Multivariate longitudinal data arisen in medical studies often exhibit complex features such as censored responses, intermittent missing values, and atypical or outlying observations. The multivariate-t linear mixed model (MtLMM) has been recognized as a powerful tool for robust modeling of multivariate longitudinal data in the presence of potential outliers or fat-tailed noises. This paper presents a generalization of MtLMM, called the MtLMM-CM, to properly adjust for censorship due to detection limits of the assay and missingness embodied within multiple outcome variables recorded at irregular occasions. An expectation conditional maximization either (ECME) algorithm is developed to compute parameter estimates using the maximum likelihood (ML) approach. The asymptotic standard errors of the ML estimators of fixed effects are obtained by inverting the empirical information matrix according to Louis' method. The techniques for the estimation of random effects and imputation of missing responses are also investigated. The proposed methodology is illustrated on two real-world examples from HIV-AIDS studies and a simulation study under a variety of scenarios.
ShahALairdNSchoenfeldD. A random-effects model for multiple characteristics with possibly missing data. J Am Stat Assoc1997; 92: 775–779.
2.
FieuwsSVerbekeG. Joint modelling of multivariate longitudinal profiles: pitfalls of the random-effects approach. Stat Med2004; 23: 3093–3104.
3.
RoyA. Estimating correlation coefficient between two variables with repeated observations using mixed effects model. Biometric J2006; 48: 286–301.
4.
WangWLFanTH. ECM-based maximum likelihood inference for multivariate linear mixed models with autoregressive errors. Computat Stat Data Analys2010; 54: 1328–1341.
5.
WangWLFanTH. Estimation in multivariate t linear mixed models for multiple longitudinal data. Stat Sinica2011; 21: 1857–1880.
6.
KotzSNadarajahS. Multivariate t distributions and their applications, Cambridge: Cambridge University Press, 2004.
7.
LairdNMWareJH. Random effects models for longitudinal data. Biometrics1982; 38: 963–974.
8.
PinheiroJCLiuCHWuYN. Efficient algorithms for robust estimation in linear mixed-effects model using the multivariate t distribution. J Computat Graph Stat2001; 10: 249–276.
9.
LairdNM. Missing data in longitudinal studies. Stat Med1988; 7: 305–315.
10.
MarshallGDe la Cruz-MesìaRBarònAE, et al.Non-linear random effects model for multivariate responses with missing data. Stat Med2006; 25: 2817–2830.
11.
SchaferJLYucelRM. Computational strategies for multivariate linear mixed-effects models with missing values. J Computat Graph Stat2002; 11: 437–457.
12.
SchluchterMD. Analysis of incomplete multivariate data using linear models with structured covariance matrices. Stat Med1988; 7: 317–324.
13.
WangWL. Multivariate t linear mixed models for irregularly observed multiple repeated measures with missing outcomes. Biometric J2013; 55: 554–571.
14.
HughesJP. Mixed-effects models with censored data with application to HIV RNA levels. Biometrics1999; 55: 625–629.
15.
VaidaFFitzgeraldAPDeGruttolaV. Efficient hybrid EM for linear and nonlinear mixed-effects models with censored response. Computat Stat Data Analys2007; 51: 5718–5730.
16.
DempsterAPLairdNMRubinDB. Maximum likelihood estimation from incomplete data via the EM algorithm (with discussion). J Royal Stat Soc Ser B1977; 39: 1–38.
17.
VaidaFLiuL. Fast implementation for normal mixed effects models with censored response. J Computat Graph Stat2009; 18: 797–817.
18.
MatosLAPratesMOChenMH, et al.Likelihood-based inference for mixed-effects models with censored response using the multivariate-t distribution. Stat Sinica2013; 23: 1323–1345.
19.
WangWLLinTILachosVH. Extending multivariate-t linear mixed models for multiple longitudinal data with censored responses and heavy tails. Stat Meth Med Res2018; 27: 48–64.
20.
LinTILachosVHWangWL. Multivariate longitudinal data analysis with censored and intermittent missing responses. Stat Med2018; 37: 2822–2835.
21.
RubinDB. Inference and missing data. Biometrika1976; 63: 581–592.
22.
Rubin DB. Multiple imputation for nonresponse in surveys. New York, NY: Wiley, 1987.
23.
WangWLLinTI. Multivariate t nonlinear mixed-effects models for multi-outcome longitudinal data with missing values. Stat Med2014; 33: 3029–3046.
24.
WangWLLinTI. Bayesian analysis of multivariate t linear mixed models with missing responses at random. J Stat Computat Simulat2015; 85: 3594–3612.
25.
LiuCRubinDB. The ECME algorithm: a simple extension of EM and ECM with faster monotone convergence. Biometrika1994; 81: 633–648.
26.
MengXLRubinDB. Maximum likelihood estimation via the ECM algorithm: a general framework. Biometrika1993; 80: 267–278.
27.
