Discovering Control Variables for Maladaptive Drinking Behavior by Analyzing the Geometry of Multi-domain Risk Factors 9/26/14 - 1/15/15

Yale University, New Haven CT

Investigators

Abstract

For the last decade, the National Institute on Alcohol Abuse and Alcoholism (NIAAA) has had an active interest in the issue of how and why behavior change occurs both within and outside the context of professionally-facilitated treatment. NIAAA issued a Broad Agency Announcement (BAA) in 2009, titled, Mechanisms of Behavior Change Initiation (MOBCI) for Drinking (Alcohol) Behavior soliciting proposals to form the Mechanisms of Behavior Change Initiation Research Consortium (MIRC). MIRC represented NIAAA's most progressive effort to supp01i highly innovative, foundational research (using existing data sets to the greatest extent possible) to investigate the causal mechanisms and processes underlying the initiation of behavior change from a maladaptive to a more healthful state. More specifically, NIAAA sought to develop practical models of the alcohol drinking behavior control system (DBCS) that might elucidate how to effect positive behavior change within the spectrum of behavioral modes and define what is achievable within any given individual's local context. Three research contracts were awarded in 2009 and 2010. In May of2010, NIAAA awarded one of these contracts entitled, Discovering Control Variables for Maladaptive Drinking Behavior by Analyzing the Geometry of Multi-domain Risk Factors. The goal of that project was to analyze population-scale prospective data on maladaptive drinking behavior by understanding the behavioral and contextual matrix in which it is embedded. The Contractor developed non-linear dimensional reduction and geometric harmonic analysis techniques to discover implicit (or silent) variables underlying problematic alcohol consumption. The Contractor applied these techniques to two populationscales, prospective epidemiological datasets (National Epidemiologic Survey of Alcohol and Related Conditions; and the National Longitudinal Study of Adolescent Health) that codified trends in behavior (including alcohol consumption), social context, social status, collateral relationships, and many other potential influences on maladaptive drinking behavior. These data were viewed geometrically: the yes/no responses to each question, when bundled into a vector, allowed each subject to be represented as a point in a high-dimensional answer space. Diffusion maps organized the answer space, and allowed the identification of subspaces that reveal commonalities among the subjects and the questions. Such analysis identified highorder and non-linear factors in the data predictive of behavior and behavior change. The information contained in these response matrices involved complex, non-linear combinations of answers not apparent from standard statistical analysis. At the point of contract expiration, the project completed preliminary analysis.