joineRmeta (version 0.1.2)

jointmetaSE: Bootstrapping function to obtain standard errors for jointmeta1 fit

Description

This function takes the results of a jointmeta1 fit and bootstraps it to find the standard errors of the parameter estimates.

Usage

jointmetaSE(
  fitted,
  n.boot,
  gpt,
  max.it,
  tol,
  print.detail = FALSE,
  overalleffects = NULL
)

Arguments

fitted

a jointmeta1 object

n.boot

the number of bootstraps to conduct. Note that confidence intervals will only be calculated if n.boot is greater than 100.

gpt

the number of quadrature points over which the integration with respect to the random effects will be performed. Will default to gpt = 3.

max.it

the maximum number of iterations that the EM algorithm will perform for each bootstrap fit. Will default to 350.

tol

the tolerance level before convergence of the algorithm is considered to have occurred. Default value is tol = 0.001.

print.detail

this argument determines the level of printing that is done during the bootstrapping. If TRUE then the parameter estimates from each bootstrap sample are output. Otherwise a progress bar is printed to indicated the proportion of bootstraps currently completed.

overalleffects

this argument indicates what if any overall effects will have their standard errors and confidence intervals calculated during the bootstrap procedure. An example of an overall effect would be the combined value of a treatment effect, and a treatment by study membership interaction. The overall treatment effect (the sum of these two values) could be of interest in an investigation. This argument is a list containing two elements, long and surv. Each of these elements contains a list of vectors, each of which contains the names of the parameters that make up the required overall effects

Value

a list containing three elements:

results

a data frame containing the estimates, standard errors and 95 and any overall effects requested.

covmat

the covariance matrix for the model parameters

bootstraps

a data frame containing the results of each bootstrap

Details

This function takes the results of a one stage joint model fit to data from multiple studies using the function jointmeta1 and performs n.boot bootstraps to determine the standard errors of the parameter estimates, and their confidence intervals if n.boot > 100.

The parameter overalleffects is designed for use in cases where interaction terms are included in the model specification, for example a model fitted using jointmeta1 which includes both treat and treat:study where treat is a binary treatment indicator variable and study is a study indicator. In this case it may be of interest to calculate the confidence interval for the value of treat + treat:study for a given study. This is done by calculating the value of the expression for each bootstrap, and calculating the standard errors for the expression in the same way as for the other parameters. Any overall effects to be calculated for the longitudinal sub-model are supplied as a list named long in the list overalleffects, with each element of this list containing a vector of the character names of the fixed effects to be summed to form an overall effect. Overall effects from the survival model are specified in a similar way to an element named surv

Examples

# NOT RUN {
jointdat<-tojointdata(longitudinal = simdat2$longitudinal,
                      survival = simdat2$survival, id = "id",
                      longoutcome = "Y",
                      timevarying = c("time","ltime"),
                      survtime = "survtime", cens = "cens",
                      time = "time")

onestagefit4 <- jointmeta1(data = jointdat,
                           long.formula = Y ~ 1 + time + treat + study,
                           long.rand.ind = c("int", "time"),
                           long.rand.stud = c("treat"),
                           sharingstrct = "randprop",
                           surv.formula = Surv(survtime, cens) ~ treat,
                           study.name = "study", strat = TRUE)

onestagefit4SE <- jointmetaSE(fitted = onestagefit4, n.boot = 200)
# }
# NOT RUN {

# }