YSPH Biostatistics Seminar: “Employing Machine Learning Models to Predict Substance Use among Adolescents"

NOTE: BIS 525 students are required to attend in person. Others are invited to attend in person, but may also attend via Zoom.

SPEAKER: Uche Aneni, MD, MHS, Assistant Professor of Child Psychiatry and Biomedical Informatics and Data Science, Yale University

TITLE: “Employing Machine Learning Models to Predict Substance Use among Adolescents"

ABSTRACT: Most adolescents in need of substance use treatment do not receive it. Adolescents with unhealthy substance use are at elevated risk of developing a substance use disorder, dying from an overdose or developing other mental and physical health problems. Thus, early identification of adolescents with unhealthy substance use can lead to timely interventions. However, barriers to screening for unhealthy substance use preclude treatment access. These barriers include lack of time for screening, provider discomfort, and lack of privacy in healthcare settings. Automating the screening process using digital tools such as games or large datasets from sources such as the electronic health record may mitigate these barriers. Games collect considerable data during play that may be used for prediction. Performance in a game is captured by in-game metrics such as reaction time, speed of task completion, or choice move in a game. These in-game metrics may reflect digital biomarkers of health outcomes. Digital biomarkers are “physiological and/or behavioral measures generated by persons that may explain, influence, or predict health outcomes.” Game performance is also influenced by cognitive processes such as working memory, inhibitory control, and decision making. These cognitive processes are implicated in the development of unhealthy substance use and, in turn, are impacted by unhealthy substance use. As such, in-game metrics may represent digital biomarkers of cognitive processes that predict unhealthy substance use. Similarly, publicly available data and vast amounts of data from the electronic health record contain physiologic and behavioral measures that may be indicative of substance use. As such these data can be leveraged to develop predictive models for substance use identification among adolescents. This talk will discuss findings from two studies that developed a predictive machine learning model for unhealthy substance use among adolescents using in-game data from an existing videogame and data from a nationally representative sample of adolescents with implications for substance use prevention among adolescents.

YSPH values inclusion and access for all participants. If you have questions about accessibility or would like to request an accommodation, please contact Charmila Fernandes at Charmila.fernandes@yale.edu. We will try to provide accommodations requested by September 26, 2024.