Ho HJ, Lin TI, Wang WL, et al. R TTmoment package: sampling and calculating the first and second moments for the doubly truncated multivariate t distribution. R Package Version 1.0, http://cran.r-project.org/web/packages/TTmoment/ (accessed 4 May 2015).
28.
LouisTA. Finding the observed information matrix when using the EM algorithm. J Royal Stat Soc Ser B1982; 44: 226–233.
29.
LedermanMMConnickELandayA, et al.Immunologic responses associated with 12 weeks of combination antiretroviral therapy consisting of zidovudine, lamivudine, and ritonavir: results of AIDS clinical trials group protocol 315. J Infect Dis1998; 178: 70–79.
30.
AcostaEPWuHWalawanderA, et al.Comparison of two indinavir/ritonavir regimens in treatment-experienced HIV-infected individuals. J Acquir Immune Deficiency Syndrome2004; 37: 1358–1366.
31.
CoombsRWSpeckCEHughesJP, et al.Association between culturable human immunodeficiency virus type-1 (HIV-1) in semen and HIV-1 RNA levels in semen and blood: evidence for compartmentalization of HIV-1 between semen and blood. J Infect Dis1998; 177: 320–330.
32.
PaxtonWBCoombsRWMcElrathMJ, et al.Longitudinal analysis of quantitative virologic measures in human immunodeficiency virus-infected subjects with ≥ 400 CD4 lymphocytes: implication for applying measurements to individual patients. J Infect Dis1997; 175: 247–254.
33.
HuangYDagneGWuL. Bayesian inference on joint models of HIV dynamics for time-to-event and longitudinal data with skewness and covariate measurement errors. Stat Med2011; 30: 2930–2946.
34.
LinTIWangWL. Multivariate-t nonlinear mixed models with application to censored multi-outcome AIDS studies. Biostatistics2017; 18: 666–681.
35.
WuHKuritzkesDRMcClernonDR, et al.Characterization of viral dynamics in human immunodeficiency virus type 1-infected patients treated with combination antiretroviral therapy: relationships to host factors, cellular restoration, and virologic end points. J Infect Dis1999; 179: 799–807.
36.
ConnickELedermanMMKotzinBL, et al.Immune reconstitution in the first year of potent antiretroviral therapy and its relationship to virologic response. J Infect Dis2000; 181: 358–363.
37.
WuL. A joint model for nonlinear mixed-effects models with censoring and covariates measured with error, with application to AIDS studies. J Am Stat Assoc2002; 97: 955–964.
38.
DagneGAHuangY. Bayesian semiparametric mixture Tobit models with left censoring, skewness, and covariate measurement errors. Stat Med2013; 32: 3881–3898.
39.
Vaida F and Liu L. R lmec package: linear mixed-effects models with censored responses. R package version 3.0.3. 2012.
40.
Pinheiro J, Bates D, DebRoy S, et al. R nlme package: linear and nonlinear mixed effects models. R package version 3.1-104. 2012.
41.
MuñozACareyVSchoutenJP, et al.A parametric family of correlation structures for the analysis of longitudinal data. Biometrics1992; 48: 733–742.
42.
Akaike H. Information theory and an extension of the maximum likelihood principle. In: 2nd International symposium on information theory, Tsahkadsor, Armenia, 2--8 September, 1971, pp.267–281.
43.
SchwarzG. Estimating the dimension of a model. Ann Stat1978; 6: 461–464.
44.
EfronBTibshiraniR. Bootstrap methods for standard errors, confidence intervals, and other measures of statistical accuracy. Stat Sci1986; 1: 54–75.
45.
HoHJLinTIChenHY, et al.Some results on the truncated multivariate t distribution. J Stat Plan Inference2012; 142: 25–40.
46.
LinTIWangWL. Multivariate skew-normal linear mixed models for multi-outcome longitudinal data. Stat Model2013; 13: 199–221.
47.
BandyopadhyayDLachosVHAbanto-VallecCA, et al.Linear mixed models for skew-normal/independent bivariate responses with an application to periodontal disease. Stat Med2010; 29: 2643–2655.
48.
AzzaliniACapitainoA. Distributions generated by perturbation of symmetry with emphasis on a multivariate skew t-distribution. J Royal Stat Soc Ser B2003; 65: 367–389.
49.
SahuSKDeyDKBrancoMD. A new class of multivariate skew distributions with applications to Bayesian regression models. Can J Stat2003; 31: 129–150.
50.
BrancoMDeyD. A general class of multivariate skew-elliptical distribution. J Multivariate Analys2001; 79: 93–113.
51.
MengXLvan DykD. The EM algorithm – an old folk-song sung to a fast new tune. J Royal Stat Soc Ser B1997; 59: 511–567.
52.
DelyonBLavielleMMoulinesE. Convergence of a stochastic approximation version of the EM algorithm. Ann Stat1999; 27: 94–128.
53.
RobinsJMRotnitzkyAZhaoLP. Analysis of semiparametric regression models for repeated outcomes in the presence of missing data. J Am Stat Assoc1995; 90: 106–121.
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
