A Bayesian approach for sensitivity analysis of incomplete multivariate longitudinal data with potential nonrandom dropout

Longitudinal designs for biomarker traits are very appealing for a number of reasons. First, the collection of repeated biomarker measurements over time allows investigators to characterize both the overall and individual trajectories in these measurements. This helps investigators to understand overall effects as well as how individuals vary around these average effects. Distinguishing individual heterogeneity from average trend is impossible without longitudinal data. This distinction is important for understanding individual between-subject in its own right, as well as in developing predictors of disease outcome from longitudinal marker data. There are many statistical issues that are unique to longitudinal data analysis ranging from correctly specifying realistic models which characterize average changes over time and individual variation, accounting for observations that are taken at unequally spaced observation times which may differ between individuals, to accounting for missing data due to occasionally missed measurements or loss to follow-up. When the reason for missing data is not related to the underlying observation process, most longitudinal data analysis methods can be routinely applied. However, complex modelling is necessary when the reason for missing an observation or dropout is directly related to the underlying outcome process [1–2]. Statistical techniques that appropriately analyze continuous and discrete longitudinal data have been discussed by many authors. Diggle, et al. [3] provides a good summary of these techniques for the practitioner, while Fitzmaurice et al. [4] discusses recent advances from a more technical perspective. The three papers in this section present innovative approaches for solving important methodological problems in biomarker studies. Most standard longitudinal data analysis techniques assume that the measurement process is independent of the outcome process. However, when the outcome is related to the underlying observation process, failure to properly account for the observation process can result in bias [5]. Schildcrout et al.[6] proposed an approach for regression modelling for longitudinal binary data where the outcome is measured with probabilities determined by an auxiliary variable that is related, but not identical to the binary outcome. Modelling multivariate longitudinal biomarker data is an increasingly common issue with the advent of an increasing number of important biomarker techniques (e.g., multiple cytokines in a multiplex assay or multiple reproductive hormones that describe complex biological phenomenon). Although various authors have proposed techniques for analyzing multivariate longitudinal data [7–8 and others], little work has focused on estimating multivariate linear mixed models under known constraints in the relationship between the multivariate outcomes. Roy et al [9] proposes such an approach for analyzing multivariate reproductive hormones over the menstrual cycle, where established relationships have solid biological justification. Finally, the collection of longitudinal biomarker data as disease predictors is becoming an increasingly important design in diagnostic medicine. Pfeiffer et al. [10] proposes flexible non-parametric methodology for combining biomarkers that exploits the longitudinal structure in formulating a score that is used for predicting a binary outcome.

Multivariate binary longitudinal data sets are comprised of repeated observations of a vector outcome and a set of multi-dimensional covariates under each of many independent clusters. The multivariate longitudinal data of this type exhibit two-way correlations:first, at a given point of time, the multivariate responses under a cluster are correlated; second, these correlated responses are repeatedly observed under a cluster over a period of time and the repeated responses are also correlated. One objective is to describe the marginal expectation of the outcome variable as a function of the covariates while accounting for the correlations among the repeated multi-dimensional responses for a given cluster. Based on the assumption that each pair of the multi-dimensional responses in a cluster follow the bivariate beta correlated binary distribution, this paper proposes a general class of estimating equations approach for inference on response probabilities and the scale parameter of the beta correlated binary distribution. The estimating equations have solutions which are consistent and asymptotically Gaussian. We illustrate the use of the estimating equations approach by analyzing a multivariate longitudinal data set on health care utilization by families in the city of St. John's, Canada.

