Growth modeling with nonignorable dropout: Alternative analyses of the STAR*D antidepressant trial.

Abstract

This paper uses a general latent variable framework to study a series of models for non-ignorable missingness due to dropout. Non-ignorable missing data modeling acknowledges that missingness may depend on not only covariates and observed outcomes at previous time points as with the standard missing at random (MAR) assumption, but also on latent variables such as values that would have been observed (missing outcomes), developmental trends (growth factors), and qualitatively different types of development (latent trajectory classes). These alternative predictors of missing data can be explored in a general latent variable framework using the Mplus program. A flexible new model uses an extended pattern-mixture approach where missingness is a function of latent dropout classes in combination with growth mixture modeling using latent trajectory classes. A new selection model allows not only an influence of the outcomes on missingness, but allows this influence to vary across latent trajectory classes. Recommendations are given for choosing models. The missing data models are applied to longitudinal data from STAR*D, the largest antidepressant clinical trial in the U.S. to date. Despite the importance of this trial, STAR*D growth model analyses using non-ignorable missing data techniques have not been explored until now. The STAR*D data are shown to feature distinct trajectory classes, including a low class corresponding to substantial improvement in depression, a minority class with a Ushaped curve corresponding to transient improvement, and a high class corresponding to no improvement. The analyses provide a new way to assess drug efficiency in the presence of dropout.

Growth mixture modeling with non‐normal distributions

“…Because of the general implementation in the Mplus software, other extensions are available as well. For example, a survival part can be used to model non-ignorable dropout as in [11], and categorical and count variables can be included in the model with parameters varying as a function of the latent trajectory classes for the continuous repeated measures. § …”

“…Growth mixture modeling was introduced in Verbeke and LeSaffre [3] and Muthén and Shedden [4] with related developments in Nagin and Land [5] and Roeder et al [6]. Following this, many extensions and applications have been presented such as Lin et al [7] considering prostate-specific antigen (PSA) biomarker trajectories with irregularly scheduled observations, Lin et al [8] adding joint estimation of survival with prostate cancer, Muthén and Brown [9] considering causal inference in randomized trials of antidepressants with placebo effects, Muthén and Asparouhov [10] adding general multilevel growth mixture modeling, and Muthén et al [11] modeling non-ignorable dropout in antidepressant trials. For overviews of methods with illustrations by a variety of applications, see [10,12].…”

Being “Them” and “Us” at the Same Time? Subgroups of Cultural Identification Change Among Adolescent Diaspora Immigrants

“…We addressed this issue by handling missing data points by full information maximum likelihood estimation as is recommended for structural equation modeling (e.g., Enders, 2006;Schafer & Graham, 2002). More importantly, we showed in additional analyses (see Enders, 2011;Muthén et al, 2011) that results on subgroups of identification change can be supposed to be unbiased by the presence and the patterning of missing points in our data set. We therefore assume that our results are robust.…”

Analysis of Multivariate Longitudinal Data Subject to Nonrandom Dropout

“…It expresses the dynamic nature of the latent attitude variable (Dunson, 2003;Cagnone et al, 2009) and accounts for the serial correlation in it in a form where the latent variable at time point 3, say, is only related to that measured at time 1 via the latent variable at time 2. Another alternative specification would be a random effects model in which a random intercept and possibly a random slope affect the time-dependent latent variables as in a standard growth mixture model for observed repeated measures; for example, see Muthén and Masyn (2005) and Muthén, Asparouhov, Hunter, and Leuchter (2011). However, this type of model is not considered here.…”