| Type: | Package |
| Title: | Scalable and Robust Latent Trajectory Class Analysis |
| Description: | Conduct latent trajectory class analysis with longitudinal data. Our method supports longitudinal continuous, binary and count data. For more methodological details, please refer to Hart, K.R., Fei, T. and Hanfelt, J.J. (2020), Scalable and robust latent trajectory class analysis using artificial likelihood. Biometrics <doi:10.1111/biom.13366>. |
| Depends: | R (≥ 3.3.0) |
| Imports: | stats, geepack, VGAM, Matrix, mvtnorm |
| Version: | 0.1.0 |
| Maintainer: | Teng Fei <tfei@emory.edu> |
| BugReports: | https://github.com/tengfei-emory/SLTCA/issues |
| License: | GPL-2 | GPL-3 [expanded from: GPL (≥ 2)] |
| Encoding: | UTF-8 |
| LazyData: | true |
| RoxygenNote: | 7.1.1 |
| NeedsCompilation: | no |
| Packaged: | 2020-09-17 18:18:58 UTC; TFEI |
| Author: | Kari Hart [aut],
Teng Fei  [cre,
aut],
John Hanfelt
[aut] |
| Repository: | CRAN |
| Date/Publication: | 2020-09-23 15:30:07 UTC |
Simulate a dataset which can be analyzed by SLTCA
Description
Simulate a dataset with longitudinal observations.
Usage
simulation(n)
Arguments
n |
Sample size. |
Value
Returns a data frame with 6 longitudinal features y.1 - y.6, including count (y.1 and y.2), binary (y.3 and y.4) and continuous (y.5 and y.6) type. Variable baselinecov is the baseline risk factor of latent classes. Variable latent is the true latent class labels.
Author(s)
Teng Fei. Email: tfei@emory.edu
References
Hart, K.R., Fei, T. and Hanfelt, J.J. (2020), Scalable and robust latent trajectory class analysis using artificial likelihood. Biometrics. Accepted Author Manuscript <doi:10.1111/biom.13366>.
Examples
dat <- simulation(500)
Scalable and Robust Latent Trajectory Class Analysis Using Artificial Likelihood
Description
Conduct latent trajectory class analysis with longitudinal observations.
Usage
SLTCA(
k = 20,
dat,
num_class,
id,
time,
num_obs,
features,
Y_dist,
covx,
ipw,
stop,
tol = 0.005,
max = 50,
varest = TRUE,
balanced = TRUE,
MSC = "EQIC",
verbose = TRUE
)
Arguments
k |
Number of random initialization to start the algorithm. |
dat |
Input data matrix. |
num_class |
Number of latent classes in the fitted model. |
id |
Column name in the data matrix 'dat' for the patient id. |
time |
Column name in the data matrix 'dat' for the time of longitudinal observations. |
num_obs |
Column name in the data matrix 'dat' for the number of longitudinal observations (number of visits). |
features |
A vector of column names in the data matrix 'dat' for the longitudinal observations. |
Y_dist |
A vector indicating the type of longitudinal observations. An element of Y_dist can be 'normal','bin', and 'poi' for continuous, binary and count data. |
covx |
A vector of column names in the data matrix 'dat' for baseline latent class risk factors. |
ipw |
Column name in the data matrix 'dat' for the inverse probability weights for missingness. ipw=1 if not specified. |
stop |
Stopping criterion for the algorithm. stop can be either 'tau' based on posterior probabilities or 'par' based on point estimation. |
tol |
A constant such that the algorithm stops if the stopping criterion is below this constant. |
max |
Maximum number of iterations if the algorithm does not converge. |
varest |
True or False: whether conduct variance estimation or not. |
balanced |
True or False: whether the longitudinal observations are equally spaced. |
MSC |
Model selection criteria: 'AQIC','BQIC' or 'EQIC'. |
verbose |
Output progress of fitting the model. |
Value
A list with point estimates (alpha, beta0, beta1, phi, gamma), variance estimates (ASE), posterior membership probabilities (tau), QICs (qic) of the latent trajectory class model, and stopping criteria (diff) at the last iteration. Point estimates and variance estimates are provided in matrix format, where columns represent latent classes and rows represent covariates or longitudinal features.
Author(s)
Teng Fei. Email: <tfei@emory.edu>
References
Hart, K.R., Fei, T. and Hanfelt, J.J. (2020), Scalable and robust latent trajectory class analysis using artificial likelihood. Biometrics. Accepted Author Manuscript <doi:10.1111/biom.13366>.
Examples
# In this illustrative example the sample size is set as n=50,
# variance estimation is skipped by setting varest=FALSE, and
# the maximum number of iterations is set as max=1 in order to pass CRAN test.
# Please use n=500, varest=TRUE and max=50 for more reliable results.
dat <- simulation(n=50)
res <- SLTCA(k=1,dat,num_class=2,"id","time","num_obs",paste("y.",1:6,sep=''),
Y_dist=c('poi','poi','bin','bin','normal','normal'),
"baselinecov",1,stop="tau",tol=0.005,max=1,
varest=FALSE,balanced=TRUE,MSC='EQIC',verbose=FALSE)