Connectome-Based Modeling of Real World Clinical Outcomes in Addictions

Explore cutting-edge research on predicting addiction outcomes using connectome-based modeling in this 42-minute lecture from the Computational Psychiatry 2020 conference. Delve into Sarah Yip's presentation from Yale University, which showcases how machine learning and predictive modeling techniques can overcome traditional limitations in clinical research. Learn about Connectome-based Predictive Modeling (CPM) and its application in forecasting real-world clinical outcomes for individuals with polysubstance use. Discover how this data-driven approach identifies specific brain networks underlying behavior, predicts cocaine and opioid abstinence, and demonstrates network stability over time. Gain insights into the dissociable anatomical substrates of different substance use types and the potential for translating these findings to clinical settings. Examine the study design, brain state manipulation techniques, and model validation methods used in this groundbreaking research.

Intro
Clinical reality
Neuroimaging of addiction outcomes
Limitations
Machine learning (aka predictive modeling)
Study design
Brain state manipulation improves prediction
Monetary incentive delay task
Model validation - predictive accuracy
Abstinence networks
Short versus long-range connectivity
Post-treatment networks predict abstinence
Cognitive control (Stroop) task
Opioid network connectivity
Theoretical opioid network model
Network identification is brain-state dependent
Cocaine network across drugs and brain states
Opioid network across drugs and brain states
Post-treatment connectivity predicts opioid use
Pathology versus prediction
Theoretical model
Healthy controls
Protracted neural change?
Second external replication
Best' metric depends on the question
Clinical workflow
Elucidation as a goal of prediction