Multivariate longitudinal data are frequently encountered in practice such as in our motivating longitudinal microbiome study. It is of general interest to associate such high-dimensional, longitudinal measures with some univariate continuous outcome. However, incomplete observations are common in a regular study design, as not all samples are measured at every time point, giving rise to the so-called blockwise missing values. Such missing structure imposes significant challenges for association analysis and defies many existing methods that require complete samples. In this paper we propose to represent multivariate longitudinal data as a three-way tensor array (i.e., sample-by-feature-by-time) and exploit a parsimonious scalar-on-tensor regression model for association analysis. We develop a regularized covariance-based estimation procedure that effectively leverages all available observations without imputation. The method achieves variable selection and smooth estimation of time-varying effects. The application to the motivating microbiome study reveals interesting links between the preterm infant's gut microbiome dynamics and their neurodevelopment. Additional numerical studies on synthetic data and a longitudinal aging study further demonstrate the efficacy of the proposed method.

A reanalysis of a longitudinal scleroderma clinical trial using non-ignorable missingness models

Analysis of healthcare utilization, such as hospitalization duration and medical costs, is crucial for policymakers and doctors in experimental and epidemiological investigations. Herein, we examine the healthcare utilization data of patients with systemic lupus erythematosus (SLE). The characteristics of the SLE data were measured over a 10-year period with outliers. Multivariate linear models with multivariate normal error distributions are commonly used to evaluate long series of multivariate longitudinal data. However, when there are outliers or heavy tails in the data, such as those based on healthcare utilization, the assumption of multivariate normality may be too strong, resulting in biased estimates. To address this, we propose multivariate t-linear models (MTLMs) with an autoregressive moving-average (ARMA) covariance matrix. Modeling the covariance matrix for multivariate longitudinal data is difficult since the covariance matrix is high dimensional and must be positive-definite. To address these, we employ a modified ARMA Cholesky decomposition and hypersphere decomposition. Several simulation studies are conducted to demonstrate the performance, robustness, and flexibility of the proposed models. The proposed MTLMs with ARMA structured covariance matrix are applied to analyze the healthcare utilization data of patients with SLE.

Multivariate longitudinal data are used in a variety of research areas not only because they allow to analyze time trajectories of multiple indicators, but also to determine how these trajectories are influenced by other covariates. In this article, we propose a mixture of longitudinal factor analyzers. This model could be used to extract latent factors representing multiple longitudinal noisy indicators in heterogeneous longitudinal data and to study the impact of one or several covariates on these latent factors. One of the advantages of this model is that it allows for measurement non-invariance, which arises in practice when the factor structure varies between groups of individuals due to cultural or physiological differences. This is achieved by estimating different factor models for different latent classes. The proposed model could also be used to extract latent classes with different latent factor trajectories over time. Other advantages of the model include its ability to take into account heteroscedasticity of errors in the factor analysis model by estimating different error variances for different latent classes. We first define the mixture of longitudinal factor analyzers and its parameters. Then, we propose an EM algorithm to estimate these parameters. We propose a Bayesian information criterion to identify both the number of components in the mixture and the number of latent factors. We then discuss the comparability of the latent factors obtained between subjects in different latent groups. Finally, we apply the model to simulated and real data of patients with chronic postoperative pain.

Mixed latent Markov (MLM) models represent an important tool of analysis of longitudinal data when response variables are affected by time-fixed and time-varying unobserved heterogeneity, in which the latter is accounted for by a hidden Markov chain. In order to avoid bias when using a model of this type in the presence of informative drop-out, we propose an event-history (EH) extension of the latent Markov approach that may be used with multivariate longitudinal data, in which one or more outcomes of a different nature are observed at each time occasion. The EH component of the resulting model is referred to the interval-censored drop-out, and bias in MLM modeling is avoided by correlated random effects, included in the different model components, which follow common latent distributions. In order to perform maximum likelihood estimation of the proposed model by the expectation-maximization algorithm, we extend the usual forward-backward recursions of Baum and Welch. The algorithm has the same complexity as the one adopted in cases of non-informative drop-out. We illustrate the proposed approach through simulations and an application based on data coming from a medical study about primary biliary cirrhosis in which there are two outcomes of interest, one continuous and the other binary.

