JointMult: Estimation of a joint model for multivariate latent processes...

Description Usage Arguments Value Author(s) References

View source: R/JointMult.R

Description

This function fits a joint model for multivariate longitudinal markers (possibly summarized into latent processes) and clinical endpoints. The estimation is performed within the Maximum Likelihood Framework benefiting from an exact likelihood formulation. More details are given below.

Each dimension (constituted of a unique longitudinal marker or of several longitudinal markers measuring the same underlying latent process) is modeled according to a mixed model that handles continuous (Gaussian or non-Gaussian, curvilinear) data. The technique for each dimension follows the methodology of lcmm and multlcmm functions (lcmm package). All the dimensions can be correlated through correlated random effects. The dimensions are linked with one or two competing clinical endpoints through a joint degradation process model. Specifically, each clinical endpoint is defined as a binary repeated endpoint either measured at various visit times or derived from a time-to-event discretized into time intervals. The model assumes that the clinical endpoint reaches 1 when its underlying degradation process becomes above a threshold to estimate, this degradation process being defined as a linear combination of the longitudinal dimensions.

The methodology is fully described in : Proust-Lima C, Philipps V, Dartigues JF (2018). A joint model for multiple dynamic processes and clinical endpoints: application to Alzheimer’s disease. https://arxiv.org/abs/1803.10043

Usage

1
2
3
4
JointMult(Y, D, data, var.time, RE = "block-diag", BM = "diag", 
    B, posfix, maxiter = 0, nproc = 1, verbose = FALSE, 
    file = "", pred = FALSE, breaks = NULL, delayed = TRUE,  
    eps = c(1e-04, 1e-04, 0.001)) 

Arguments

Y

a list of multlcmm objects. Each multlcmm object defines the outcomes and longitudinal structure of one dimension of the model. The number of outcomes, covariates or random effects can differ between dimensions.

D

a list of two-sided formula defining the event part of the model. The left side should be either longDiag(time0, time, indicator) with time0 the variable indicating entry time (which will only be used for delayed entry), time the repeated measurement times, indicator the repeated event indicator at each time or survBreak(T0, T, Event) with T0 the entry time, T the event time and Event the event indicator.

var.time

a character vector indicating the name of the time variable of each dimension. The scales of these different time variables should be the same.

RE

an indicator of the random effect structure between dimensions. Should be either "block-diag" for independent random effects between dimensions (the internal structure being defined in the multlcmm object) or "full" for correlated random effects between dimensions.

BM

in the case where Brownian motions are included in the multlcmm objects, an indicator of the correlation of the Brownian motions. Should be "diag" for independence, or "full" for correlated Brownian motions between dimensions.

breaks

optional vector specifying the break points in the case where the event time is discretized into time intervals.

delayed

logical vector indicating, for each event, if delayed entry should be accounted for.

B

optional specification for the initial values for the parameters.

posfix

optional vector specifying the indices in vector B of the parameters that should not be estimated. By default, all parameters are estimated.

maxiter

optional maximum number of iterations for the Marquardt iterative algorithm. By default, maxiter=XXX.

eps

optional thresholds for the convergence criteria. Default is set to 0.0001 for the parameters stability, to 0.0001 for the log-likelihood stability, and to 0.001 for the criterion based on second derivatives.

nproc

optional integer indicating the number of processors to be used for parallel computation. Default is set to 1 (i.e., the algorithm runs sequentially).

verbose

logical indicating if information about computation should be reported. Default to FALSE.

file

optional filename in which to report information about computation (if verbose=TRUE).

pred

logical indicating if subject-specific predictions should be computed. Default is set to FALSE.

Value

istop

convergence status: 1 if the model converged properly, 2 if the maximum number of iterations was reached without convergence, >2 if an error occurred.

ni

number of iterations

loglik

log-likelihood of the model

b

vector of estimated parameters

v

estimated variance matrix of the estimated parameters

convcrit

convergence criteria at stop point

time

estimation time

nproc

number of processors

bopt

total vector of estimated and fixed parameters

nef

number of fixed effects parameters

ncontr

number of contrasts parameters

nea

number of random effects variables

nvc

number of random effects parameters

idiag

indicator of intra dimension correlation between random effects

ntr

number of link function parameters

ntrtot

number of link function parameters

ncor

number of Brownian motion parameters

nalea

number of outcome-specific random effects parameters

ny

number of longitudinal outcomes

link

type of link functions

nodes

nodes for the link functions

nRE

number of correlation parameters for the random effects between dimensions

nRM

number of correlation for the Brownian motions between dimensions

varD

time independent covariates in the event model

vardept

time dependent covariates in the event model

nvarD

number of time independent covariates in the event model

ndept

number of time dependent covariates in the event model

idcause

indicator of presence of the covariates in the event model

call

the model's call

fix

indicator of fixed parameters

chol

indicator of Cholesky transformation

Ynames

name of the longitudinal outcomes

Xnames

name of the covariates in the longitudinal part

ns

number of subjects

nbevt

number of observed events

nbmes

mean number of measurement

entreRetard

indicator of left truncation

discretise

indicator of dicretization

breaks

list of break points

VRE

variance-covariance matrix of the random effects

corRE

correlation matrix of the random effects

mod

list of updated multlcmm objects

Author(s)

Cecile Proust-Lima and Viviane Philipps

References

Proust-Lima C, Philipps V, Dartigues JF (2018). A joint model for multiple dynamic processes and clinical endpoints: application to Alzheimer’s disease. https://arxiv.org/abs/1803.10043


VivianePhilipps/multLPM documentation built on Oct. 11, 2018, 12:22 a.m.