Faculty Bibliography

The amount and complexity of patient-level data being collected in randomized-controlled trials offer both opportunities and challenges for developing personalized rules for assigning treatment for a given disease or ailment. For example, trials examining treatments for major depressive disorder are not only collecting typical baseline data such as age, gender, or scores on various tests, but also data that measure the structure and function of the brain such as images from magnetic resonance imaging (MRI), functional MRI (fMRI), or electroencephalography (EEG). These latter types of data have an inherent structure and may be considered as functional data. We propose an approach that uses baseline covariates, both scalars and functions, to aid in the selection of an optimal treatment. In addition to providing information on which treatment should be selected for a new patient, the estimated regime has the potential to provide insight into the relationship between treatment response and the set of baseline covariates. Our approach can be viewed as an extension of "advantage learning" to include both scalar and functional covariates. We describe our method and how to implement it using existing software. Empirical performance of our method is evaluated with simulated data in a variety of settings and also applied to data arising from a study of patients with major depressive disorder from whom baseline scalar covariates as well as functional data from EEG are available.

This paper presents a counterintuitive result regarding the estimation of a regression slope co-efficient. Paradoxically, the precision of the slope estimator can deteriorate when additional information is used to estimate its value. In a randomized experiment, the distribution of baseline variables should be identical across treatments due to randomization. The motivation for this paper came from noting that the precision of slope estimators deteriorated when pooling baseline predictors across treatment groups.

Many techniques of functional data analysis require choosing a measure of distance between functions, with the most common choice being L2 distance. In this article we show that using a weighted L2 distance, with a judiciously chosen weight function, can improve the performance of various statistical methods for functional data, including k-medoids clustering, nonparametric classification, and permutation testing. Assuming a quadratically penalized (e.g., spline) basis representation for the functional data, we consider three nontrivial weight functions: design density weights, inverse-variance weights, and a new weight function that minimizes the coefficient of variation of the resulting squared distance by means of an efficient iterative procedure. The benefits of weighting, in particular with the proposed weight function, are demonstrated both in simulation studies and in applications to the Berkeley growth data and a functional magnetic resonance imaging data set.

We propose a penalized spline approach to performing large numbers of parallel non-parametric analyses of either of two types: restricted likelihood ratio tests of a parametric regression model versus a general smooth alternative, and nonparametric regression. Compared with naively performing each analysis in turn, our techniques reduce computation time dramatically. Viewing the large collection of scatterplot smooths produced by our methods as functional data, we develop a clustering approach to summarize and visualize these results. Our approach is applicable to ultra-high-dimensional data, particularly data acquired by neuroimaging; we illustrate it with an analysis of developmental trajectories of functional connectivity at each of approximately 70000 brain locations. Supplementary materials, including an appendix and an R package, are available online.

A recent meta-regression of antidepressant efficacy on baseline depression severity has caused considerable controversy in the popular media. A central source of the controversy is a lack of clarity about the relation of meta-regression parameters to corresponding parameters in models for subject-level data. This paper focuses on a linear regression with continuous outcome and predictor, a case that is often considered less problematic. We frame meta-regression in a general mixture setting that encompasses both finite and infinite mixture models. In many applications of meta-analysis, the goal is to evaluate the efficacy of a treatment from several studies, and authors use meta-regression on grouped data to explain variations in the treatment efficacy by study features. When the study feature is a characteristic that has been averaged over subjects, it is difficult not to interpret the meta-regression results on a subject level, a practice that is still widespread in medical research. Although much of the attention in the literature is on methods of estimating meta-regression model parameters, our results illustrate that estimation methods cannot protect against erroneous interpretations of meta-regression on grouped data. We derive relations between meta-regression parameters and within-study model parameters and show that the conditions under which slopes from these models are equal cannot be verified on the basis of group-level information only. The effects of these model violations cannot be known without subject-level data. We conclude that interpretations of meta-regression results are highly problematic when the predictor is a subject-level characteristic that has been averaged over study subjects.

Finite mixture models have come to play a very prominent role in modelling data. The finite mixture model is predicated on the assumption that distinct latent groups exist in the population. The finite mixture model therefore is based on a categorical latent variable that distinguishes the different groups. Often in practice distinct sub-populations do not actually exist. For example, disease severity (e.g. depression) may vary continuously and therefore, a distinction of diseased and not-diseased may not be based on the existence of distinct sub-populations. Thus, what is needed is a generalization of the finite mixture's discrete latent predictor to a continuous latent predictor. We cast the finite mixture model as a regression model with a latent Bernoulli predictor. A latent regression model is proposed by replacing the discrete Bernoulli predictor by a continuous latent predictor with a beta distribution. Motivation for the latent regression model arises from applications where distinct latent classes do not exist, but instead individuals vary according to a continuous latent variable. The shapes of the beta density are very flexible and can approximate the discrete Bernoulli distribution. Examples and a simulation are provided to illustrate the latent regression model. In particular, the latent regression model is used to model placebo effect among drug treated subjects in a depression study

Non-specific treatment response, also known as placebo response, is ubiquitous in the treatment of mental illness, particularly in treating depression. The study of placebo effect is complicated because the factors that constitute non-specific treatment effects are latent and not directly observed. A flexible infinite mixture model is introduced to model these nonspecific treatment effects. The infinite mixture model stipulates that the non-specific treatment effects are continuous and this is contrasted with a finite mixture model that is based on the assumption that the non-specific treatment effects are discrete. Data from a depression clinical trial is used to illustrate the model and to study the evolution of the placebo effect over the course of treatment.

Principal points are cluster means for theoretical distributions. A discriminant methodology based on principal points is introduced. The principal point classification method is useful in clinical trials where the goal is to distinguish and differentiate between different treatment effects. Particularly, in psychiatric studies where placebo response rates can be very high, the principal point classification is illustrated to distinguish specific drug responders from non-specific placebo responders

A long-standing problem in clinical research is distinguishing drug treated subjects that respond due to specific effects of the drug from those that respond to non-specific (or placebo) effects of the treatment. Linear mixed effect models are commonly used to model longitudinal clinical trial data. In this paper we present a solution to the problem of identifying placebo responders using an optimal partitioning methodology for linear mixed effects models. Since individual outcomes in a longitudinal study correspond to curves, the optimal partitioning methodology produces a set of prototypical outcome profiles. The optimal partitioning methodology can accommodate both continuous and discrete covariates. The proposed partitioning strategy is compared and contrasted with the growth mixture modelling approach. The methodology is applied to a two-phase depression clinical trial where subjects in a first phase were treated openly for 12 weeks with fluoxetine followed by a double blind discontinuation phase where responders to treatment in the first phase were randomized to either stay on fluoxetine or switched to a placebo. The optimal partitioning methodology is applied to the first phase to identify prototypical outcome profiles. Using time to relapse in the second phase of the study, a survival analysis is performed on the partitioned data. The optimal partitioning results identify prototypical profiles that distinguish whether subjects relapse depending on whether or not they stay on the drug or are randomized to a placebo.