| Type: | Package |
| Title: | TBF Methodology Extension for Multinomial Outcomes |
| Version: | 0.1.3 |
| VignetteBuilder: | knitr |
| Description: | Extends the test-based Bayes factor (TBF) methodology to multinomial regression models and discrete time-to-event models with competing risks. The TBF methodology has been well developed and implemented for the generalised linear model [Held et al. (2015) <doi:10.1214/14-STS510>] and for the Cox model [Held et al. (2016) <doi:10.1002/sim.7089>]. |
| Depends: | VGAM, nnet, parallel, stats, stringr, plotrix, methods |
| Suggests: | knitr, splines |
| License: | GPL-2 | GPL-3 [expanded from: GPL (≥ 2)] |
| Encoding: | UTF-8 |
| LazyData: | true |
| RoxygenNote: | 6.0.1 |
| Author: | Rachel Heyard [aut, cre] |
| Maintainer: | Rachel Heyard <rachel.heyard@uzh.ch> |
| NeedsCompilation: | no |
| Packaged: | 2018-10-09 09:02:31 UTC; rachel |
| Repository: | CRAN |
| Date/Publication: | 2018-10-12 13:30:06 UTC |
Objective Bayesian variable selection for multinomial regression and discrete time-to-event models with competing risks
Description
Extension of the TBF methdology introduced by Held et al. (2015) <doi:10.1214/14-STS510> to discrete time-to-event models with competing risks (also applicable to the multinomial regression model)
Author(s)
Rachel Heyard rachel.heyard@uzh.ch
Marginal likelihoods based on AIC or BIC
Description
This function computes the marginal likelihoods based on the AIC or on the BIC, that will later be used to calculate the TBF.
Usage
AIC_BIC_based_marginalLikelihood(fullModel = NULL, candidateModels = NULL,
data, discreteSurv = TRUE, AIC = TRUE, package = "nnet", maxit = 150,
numberCores = 1)
Arguments
fullModel |
formula of the model including all potential variables |
candidateModels |
Instead of defining the full model we can also specify the candidate models whose deviance statistic and d.o.f should be computed |
data |
the data |
discreteSurv |
Boolean variable telling us whether a ‘simple’ multinomial regression is looked for or if the goal is a discrete survival-time model for multiple modes of failure is needed. |
AIC |
if |
package |
Which package should be used to fit the models; by default
the |
maxit |
Only needs to be specified with package |
numberCores |
How many cores should be used in parallel? |
Value
a vector with the marginal likelihoods of all candidate models
Author(s)
Rachel Heyard
Examples
# data extraction:
data("VAP_data")
# the definition of the full model with three potential predictors:
FULL <- outcome ~ ns(day, df = 4) + gender + type + SOFA
# here the define time as a spline with 3 knots
# now we can compute the marginal likelihoods based on the AIC f.ex:
mL_AIC <-
AIC_BIC_based_marginalLikelihood(fullModel = FULL,
data = VAP_data,
discreteSurv = TRUE,
AIC = TRUE)
Cause-specific variable selection (CSVS)
Description
This function performs CSVS given a model fitted using the multinom()
function of the nnet package or the vglm() function of the
VGAM package.
Usage
CSVS(g, model, discreteSurv = TRUE, nbIntercepts = NULL, package = "nnet")
Arguments
g |
the estimated g, must be fixed to one value |
model |
the model fitted using either |
discreteSurv |
Boolean variable telling us whether a 'simple' multinomial regression is looked for or if the goal is a discrete survival-time model for multiple modes of failure is needed. |
nbIntercepts |
how many cause-specific intercepts are there? they |
package |
Which package has been used to fit the model, |
Author(s)
Rachel Heyard
Examples
# data extraction:
data("VAP_data")
# the definition of the full model with three potential predictors:
FULL <- outcome ~ ns(day, df = 4) + gender + type + SOFA
# here the define time as a spline with 3 knots
# we first need to fit the multinomial model:
model_full <- multinom(formula = FULL, data = VAP_data,
maxit = 150, trace = FALSE)
G <- 9 # let's suppose g equals to nine
# then we proceed to CSVS
CSVS_nnet <- CSVS(g = G, model = model_full,
discreteSurv = TRUE, package = 'nnet')
Posterior inclusion probabilities (PIPs) by landmarking
Description
This function gives us the PIPs for each landmark.
