Status: Active

Despite decades of mental health research, accurately predicting psychiatric outcomes for individuals remains elusive.  Recent advances in artificial intelligence (AI) are making this goal more attainable, and the potential public health significance of these efforts cannot be overstated.  Accurate individualized predictions could lead to a transformative shift in psychiatric services, particularly if predictive capacities can point to actionable, preventive efforts that circumvent negative outcomes.

Given the ongoing mental health crisis among youth, there is a critical need for tools that can predict which youth are most likely to develop psychiatric illnesses. This predictive capacity would allow care systems to optimize their allocation of mental health services, targeting youth most likely to develop a mental illness without misdirecting resources to those unlikely to become ill.  This transformation of care would have significant impact if predictive capacities were built on measures that are easy to acquire, affordable, and – most critically – able to be equitably implemented across demographics. Moreover, while predictive capacities would be of benefit across the lifespan, the impact is amplified for youth and emerging adults. 

Adolescence is a key developmental period when not only is there a sharp rise in mental illness but also curtailing its onset could have lifelong implications.  Preventing a teenager from developing a mental illness could spare that youth the negative effects of psychiatric illness that often ripple through adulthood.

With these goals in mind, our research team leveraged advances in artificial intelligence (AI) coupled with a large open-access dataset, the Adolescent Brain and Cognitive Development (ABCD) study. We developed the Duke Predictive Model of Adolescent Mental Health (Duke-PMA), a predictive tool capable of predicting which youth are likely to become high risk for mental illness—such as moving into the highest percentile of psychiatric symptomatology—one year after an assessment.

Research supported by: Internal funding