confintmlx: Confidence intervals for population parameters
In MarcLavielle/Rsmlx: R Speaks 'Monolix'

confintmlx

R Documentation

Confidence intervals for population parameters

Description

Compute confidence intervals for the population parameters estimated by Monolix.

Usage

confintmlx(
  project,
  parameters = "all",
  method = "fim",
  level = 0.9,
  linearization = TRUE,
  nboot = 100,
  parametric = FALSE,
  settings = NULL
)

Arguments

`project`	a Monolix project
`parameters`	list of parameters for which confidence intervals are computed (default="all")
`method`	method `c("fim", "proflike", "bootstrap")` (default="fim")
`level`	confidence level, a real number between 0 and 1 (default=0.90)
`linearization`	TRUE/FALSE whether the calculation of the standard errors (default=TRUE) or the profile likelihood is based on a linearization of the model (default=TRUE)
`nboot`	number of bootstrat replicates (default=100, used when method="bootstrap")
`parametric`	boolean to define if parametric bootstrap is performed (new data is drawn from the model), (default: FALSE)
`settings`	a list of settings for the profile likelihood method: `max.iter` maximum number of iterations to find the solution (default=10) `tol.LL` absolute tolerance for -2LL (default=0.001) `tol.param` relative tolerance for the parameter (default=0.01) `print` TRUE/FALSE display the results (default=TRUE)

Details

Most functionality to compute confidence intervals (other than profile likelihood) is now available directly in the lixoftConnectors package. Please migrate the following uses of this function:

confintmlx method		lixoftConnectors function
`"fim"`	`linearizarion = TRUE`	`getEstimatedConfidenceIntervals(method = "linearization")`
`"fim"`	`linearizarion = FALSE`	`getEstimatedConfidenceIntervals(method = "stochasticApproximation")`
`"bootstrap"`	`parametric = TRUE`	`runBootstrap(method = "parametric")`
`"bootstrap"`	`parametric = FALSE`	`runBootstrap(method = "nonparametric")`

For method="proflike", continue using this function.

The method used for computing the confidence intervals can be either based on the standard errors derived from an estimation of the Fisher Information Matrix ("fim"), on the profile likelihood ("proflike") or on nonparametric bootstrap estimate ("bootstrap"). method="fim" is used by default.

When method="fim", the FIM can be either estimated using a linearization of the model or a stochastic approximation. When method="proflike", the observed likelihood can be either estimated using a linearization of the model or an importance sampling Monte Carlo procedure. When method="bootstrap", the bootstrap estimates are obtained using the bootmlx function

Value

a list with the computed confidence intervals, the method used and the level.

Examples

# RsmlxDemo2.mlxtran is a Monolix project for modelling the PK of warfarin using a PK model 
# with parameters ka, V, Cl.

# confintmlx will compute a 90% confidence interval for all the population parameters 
# using the population estimates obtained by Monolix and the Fisher Information Matrix 
# estimated by linearization
r1 <- confintmlx(project="RsmlxDemo2.mlxtran") 

# 95% confidence intervals are now computed, using the FIM estimated by Monolix using a 
# stochastic approximation algorithm:
r2 <- confintmlx(project="RsmlxDemo2.mlxtran", linearization=FALSE, level=0.95) 

# Confidence intervals are computed for ka_pop and omega_ka only, 
# using the profile likelihood method:
r <- confintmlx(project    = "RsmlxDemo2.mlxtran", 
                method     = "proflike", 
                parameters = c("ka_pop","omega_ka")) 

# Confidence intervals are computed using 200 bootstrap samples:
r3 <- confintmlx(project="RsmlxDemo2.mlxtran", method="bootstrap", nboot=200)

# See http://monolix.lixoft.com/rsmlx/confintmlx/ for detailed examples of use of confintmlx
# Download the demo examples here: http://monolix.lixoft.com/rsmlx/installation

MarcLavielle/Rsmlx documentation built on March 1, 2024, 2:01 a.m.