fMmsm: Flexible transition intensity based models for two dependant multistate processes

Description

XXXX.

Usage

fMmsm(formula1, data1, id1, state1,
       params1 = NULL, spP1 = NULL, constraint1 = NULL,
       formula2, data2, id2, state2,
       params2 = NULL, spP2 = NULL, constraint2 = NULL,
       phi = NULL,
       pmethod = 'eigendecomp',
       aggregate = TRUE, sp.method = 'perf', iterlimsp = 50,
       Q.diagnostics = TRUE, iterlim = 100, verbose,
       tolsp = 1e-7, tolsp.EFS = 0.1, parallel = FALSE, no_cores = 2)

Value

The function returns an object of class fmsm as described in fmsmObject.

Arguments

formula1

Model specification for the transition intensities of the first process.

data1

Dataset of the first process.

id1

Name of the variable in the dataset representing the unique code associated with each patient in the first process.

state1

Name of the variable in the first process dataset representing the state occupied by the patient at the given time.

params1

XXX.

spP1

Smoothing parameter for the first process.

constraint1

XXX.

formula2

Model specification for the transition intensities of the second process.

data2

Dataset of the second process.

id2

Name of the variable in the dataset representing the unique code associated with each patient in the second process.

state2

Name of the variable in the second process dataset representing the state occupied by the patient at the given time.

params2

XXX.

spP2

Smoothing parameter for the second process.

constraint2

XXX.

phi

XXX.

pmethod

Which method should be used for the computation of the transition probability matrix. Available options are

'eigendecomp' (default): this method is based on the eigendecomposition of the transition intensity matrix (from Kalbfleisch & Lawless 1985);
'analytic': uses analytic expressions of the transition probabilities, obtained by solving the Kolmogorov forward differential equation, only implemented for IDMs for now;
'scaling&squaring': this is the scaling and squaring method implemented as proposed in Fung (2004).This is inefficient, so its use is not recommended. Can be used to investigate convergence errors.

aggregate

Whether or not data should be aggregated (this slightly improves efficiency as redundancies in the data are eliminated). The default is TRUE.

sp.method

Method to be used for smoothing parameter estimation. The default is magic, the automatic multiple smoothing parameter selection algorithm. Alternatively, efs can be used for the Fellner-Schall method. To suppress the smoothing parameter estimation set this to NULL.

iterlimsp

Maximum allowed iterations for smoothing parameter estimation.

Q.diagnostics

If TRUE, diagnostics information on the Q matrix are saved. The default TRUE.

iterlim

Maximum allowed iterations for trust region algorithm.

verbose

XXX.

tolsp

Convergence criterion used in magic based smoothing parameter estimation.

tolsp.EFS

Convergence criterion used in efs based smoothing parameter estimation.

parallel

If TRUE parallel computing is used during estimation. This can only be used by Windows users for now.

no_cores

Number of cores used if parallel computing chosen. The default is 2. If NULL, all available cores are used.