Abstract
Individuals at clinical high risk for psychosis (CHR) are cognitively and neurobiologically heterogeneous, which encourages the use of a clustering approach to parse this heterogeneity. Multimodal approaches are assumed to be superior to unimodal approaches in identifying subgroups. With the success of the use of cognition and electrophysiological measures collectively in established psychotic disorders, and the lack of such an approach in CHR, we were motivated to address this gap. Using the North American Psychosis-Risk Longitudinal Study (NAPLS) 2 consortia (CHR (N=764)), we applied unsupervised cluster analysis on the combined cognitive and electrophysiology measures to identify CHR subgroups and assess their relationship with clinical and functional outcomes. A two-cluster solution with modest separability was found, which prompted the use of an alternative probabilistic, rather than discrete, clustering approach. Individuals who were more likely to be in Cluster 1 exhibited poorer cognitive performance, larger N100, mismatch negativity, and P300 amplitudes, and worse functioning, as well as a younger age of onset. These findings were largely replicated in NAPLS 3 (CHR (N=628)). Taken together, the results of our previous study of cognition-only clustering and the current study of combining cognition and electrophysiology indicate that multimodal clustering, if not developmentally informed, may obscure meaningful subtyping.