MCMC-based estimation methods for continuous longitudinal data with non-random (non)-monotone missingness
Introduction
Clinical studies typically entail the evaluation of a particular response of interest over a period of time. Often, the longitudinal nature renders the data arising from such studies more susceptible to incompleteness, primarily in the response, the extent and nature of which can have serious implications on the resulting conclusions. Any meaningful analysis should therefore attempt to address the incompleteness in the data. With the recent developments in the area of missing data, it has been shown that ignoring the missing data (i.e., analyzing only the complete cases) would be valid only under the most restrictive, unrealistic assumptions (Rubin, 1976). A more principled approach should account for, rather than ignore, the mechanism driving the missingness, , in addition to that driving the responses, , which can be partitioned into its observed and missing parts as . Such an approach would thereby entail working with the joint distribution, , where and are the parameter vectors characterizing the response and non-response processes, respectively.
The missingness or non-response process, , can be broadly classified as being either monotone (or dropout), in which the unobserved measurements within a longitudinal series all occur after a particular measurement occasion, or non-monotone, for which missing values arise intermittently within the series. In addition, the missingness process, be it monotone or not, can be classified using the taxonomy introduced by Rubin (1976). A mechanism is said to be missing completely at random (MCAR) if the processes governing the missingness and the outcomes are independent, possibly conditionally on covariates. A less rigid assumption would be missing at random (MAR), for which the missingness may depend on the observed outcomes and on covariates but, given these, not further on the unobserved outcomes. When, in addition to such dependencies, the unobserved data provide further information about the missing data mechanism, then the mechanism is referred to as being missing not at random (MNAR).
When data are incomplete, likelihood-based approaches involve working with the observed data likelihood, which is obtained by integrating the full data likelihood over the missing components. As shown by Rubin (1976), the resulting expression simplifies under MCAR and MAR to and inferences about can be drawn independently of the missing data model. This property, termed ignorability allows for the use of standard software, provided it can handle unbalanced data. The MNAR case, however, does not admit simplification and integration over the missing components is explicitly necessary. Depending on the complexity of the model used, not only is the approximation/evaluation of the integral itself involved, but its subsequent maximization can be computationally demanding.
Broadly speaking, the evaluation and subsequent maximization of the observed data likelihood under non-ignorable missingness may be approached using either a stochastic or a non-stochastic solution. Non-stochastic solutions would involve maximization of either a direct evaluation or a numerical approximation of the integral in going from to . Diggle and Kenward (1994) provide such a solution for continuous Gaussian data with monotone missingness, while Troxel et al. (1998) propose a method for the non-monotone case. The latter authors combined a Gaussian model with first-order anti-dependence structure for the outcomes with logistic regressions for the non-response indicators, further assuming independence among the latter. And yet despite these simplifying, but somewhat restrictive, assumptions, Troxel et al. (1998) reported that computations can become quite complicated and intractable for more than three or four repeated measurements. Stochastic methods, on the other hand, involve (iteratively) simulating random draws from the underlying non-standard distribution via Markov chains to fill in the missing data, and subsequently maximizing the likelihood using the completed data. For this reason, the stochastic approach offers a less computationally demanding alternative because it precludes any numerical integration.
This paper focuses on the use of the stochastic expectation–maximization (SEM) algorithm (Celeux and Diebolt, 1985) for fitting selection models for continuous longitudinal data with MNAR missingness, be it monotone or non-monotone, with subsequent comparison of resulting stochastic solutions with their non-stochastic counterparts. Through a simulation study (Section 5), the merits of the stochastic approach over non-stochastic methods are highlighted, thereby emphasizing the value of the former as a practical alternative to the latter. The stochastic EM algorithm was further applied to fit models for two case studies (described in Section 2 and the results of which are presented in Section 6). The various modeling frameworks considered are described in Section 3, while estimation procedures are presented in Section 4. Some concluding remarks are given in Section 7.
Section snippets
Age-related macular degeneration trial
The first case study data were obtained from a randomized multicenter clinical trial comparing an experimental treatment (interferon-) with a corresponding placebo in the treatment of patients with age-related macular degeneration (ARMD) (Pharmacological Therapy for Macular Degeneration Study Group, 1997), a condition under which patients progressively lose vision. In the trial, a patient’s visual acuity was assessed at 4, 12, 24, and 52 weeks by his/her ability to read lines of letters on
Model frameworks
Suppose that for subject , a sequence of measurements is designed to be measured at time points , resulting in a vector of measurements for each participant. Further, for each measurement in the series, define the response indicators as: if is observed and otherwise, which are organized into a vector of parallel structure to . Typically, can be partitioned into two sub-vectors: consisting of those
Estimation methodology: The stochastic EM algorithm
The expectation–maximization (EM) algorithm (Dempster et al., 1977), which is a general purpose iterative algorithm for calculating maximum likelihood estimates, has become quite popular in handling incomplete data problems. The fundamental idea behind the EM algorithm is to associate with the given incomplete data problem, a complete data problem for which maximum likelihood estimation is computationally more tractable. The algorithm alternates the following 2 steps. The expectation or E-step
Simulation study
To evaluate the performance of the SEM algorithm in fitting non-ignorable models for incomplete Gaussian longitudinal data, simulations were conducted for both the monotone and non-monotone situations. Data generating models, simulation settings and subsequent results under each case are described in the following subsections.
SEM results were compared with those obtained using Newton–Raphson with numerical integration, which will hereinafter be referred to as the direct likelihood approach. The
ARMD: monotone case
A subset of the ARMD data consisting of 226 subjects having either a complete set of responses or dropout type of missingness was analyzed. For the 4 repeated measurements, a saturated means model was formulated, with the treatment (interferon- vs. placebo) received by the patient, denoted , as covariate, and with an unstructured covariance, i.e., for and . For dropout, a model of the form (1) was considered. The
Discussion
In this paper, the performance of the stochastic EM algorithm in fitting non-ignorable selection models for continuous longitudinal data with monotone or non-monotone missingness was assessed. Stochastic EM is a convenient alternative to the deterministic EM algorithm because, by replacement of the E-step of the latter with a simulation step under the former, integration of the complete data likelihood over the missing values is altogether avoided since the S (simulation) step renders the data
Acknowledgement
The authors gratefully acknowledge financial support from the Interuniversity Attraction Pole Research Network P6/03 of the Belgian Government (Belgian Science Policy).
References (26)
- et al.
Analysis of longitudinal data with intermittent missing values using the stochastic EM algorithm
Computational Statistics and Data Analysis
(2006) - et al.
The nature of sensitivity in missing not at random models
Computational Statistics and Data Analysis
(2006) - Celeux, G., Chauveau, D., Diebolt, J., 1995. On stochastic versions of the EM algorithm. Technical Report 2514. INRIA...
- et al.
The SEM algorithm: a probabilistic teacher algorithm derived from the EM algorithm for the mixture problem
Computational Statistics Quarterly
(1985) - et al.
A stochastic approximation type EM algorithm for the mixture problem
Stochastics and Stochastics Reports
(1992) - et al.
The common structure of several recent statistical models for dropout in repeated continuous responses
Statistical Modelling
(2002) Global cross-ratio models for bivariate, discrete, ordered responses
Biometrics
(1986)- et al.
Maximum likelihood from incomplete data via the EM algorithm (with discussion)
Journal of the Royal Statistical Society, Series B
(1977) - et al.
Stochastic EM: method and application
- et al.
Informative dropout in longitudinal data analysis (with discussion)
Applied Statistics
(1994)