Personalized relapse prediction in Alcohol Use Disorder

点击翻译按钮获取中文摘要

PROJECT SUMMARY Background: Alcohol use disorder (AUD) has a lifetime prevalence of nearly 30%, with 14.4 million adults in the US currently in need of treatment. Even with treatment, 50-80% of individuals relapse within a year. Mechanisms underlying recovery are still not well understood, specifically individual differences underlying relapse risk. Preliminary data: The work of others and our preliminary data support the involvement of at least three neuro-behavioral mechanisms in the maintenance of AUD: 1) reward reactivity, 2) aversive reactivity and 3) executive control. Using a data-driven machine learning approach in a non-clinical community sample (N=1204; 46% male), we demonstrated that the top predictors of alcohol abuse constituted independent, additive factors of this three-domain model. Our preliminary analyses on subtyping in chronic poly-drug users (N=40; 75% male) and individuals with past AUD (N=74; 32% male), demonstrated that data-driven machine learning approaches can be used to study individual differences in these multi-factorial impairments. We found three distinct ‘subtypes’ in AUD: a “reward drinker” type (increased reward reactivity), a “relief drinker” type (increased aversive reactivity) and a “low functioning drinker” type (low executive control). Goals and Hypothesis: The immediate goal of this project in AUD is to develop a sparse personalized relapse prediction tool that can be employed in a treatment setting to continuously track relapse risk over time. The long-term goal is to determine if relapse prevention interventions can be personalized. The underlying hypothesis is that different combinations of independent factors underlie AUD maintenance and relapse per individual. We will test this by Aim I: determining individual differences in function on three domains (reward reactivity, aversive reactivity, executive control), and Aim II: evaluating the predictive power of subtype- or domain-specific relapse prediction models versus an AUD-general model, to determine their respective clinical utility. Specific Aims: In Aim 1, we will assess individual differences underlying AUD across the three domains of interest using a multi- method approach (personality, neurocognition, clinical assessments, task/resting fMRI brain function) in a large treatment cohort (N=200 AUD, 2-5 weeks into treatment, >40% female; N=100 controls). In Aim 2, we will follow our sample clinically (+6, +12 months) and employ machine learning methods to evaluate if the patterns of impairments underlying relapse risk are distinctly different between individuals. Innovation: This study provides a) a systematic, multi-method assessment of the individual heterogeneity in the neuro-behavioral mechanisms underlying AUD; b) an application of big data analytical approaches for relapse prediction to an AUD dataset; and c) the development of sparse yet highly informative personalized relapse prediction tools. Summary: This study will pave the way for the development of personalized relapse prediction tools that track relapse risk in AUD over time. This can ultimately lead to the development of personalized treatment approaches, with the potential to dramatically transform the current treatment landscape.