Usage
PIPs_by_landmarking(fullModel, data, discreteSurv = TRUE, numberCores = 1,
package = "nnet", maxit = 150, prior = "flat", method = "LEB",
landmarkLength = 1, lastlandmark, timeVariableName)
Arguments
fullModel |
formula of the model including all potential variables |
data |
the data frame with all the information |
discreteSurv |
Boolean variable telling us whether a 'simple' multinomial regression is looked for or if the goal is a discrete survival-time model for multiple modes of failure is needed. |
numberCores |
How many cores should be used in parallel? |
package |
Which package should be used to fit the models; by default
the |
maxit |
Only needs to be specified with package |
prior |
Prior on the model space |
method |
Method for the g definition |
landmarkLength |
Length of the landmark, by default we use each day |
lastlandmark |
Where will be the last landmark? |
timeVariableName |
What is the name of the variable indicating time? |
Value
a list with the PIPs for each landmark
Author(s)
Rachel Heyard
Examples
# extract the data:
data("VAP_data")
# the definition of the full model with three potential predictors:
FULL <- outcome ~ ns(day, df = 4) + gender + type + SOFA
# here we define time as a spline with 3 knots
PIPs_landmark <- PIPs_by_landmarking(fullModel = FULL, data = VAP_data,
discreteSurv = TRUE, numberCores = 1,
package = 'nnet', maxit = 150,
prior = 'flat', method = 'LEB',
landmarkLength = 7, lastlandmark = 21,
timeVariableName = 'day')
Posterior model probability
Description
This function computes the posterior probability of all candidate models
Usage
PMP(fullModel = NULL, candidateModels = NULL, data = NULL,
discreteSurv = TRUE, modelPrior = NULL, method = "LEB",
prior = "flat", package = "nnet", maxit = 150, numberCores = 1)
Arguments
fullModel |
formula of the model including all potential variables |
candidateModels |
Instead of defining the full model we can also specify the candidate models whose deviance statistic and d.o.f should be computed |
data |
the data frame with all the information |
discreteSurv |
Boolean variable telling us whether a 'simple' multinomial regression is looked for or if the goal is a discrete survival-time model for multiple modes of failure is needed. |
modelPrior |
optionaly the model priors can be computed before if candidateModels is different from NULL. |
method |
tells us which method for the definition of g should be
used. Possibilities are: |
prior |
should a dependent or a flat prior be used on the model space?
Only needed if |
package |
Which package should be used to fit the models; by default
the |
maxit |
Only needs to be specified with package |
numberCores |
How many cores should be used in parallel? |
Value
an object of class TBF.ingredients
Author(s)
Rachel Heyard
Examples
# extract the data:
data("VAP_data")
# the definition of the full model with three potential predictors:
FULL <- outcome ~ ns(day, df = 4) + gender + type + SOFA
# here we define time as a spline with 3 knots
# computation of the posterior model probabilities:
test <- PMP(fullModel = FULL, data = VAP_data,
discreteSurv = TRUE, maxit = 150)
class(test)
Class for PMP objects
Description
Class for PMP objects
Test-based Bayes factor
Description
This function computes the TBF as well as g
Usage
TBF(ingredients = NULL, fullModel = NULL, method = "LEB", data = NULL,
discreteSurv = TRUE, prior = NULL, package = "nnet", maxit = 150)
Arguments
ingredients |
|
fullModel |
if |
method |
tells us which method for the definition of g should be
used. Possibilities are: |
data |
the data frame with all the information. Only needed if
|
discreteSurv |
Boolean variable telling us whether a 'simple' multinomial regression is looked for or if the goal is a discrete survival-time model for multiple modes of failure is needed. |
prior |
should a dependent or a flat prior be used on the model space?
Only needed if |
package |
Which package should be used to fit the models; by default
the |
maxit |
Only needs to be specified with package |
Value
A list with the TBF and the g (if it is fixed) for all the candidate models.
Author(s)
Rachel Heyard
Ingredients to calculate the TBF
Description
This function calculates the ingredients needed to compute the TBFs: like the deviances with their degrees of freedom of the relevant candidate models.
Usage
TBF_ingredients(fullModel = NULL, data, discreteSurv = FALSE,
numberCores = 1, candidateModels = NULL, package = "nnet",
maxit = 150)
Arguments
fullModel |
formula of the model including all potential variables |
data |
the data frame with all the information |
discreteSurv |
Boolean variable telling us whether a 'simple' multinomial regression is looked for or if the goal is a discrete survival-time model for multiple modes of failure is needed. |
numberCores |
How many cores should be used in parallel? |
candidateModels |
Instead of defining the full model we can also specify the candidate models whose deviance statistic and d.o.f should be computed |
package |
Which package should be used to fit the models; by default
the |
maxit |
Only needs to be specified with package |
Value
an object of class TBF.ingredients
Author(s)
Rachel Heyard
Data on VAP acquistion in one ICU
Description
It is a tiny subset of the OUTCOMEREA database whose only perhaps will be to test an illustrate the functions of this package.
