Summary
Research suggests that early identification of individuals at clinical high risk (CHR) for psychosis may be
able to improve illness course. Studies suggest that early identification of CHR using specialized interviews
with help-seeking individuals (with attenuated psychosis symptoms) is a useful approach. This work has two
major limitations: 1) interview methods have limited specificity as only 20% of CHR individuals convert to
psychosis, and 2) the expertise needed to make CHR diagnosis is only accessible in a few academic centers.
We propose to develop a new psychosis symptom domain sensitive (PSDS) battery, prioritizing tasks that
show correlations with the symptoms that define psychosis and are tied to the neurobiological systems and
computational mechanisms implicated in these symptoms. To promote accessibility, we utilize behavioral tasks
that could be administered over the internet; this will set the stage for later research testing widespread
screening that would identify those most in need of in-depth assessment. To reach that goal we first need
determine which tasks are effective for predicting illness course and how this strategy compares to published
prediction methods. We propose to recruit 500 CHR participants, 500 help-seeking individuals, and 500
healthy controls across 5 sites with the following Aims: Aim 1A) To develop a psychosis risk calculator through
the application of machine learning (ML) methods to the measures from the PSDS battery. In an exploratory
ML historical analysis, we will determine the added value of combining the PSDS with self-report measures and
predicators;Aim 1B) We will evaluate group differences on the risk calculator score and hypothesize
that the risk calculator score of the CHR group will differ from help-seeking and healthy controls. We further
hypothesize that the risk calculator score of the CHR converters will differ significantly from groups of CHR
nonconverters, help-seeking and healthy controls. The inclusion of a help-seeking group is critical for
translating the risk-calculator into clinical practice, where the goal is to differentiate those at greatest risk for
psychosis from those with other forms of psychopathology; Aim 1C): Evaluate how baseline PSDS
performance relates to symptomatic outcome 2 years later examining: 1) symptomatic worsening treated as a
continuous variable, and 2) conversion to psychosis. We hypothesize that the PSDS calculator: 1) will predict
symptom course and, 2) that the differences observed between converters and nonconverters will be larger on
the PSDS calculator than on the NAPLS calculator. Aim 2) Use ML methods, as above, to develop calculators
that predict: 2A) social, and, 2B) role function deterioration, both observed over two years. Because negative
are more strongly linked to functional outcome than positive symptoms, we predict that negative mechanism tasks will be the strongest predictor of functional decline in both domains.This project will provide a next-generation CHR battery, tied to illness mechanisms and powered by cutting-edge computational methods that can be used to facilitate the earliest possible detection of psychosis risk.