Efficacy in linguistic characteristics: I‐CONECT project

Abstract

AbstractBackgroundResearchers have documented ample evidence suggesting progressive decline in linguistic and syntactic abilities due to cognitive decline. Among measures of syntactic complexity derived from natural language processing (NLP) approaches, we hypothesized that the number of coordinated phrases (CPs, an indicator of sentence complexity) would improve from an intervention which intended to enhance cognitive function by providing social interactions through frequent video‐chats.MethodIn the I‐CONECT project, the experimental group engaged in a high dose of semi‐structured conversations (four times per week) via video‐chats, with additional weekly 10 minutes telephone check‐ins. The control group received only the weekly telephone check‐ins. We analyzed language samples derived from the weekly phone check‐ins (WPCs) in this study. WPCs mainly ask about social activities engaged by the participant during the previous week with some semi‐structured open‐ended questions. Out of 186 subjects who were randomized to the trial, 113 subjects had at least one recording of WPCs with sufficient recording quality during the 2nd, 3rd and 4th weeks of the intervention (baseline) and during the 3 weeks (i.e., 22, 23, 24th) before the month 6 (M6) assessments. We extracted the number of CPs from WPCs of 52 experimental (25 MCI, 27 normal) and 61 control (34 MCI, 27 normal) participants. We regressed CP scores at M6 on age, gender, years of education, the experimental (vs. control) group, controlling for baseline CP scores using a linear regression model.ResultAmong the MCI group, the experimental group’s CP scores were 0.60 points higher than the control group (p = 0.046) at M6, which is equivalent to over 10 years of age difference in CP scores (i.e., ‐0.055×10 = ‐0.55), controlling for covariates listed above (see Figure 1). Among those with normal cognition, we did not find any significant differences between the two groups at M6 (p = 0.414).ConclusionWe found that an indicator of sentence complexity improved among the experimental group compared with the control group among the MCI participants. Our results indicated that language measures are capable of tracking clinically meaningful changes in language samples capitalizing their potential use in clinical trials as automated outcome measures.