Usage
data(VAP_data)
Format
A data frame with 1640 rows and 7 variables on 90 distinct patients:
- ID
distinct ID for each patient
- day
day of ventilation, day = 1 is the first day of ventilation
- type
is it a medical or a surgical patient
- gender
gender of the patient, 1 = male, 0 = female
- SAPSadmission
the SAPS 2 score at admission to the ICU
- SOFA
the daily SOFA score
- outcome
final outcome after the first observation period
Formulas of all the candidate models
Description
This function retrieves the formulas of all the candidate models if the reference model is the null / baseline model.
Usage
all_formulas(fullModel, discreteSurv = TRUE)
Arguments
fullModel |
formula of the model including all potential variables |
discreteSurv |
Boolean variable telling us whether a ‘simple’ multinomial regression is looked for or if the goal is a discrete survival-time model for multiple modes of failure is needed. |
Value
character vector with all the formulas; the first one will be the reference model; the last element will be the full model.
Author(s)
Rachel Heyard
Examples
data("VAP_data")
FULL <- outcome ~ ns(day, df = 4) + male + type + SOFA
models <- TBFmultinomial:::all_formulas(fullModel = FULL,
discreteSurv = TRUE)
# models
Convert a PMP object into a data frame
Description
This function takes a PMP object an returns a data.frame summarising
the information.
Usage
## S3 method for class 'PMP'
as.data.frame(x, ...)
Arguments
x |
valid |
... |
arguments to be passed to |
Value
a data.frame with the posterior and prior probabilities as
well as the definition of the models
Author(s)
Rachel Heyard
Prior model probability
Description
This function computes the prior model probabilities of the candidate models
Usage
model_priors(fullModel, discreteSurv = TRUE, modelPrior = "flat")
Arguments
fullModel |
formula of the model including all potential variables |
discreteSurv |
Boolean var telling us whether a 'simple' multinomial regression is looked for or if the goal is a discrete survival-time model for multiple modes of failure is needed. |
modelPrior |
what prior should be used on the model space?
|
Value
a numerical vector with the prior model probabilities
Author(s)
Rachel Heyard
Examples
# the definition of the full model with three potential predictors:
FULL <- outcome ~ ns(day, df = 4) + gender + type + SOFA
# here we define time as a spline with 3 knots
priors <- model_priors(fullModel = FULL, discreteSurv = TRUE,
modelPrior = 'dependent')
Plot a CSVS object
Description
Plot a CSVS object
Usage
plot_CSVS(CSVSobject, namesVar = NULL, shrunken = FALSE,
standardized = FALSE, numberIntercepts, ...)
Arguments
CSVSobject |
valid |
namesVar |
names of the variables |
shrunken |
should the coefficients be shrunken? |
standardized |
should the coefficients be standardized? |
numberIntercepts |
how many cause-specific intercepts are in the model for each outcome |
... |
parameters for plot |
Author(s)
Rachel Heyard
Posterior inclusion probability (PIP)
Description
This function computes the PIPs of all potential predictors
Usage
postInclusionProb(object)
Arguments
object |
An object of class |
Value
an named vector with all PIPs
Author(s)
Rachel Heyard
Examples
# extract the data:
data("VAP_data")
# the definition of the full model with three potential predictors:
FULL <- outcome ~ ns(day, df = 4) + gender + type + SOFA
# here we define time as a spline with 3 knots
# computation of the posterior model probabilities:
test <- PMP(fullModel = FULL, data = VAP_data,
discreteSurv = TRUE, maxit = 150)
class(test)
#computation of the posterior inclusion probabilities:
postInclusionProb(test)
Samples from a PMP object
Description
This function samples from a specific model inside a PMP object.
Usage
sample_multinomial(PMP_object, shrink = TRUE, data, which = "MPM",
discreteSurv = TRUE)
Arguments
PMP_object |
formula of the model including all potential variables |
shrink |
should the coefficients be shrunken towards their prior mean? |
data |
the (training) data frame with all the information |
which |
which model should be sampled from? either an integer, 'MPM' or 'MAP' |
discreteSurv |
Boolean variable telling us whether a 'simple' multinomial regression is looked for or if the goal is a discrete survival-time model for multiple modes of failure is needed. |
Value
returns an object with the model coefficients and supplementary information
Author(s)
Rachel Heyard