mHealth in Mental Health: Challenges and Opportunities
(Samprit Banerjee): Smartphones provide an interactive interface that can passively measure various aspects of the user’s behavior from device sensors, as well as actively collect self-ratings (e.g. mood, stress etc.) obtained via daily ecological momentary assessment. Taken together with traditional clinical assessments, these measures have the potential to provide unique insight into the outcome trajectory of patients with major depressive disorder undergoing treatment. The potential to predict patient compliance and adherence to treatment in clinical trials of psychotherapy is a necessary step to modify future sessions in order to improve compliance and therefore efficacy. There are unique challenges of such predictions due to the noisy nature (missing or under-reporting) of passive and actively mHealth data. This talk will demonstrate some of these problems and provide strategies and solutions to handle them.
Discovering responders of psychiatric interventions using LGMM in the presence of missing data
(Jihui Lee, Elizabeth Mauer, Yiyuan Wu): Clinical trials in mental health research often involve testing the effect of an intervention on longitudinal outcomes of symptoms (e.g. depressive symptoms). There are two methodological issues that are common to such trials - one, there is a high degree of missing data in the longitudinal outcomes and two, not all participants respond equally to the intervention. In this tutorial we focus on these two issues and demonstrate methodological solutions to overcome these challenges. First, it is imperative to obtain an overall estimate of treatment effect that may be biased due to the presence of missing outcomes that are non-ignorable or missing not at random. To this end, we present random-effects pattern-mixture models (PMM) that has the ability to test for missing data mechanism that depend on the dropout pattern and correct the bias in the estimated treatment effect that is introduced due the presence of such non-ignorable missing data. Second, even if there is a significant treatment effect, not all participants respond equally to the treatment. Therefore it is necessary to understand identify sub-groups of patients who do not respond to treatment so that future treatment can be personalized and future interventions can be tailored to such sub-groups. To this end, we demonstrate the use of Latent Growth Mixture Models (LGMM) that cluster trajectories of treatment response into sub-groups that show different patterns of response (e.g. early response, non-response, late response). We also address key methodological concepts such as choosing the number of subgroups, modeling time in different polynomial terms, and extensions to non-continuous outcomes. Finally, we demonstrate how one can utilize baseline clinical and demographic characteristics of participants to predict these sub-groups show different patterns of response.
Causal Mediation Analysis for Stochastic Interventions
(Iván Díaz): Mediation analysis in causal inference has traditionally focused on binary exposures and deterministic interventions, and a decomposition of the average treatment effect in terms of direct and indirect effects. We present an analogous decomposition of the population intervention effect, defined through stochastic interventions on the exposure. Population intervention effects provide a generalized framework in which a variety of interesting causal contrasts can be defined, including effects for continuous and categorical exposures. We show that identification of direct and indirect effects.
Analyzing Medical Big Data: Challenges and Applications
(Wenna Xi, Jie Xu, Zhenxing Xu, and Yufang Huang) The recent advancement of technology and computational power have encouraged more researchers to analyze big data in fields of biomedical sciences. Medical big data are commonly used for predictive modeling, disease clustering, and providing clinical decision support. Analyzing medical big data is a valuable field which requires talents with knowledge of both machine learning/statistics and health, as well as a deep understanding of its challenges -- although a promising field, analysis of medical big data suffers from typical big data issues, such as the trade-off between data quality vs. data quantity, as well as inherent limitations, such as being observational studies by design. Common methods of analyzing medical big data include regression, clustering, and classification. In this talk, we will first discuss the promises and challenges of analyzing medical big data. Then, we will take electronic health record (EHR) data as an example and illustrate some applications of analyzing medical big data.