Machine learning-assisted precision mental health for cigarette smoking cessation

点击翻译按钮获取中文摘要

Project Summary The goal of this fellowship application is to support and facilitate the necessary training for the applicant to develop an independent research career in precision mental health for substance use disorders. Her long- term program of research will 1) identify and harness data sources that may allow for differential treatment selection, 2) build machine learning models to select among treatments, 3) bring these models forward for use in clinical practice, and 4) use these models to inform the development of novel treatments to fill gaps in care. Through the proposed training goals, guided mentorship, and complementary experiences, this fellowship will strategically advance the applicant’s career. She will increase her knowledge of substance use disorders and treatments. She will gain expertise in advanced quantitative methods including feature engineering, genetic analyses, and machine learning. She will complete her training with the skills to conduct and disseminate interdisciplinary research poised to further precision mental health research for substance use disorders. The proposed project seeks to apply machine learning to precision mental health for cigarette smoking cessation. Precision mental health is the application of the precision medicine paradigm to mental health conditions. Precision mental health guides treatment selection using individual differences characteristics likely to predict treatment success. Several factors have hindered progress towards successful treatment selection via precision mental health. First, traditional analytic techniques have been insufficient to account for the real- world complexities that underlie treatment response and recovery. Second, precision mental health models are built and evaluated in the same sample and consequently do not generalize to new data (i.e., new patients). Third, precision mental health research has rarely included genetic features (predictors) alongside and in interaction with clinical (non-genetic) data, preventing integration of knowledge across domains. Contemporary machine learning approaches are well-suited to address these limitations. Machine learning models can accommodate high-dimensional arrays of features across data sources, extract reliable prediction signals that generalize robustly to new samples of patients, and incorporate genetic and non-genetic features simultaneously. The proposed project will apply machine learning to the precision mental health paradigm for cigarette smoking cessation. Cigarette smoking remains a critical and costly public health crisis for which existing treatments are only moderately effective at best. The proposed project will produce a model that can guide treatment selection among several first-line (i.e., FDA-approved) smoking cessation medications. Successful application of the precision mental health paradigm to cigarette smoking cessation would have immediate clinical impact by accelerating and optimizing therapeutic benefit. It would also serve as a template for how to improve treatment outcomes across substance use disorders.