Novel computational measure of semantic fluency performance associated with first-episode of psychosis

Abstract

The semantic fluency task is easy to administer and predictive of multiple pathologies. The variables used to analyze its output, however, suffer from 1) high correlations with the task's primary variable (list length: number of domain-specific words), 2) the need to exclude large portions of data, or 3) the inability to conduct participant-level analyses. In this article, we exploited network science methods to create a novel computational measure that avoided all three drawbacks. A semantic network, in which words are connected based on their inter-similarity, was constructed using the lists of words verbally produced by 314 healthy participants in the animal fluency variant of the semantic fluency task. From this network we derived jump probability, which entails the probability that a given participant's fluency list makes random jumps within the semantic network. A correlation analysis found that jump probability did not significantly correlate with list length, unlike the alternative measures clustering and switching. A linear regression found that jump probability significantly predicted whether fluency lists originated from healthy controls (N = 154) or patients with a first episode of psychosis (N = 144). Further, a patient-level analysis revealed a significant interaction between positive symptoms and diagnosis. Patients given a non-affective schizophrenia diagnosis with greater severity of positive symptoms tended to produce fluency lists of higher jump probability. Jump probability is a promising new method of assessing semantic processing among patients on the schizophrenia spectrum.