The total score, which is calculated as the sum of scores in multiple items or questions, is repeatedly measured in longitudinal clinical studies. A mixed effects model for repeated measures method is often used to analyze these data; however, if one or more individual items are not measured, the method cannot be directly applied to the total score. We develop two simple and interpretable procedures that infer fixed effects for a longitudinal continuous composite variable. These procedures consider that the items that compose the total score are multivariate longitudinal continuous data and, simultaneously, handle subject-level and item-level missing data. One procedure is based on a multivariate marginalized random effects model with a multiple of Kronecker product covariance matrices for serial time dependence and correlation among items. The other procedure is based on a multiple imputation approach with a multivariate normal model. In terms of the type-1 error rate and the bias of treatment effect in total score, the marginalized random effects model and multiple imputation procedures performed better than the standard mixed effects model for repeated measures analysis with listwise deletion and single imputations for handling item-level missing data. In particular, the mixed effects model for repeated measures with listwise deletion resulted in substantial inflation of the type-1 error rate. The marginalized random effects model and multiple imputation methods provide for a more efficient analysis by fully utilizing the partially available data, compared to the mixed effects model for repeated measures method with listwise deletion.

Mixed-model techniques are applied to incomplete longitudinal data from a double-blind, placebo-controlled, parallel-group, high-dose study of tacrine in patients with Alzheimer's disease. The study consisted of a 30-week double-blind treatment period. Patients were randomized to one of four treatment groups. Dosing was initiated at 40 mg/day and increased in increments of 40 mg/day every 6 weeks until the target dose was achieved. If the study medication was not well tolerated or there were significant elevations in alanine aminotransferase, patients were withdrawn from the study. The use of SAS procedure PROC MIXED for the analysis of incomplete longitudinal data is discussed. This approach is used to evaluate disease progression over time and the effect of incremental dose increases of tacrine on changes on the Alzheimer's Disease Assessment Scale-Cognitive subscale.

Conducting longitudinal research has been encouraged among family scholars as a way to better understand family process and to measure changes in family structure and relationships over time. When doing longitudinal research, a common problem is the loss of sample members between the first wave of data collection and subsequent waves. Some subjects move between data points and cannot be located, whereas others refuse to continue their participation. Some members of the original sample, especially if it includes older adults, will have died or have become too disabled to participate further in the study. These losses represent an attrition of the original sample over time. These various forms of attrition of a sample result in a potential threat of bias if those who drop out have unique characteristics such that the remaining sample ceases to be representative of the original sample. This bias, known as attrition bias, is assumed to exist when there are significant differences between the initial and later samples. Because the loss of sample members between the first wave of data collection and subsequent waves is relatively common, researchers in longitudinal studies should check for attrition bias. Unchecked bias may be the major threat to longitudinal research (Markides, Dickson, & Pappas, 1982; Norris, 1985). Larzelere and Klein (1987) listed sample attrition as one of the three major problems that researchers face in analyzing longitudinal data. Despite the threat of attrition bias, however, it has received little attention in longitudinal family research. Indeed, few family researchers address this issue when reporting the results of their analyses of longitudinal data. There are two ways attrition can bias a sample. Firs, it may alter the characteristics of the sample, making it no longer representative of the original sample. Because the results of later waves of data are not generalizable to the original population that was sampled, the external validity of the study is threatened. For example, a longitudinal study examining the effects of divorce on the mental health of former spouses may fail to retain those subjects who have become too depressed to respond to the questionnaire. The nonparticipation of this subsample could bias the findings towards a minimization of depression as an outcome of divorce. The second way that selective attrition can bias longitudinal data is by altering the covariance of variables (Goudy, 1985; Norris, 1987). This problem occurs when the underrepresentation of some groups in the longitudinal sample leads to correlations between variables that are different than the true correlations in the original sample. Referring back to the previous example, the underrepresentation of depressed people in the second wave of the study may distort the statistical relation between length of marriage and depression. This poses a potentially severe threat to theory building because of the effect of these distortions on internal validity. TESTING FOR ATTRITION BIAS The most common method for detecting attrition bias in the characteristics of the sample is to use t tests to compare those subjects who responded to all waves of the study with those who dropped out after only one wave. The means of important variables from the first wave, such as socioeconomic status (SES), age, marital satisfaction, and depression, are compared between the two samples to determine if the differences are statistically significantly different. Significant differences in the means of one or more variables indicates the existence of attrition bias. Differences in categorical variables, such as race and marital status, can be determined by using the chi-square statistic. Logit analysis is a second method for detecting differences in characteristics of the sample. A logit equation is developed to estimate the probability that each first wave respondent will participate in later waves. …

