Identifying treatment responders in medication trials for AUD using machine learning approaches

点击翻译按钮获取中文摘要

ABSTRACT Alcohol use disorder (AUD), as defined in DSM-5, represents a highly prevalent, costly, and often untreated condition in the United States. Pharmacotherapy offers a promising avenue for treating AUD and for improving clinical outcomes for this debilitating disorder. While developing novel medications to treat AUD remains a high priority research area, there remain major opportunities to further elucidate clinical response in completed medication trials. To that end, a key question in randomized clinical trials (RCTs) is which patients respond to a given pharmacotherapy. Identifying treatment responders provides major opportunities to advance clinical care for AUD by personalizing medication practices on the bases of variables/predictors of good clinical response. For example, while the effect size for medications such as naltrexone is deemed small-to-moderate, a host of studies over the past decade have shown that its effect size may be considerably larger for certain subgroups of patients. Towards advancing precision medicine for AUD and leveraging data from a host of carefully conducted RCTs for AUD, this R03 application seeks to conduct secondary data analysis. Specifically, we propose to analyze data from four RCTs conducted by the NIAAA Clinical Investigations Group (NCIG). These state-of-the-art RCTs for AUD have tested the following pharmacotherapies: (a) quetiapine, (b) Levetiracetam XR (Keppra XR®), (c) Varenicline (Chantix®), and (d) HORIZANT® (Gabapentin Enacarbil) Extended-Release. In this R03 application, we propose to use a machine learning approach to identify treatment responders in the NCIG RCTs. Machine learning represents a highly promising and underutilized data analytic strategy in the field of AUD treatment response. Machine learning models prioritize the ability to predict future outcomes over creating perfectly fitting models for the data at hand. This results in models which are more generalizable to future observations, which fits well with our goal of identifying responders in RCTs. Leveraging data from these pivotal RCTs through secondary data analysis and using novel analytic methods, namely machine learning, provides a cost-effective approach to identifying AUD pharmacotherapy responders.