = "tstop",
  idVar = "id",
  na.action = na.omit,
  link = c("log", "identity", "yj"),
  SANN.init = 0,
  noEvent = c("", NA),
  robust = FALSE,
  method = c("BFGS", "Nelder-Mead", "CG", "SANN"),
  method.seed = NULL,
  hazardPrefix = "nPrior.",
  maxit,
  trace = FALSE,
  fitDetails = FALSE,
  hessian = c("optim", "numDeriv")
)

Arguments

...	one or more models (see details)
data	the dataset
eventVar	character, the name of the event variable in the data= dataset
startTimeVar	character, the name of the variable that holds the start time of each interval in the data= dataset
stopTimeVar	character, the name of the variable that holds the stop time of each interval in the data= dataset
idVar	character, the name of the id variable which uniquely identifies individuals in the data= dataset
na.action	function, a function which indicates what should happen when the data contain NAs. The default is na.omit.
link	the link used to model the hazard
SANN.init	integer, if this is greater than 0 it indicates the number of simulated annealing iterations used to refine the initial estimate of the parameters before calling the method specified in method=.
noEvent	character, the value(s) of the event variable that indicate that no event occurred in an interval
robust	logical, if TRUE robust standard errors are used
method	character, the optimization method see the optim function for details
method.seed	numeric, the seed used
hazardPrefix	character, the prefix used to denote prior event functions in the formulas for the hazard
maxit	integer, the maximum number of iteration for the optimization method. Defaults to 500 for Nelder-Mead, 100 for all other methods.
trace	logical, if TRUE additional information is printed by the algorithm
fitDetails	logical, if TRUE the returned fit object contains additional information used to create fit diagnostics.
hessian	character, use the Hessian calculated in optim or calculated separately in numDeriv::hessian().

Details

models for the hazard are of the form

lhs ~ rhs

where lhs is one of the possible events and rhs is the usual model right hand side, with one addition: the pseudo-function "nPrior." can be used to indicate a dependence of the hazard on prior events. This function has the form

nPrior.event(~formula, alpha=value)

where

event

should be replaced by the appropriate event. For example, "nPrior.stroke" indicates a dependence of the hazard on the number of strokes prior to the current interval

formula

is a possibly empty formula which expresses the possible covariate effects on the extent to which prior events impact the hazard. For example "nPrior.stroke(~hba1c+sbp)" indicates a dependence on the number of prior strokes which may be modulated by HbA1c (variable hba1c) and systolic blood pressure (variable sbp)

value

is the power to which the number of prior events should be raised. The default is 1, i.e. the hazard depends simply on the number of prior events. A value of 0.5 means that the hazard depends on the square root of the number of prior events, i.e. that additional events have a diminishing effect on the hazard. Specifying alpha=NA means that the power is unknown and should be estimated.

For example a model of myocardial infaction (event='mi') might be

mi ~ bmi + nPrior.mi(alpha=0.5) + nPrior.stroke(~hba1c+sbp)

to indicate that the hazard of mi depends on body mass index (variable bmi), on the square root of the number of prior MI's and on the number of prior strokes, modulated by hba1c and systolic blood pressure.

Value

an object of type agMTRE

Examples

# Fit a model for the multiRecCVD2 dataset
fit = multiRec(hf   ~ nPrior.afib(~male),
               afib ~ age + nPrior.afib() + nPrior.hf(),
               data=multiRecCVD2,
               idVar='id',
               link='log')

# Fit the model with robust variances
fit = multiRec(hf   ~ nPrior.afib(~male),
               afib ~ age + nPrior.afib() + nPrior.hf(),
               data=multiRecCVD2,
               idVar='id',
               link='log',
               robust = TRUE)   # Request robust (sandwich) variances

# Use a specified power
fit = multiRec(hf   ~ nPrior.afib(~male),
               afib ~ age + nPrior.afib(alpha=0.5) + nPrior.hf(),
               data=multiRecCVD2,
               idVar='id',
               link='log',
               robust = TRUE)

# Display the coefficients
coef(fit)
confint(fit)

# Display intervals with a poor fit
boxplot(resid(fit))

multiRec documentation built on Feb. 3, 2026, 5:06 p.m.