In longitudinal studies data are collected on the same set of units for more than one occasion. In medical studies it is very common to have mixed Poisson and continuous longitudinal data. In such studies, for different reasons, some intended measurements might not be available resulting in a missing data setting. When the probability of missingness is related to the missing values, the missingness mechanism is termed nonrandom. The stochastic expectation-maximization (SEM) algorithm and the parametric fractional imputation (PFI) method are developed to handle nonrandom missingness in mixed discrete and continuous longitudinal data assuming different covariance structures for the continuous outcome. The proposed techniques are evaluated using simulation studies. Also, the proposed techniques are applied to the interstitial cystitis data base (ICDB) data.

Public management scholars are looking to longitudinal research designs and data to help overcome the many limitations associated with cross-sectional research. However, far less attention has been given to time itself as a research lens for scholars to consider. This article seeks to integrate time as a construct of theoretical importance into a discussion of longitudinal design, data, and public management research. First, I discuss the relative advantages of longitudinal design and data, but also the challenges, limitations, and issues researchers need to consider. Second, I consider the importance of time as a theoretical construct of interest in the pursuit of longitudinal public management research. Third, I offer a brief look at the use of longitudinal design and panel data analyses in the current public management literature. The overview demonstrates a notable absence of public management research considering the attitudes, motives, perceptions, and experiences of individual public employees and managers. Finally, I consider why there are so few longitudinal studies of public employees and point out the issues public management researchers interested in individual employee-level phenomena need to consider when advancing their own longitudinal research designs.

We illustrate a class of conditional models for the analysis of longitudinal data suffering attrition in random effects models framework, where the subject-specific random effects are assumed to be discrete and to follow a time-dependent latent process. The latent process accounts for unobserved heterogeneity and correlation between individuals in a dynamic fashion, and for dependence between the observed process and the missing data mechanism. Of particular interest is the case where the missing mechanism is non-ignorable. To deal with the topic we introduce a conditional to dropout model. A shape change in the random effects distribution is considered by directly modeling the effect of the missing data process on the evolution of the latent structure. To estimate the resulting model, we rely on the conditional maximum likelihood approach and for this aim we outline an EM algorithm. The proposal is illustrated via simulations and then applied on a dataset concerning skin cancers. Comparisons with other well-established methods are provided as well.

The finite mixture model is considered as an appropriate instrument for data clustering.Different parsimonious multivariate mixture distributions are introduced for skewed and/or heavy-tailed longitudinal data.The eigenvalue or modified Cholesky decomposition of covariance matrices develops the families of parsimonious mixture models.Thus, the finite mixture of matrix-variate t-distributions for clustering a three-way dataset with heavy-tailed or outlier observations (e.g., multivariate longitudinal data) is more appropriate compared to matrix-variate normal distributions.Accordingly, the present study considered a parsimonious family of the finite mixture of matrix-variate t-distributions using the eigenvalue and modified Cholesky decomposition for within and between covariance matrices, respectively.Finally, parameter estimates were calculated using the expectation-maximization algorithm, and simulations studies and real data analyses were conducted to confirm the obtained results.

Covariance miss-specification and the local influence approach in sensitivity analyses of longitudinal data with drop-outs