Automated Indices of Clustering and Switching of Semantic Verbal Fluency in Parkinson's Disease.

Abstract

Deficits in semantic verbal fluency (SVF) can stem from dysfunction of an executive control system and/or of semantic knowledge. Previous analyses of SVF responses were devised to characterize these two components including switching and mean cluster size (MCS) indices, but these rely on subjective experimenter-based assessment of the words' relatedness. To address this limitation, computational data-driven SVF indices have been developed. Our aim is to assess the validity and usefulness of these automated indices in the context of cognitive decline in Parkinson's disease (PD). This is a retrospective study including 50 advanced PD patients with (n=28) or without (n=22) mild cognitive impairment (PD-MCI). We analyzed animal SVF outputs using an automated computational approach yielding switching, MCS, and cumulative relatedness (CuRel) indices. We compared these indices to the classic experimenter-based switching and MCS indices to assess concurrent validity, and against neuropsychological measures of executive functioning and semantic knowledge to assess construct validity. We also examined whether these indices were impaired and predicted PD-MCI. Automated switching indices, but not MCS or CuRel, showed evidence of concurrent and construct validity, and characterized individual difference in advanced PD. Automated switching indices also outperformed the experimenter-dependent index in predicting the presence of PD-MCI. Computational methods hold promise as fine-grained, unbiased indices reflecting the executive component of SVF, but none of the methods provided valid measures of semantic knowledge in PD. Our data also confirm that SVF are not adequate tests of semantic memory in patients with executive dysfunction such as PD. (JINS, 2018, 24, 1047-1056).