Network analysis reveals specific patterns of relationships between connected speech features in subjective cognitive decline

Subtle language changes occur early in Alzheimer's disease (AD). Early detection of these changes therefore represents a potential tool for the early detection of individuals who could develop the disease. Knowledge on connected speech (CS) in Subjective Cognitive Decline (SCD, a preclinical stage of AD for at least some individuals, characterized by cognitive complaints without objective cognitive deficits), remains limited. In this study, we used network analysis to examine the relationships between CS features extracted with Natural Language Processing techniques. This method allowed us to determine whether the patterns emerging from the SCD CS network (i.e., a network modelling the relationships between CS features) differed from those observed in similar networks constructed for controls and individuals with Mild Cognitive Impairment (MCI, a prodromal stage of AD for many individuals, characterized by objective cognitive deficits but preserved daily functioning). We found minimal differences between individuals with SCD and controls and individuals with MCI in raw CS features. Nevertheless, our network analysis revealed a stronger relative subdivision of the SCD network into different dimensions (representing linguistic domains), as well as relative reductions in within-dimension density and between-dimension strength compared to the Control and MCI networks. There were also relative differences in how the CS features forming the Semantic conciseness and Lexical and syntactic complexity dimensions associated with other features in the SCD network. Results suggest that network analysis might be more sensitive than other methods to subtle CS differences in SCD and that this population might use distinct CS production mechanisms.