UC Berkeley

to Predict Behavioral Outcomes

By

Susan Whitfield-Gabrieli

Doctor of Philosophy in Psychology

University of California, Berkeley

Professor Silvia Bunge, Chair

The emerging field of “neuroprediction” or “predictive analytics” in mental health has promise

for revolutionizing clinical practice by moving towards personalized or precision medicine. The

core idea is that brain measures at a given time may predict individual future behavioral

outcomes, presumably because specific structural or functional brain characteristics constrain the

trajectories of evolving behavior over time. For basic science, discovery of such brain measures

identifies particular neural circuits that constrain specific future behaviors. For clinical science,

such brain measures may support identification of vulnerabilities that could be treated

preventively to minimize poor mental health outcomes.

In this thesis, I first provide an overview of the key questions and challenges in the field

of predictive analytics, aiming to (1) propose general guidelines for predictive analytics projects

in psychiatry, (2) provide a conceptual introduction to core aspects of predictive modeling

technology, and (3) foster a broad and informed discussion involving all stakeholders including

researchers, clinicians, patients, funding bodies and policymakers. Next, I discuss two strategies

for identifying, in a developmental context with children, brain vulnerabilities for future mental

health difficulties. First, I used resting state functional connectivity, measured via functional

magnetic resonance imaging (fMRI), to discover whether children without depression but with

heightened familial risk for major depression disorder (MDD) had brain differences indicative of

risk for depression. At-risk children, compared to children not at familial risk, exhibited

significant differences in functional brain connectivity in three brain networks. Classification

between at-risk versus control children based on resting-state connectivity yielded high accuracy

with high sensitivity and specificity that was superior to traditional clinical rating scales.

Second, I examined whether variation in functional connectivity could predict the trajectory of

clinical symptomology over the ensuing four years in a longitudinal study with a normative child

sample. Variation at age 7 in specific networks predicted individual children’s developmental

trajectories at age 11 towards attentional problems characteristic of Attention Deficit

Hyperactivity Disorder (ADHD) or internalizing problems characteristic of MDD. The predictive

network for internalizing problems was one of the networks that had been atypical in children at

familial risk for MDD. These studies identify variation in brain networks indicative of risk for

two of the most common disorders of adolescent mental health, and suggest that such measures

may support targeted early and preventive interventions. The conclusion of the thesis provides a

discussion of these findings, future directions, theoretical implications, clinical applications and

ethical considerations.