Empirically derived “discourse phenotypes” are associated with neuropsychological performance, cognitive status, and global atrophy: A latent profile analysis of connected speech measures

Abstract

AbstractBackgroundEvidence suggests that “connected speech” (i.e., spoken language, as in discourse) is related to early cognitive decline and early Alzheimer’s disease and related dementias (ADRD), yet performance on speech tasks and cognitive performance in ADRD are notably heterogeneous. Here we examined whether unique spoken language phenotypes could be ascertained using latent profile analysis (LPA), and we examined the association of identified clusters to demographic, clinical and cognitive characteristics. In the subsets with amyloid PET and structural MR, we analyzed cluster associations with estimated amyloid DVR and/or global atrophy at or near the speech sample.Method1,039 participants from the Wisconsin Registry for Alzheimer’s Prevention (WRAP) completed comprehensive neuropsychological testing and provided a spoken language digital recording describing the Cookie Theft picture from the Boston Diagnostic Aphasia Examination. Cognitive status was determined by consensus conference. Five speech‐language metrics based on previous literature (Table 1) were automatically extracted from transcriptions of recordings utilizing Computerized Language Analysis (CLAN), including: semantic content, sentence length, speaking rate, and fluency. Baseline speech metric Z‐scores were included in the LPA; model fits were evaluated using BIC criteria. T‐tests, chi‐square tests, and ANCOVAs were used to the compare demographic, clinical, cognitive and imaging characteristics of the resulting discourse phenotypes (alpha = .05).ResultFit indices suggested a 3‐profile model: normal content‐normal fluency (n = 558), low content‐normal fluency (n = 224), and low content‐low fluency (n = 257) (Figure 1). The three phenotypes differed on gender and cognitive status, with more men and a higher proportion of MCI in the low content‐low fluency phenotype than the other groups; the phenotypes also differed on concurrent neuropsychological tests of language, memory and executive function. In the subset with imaging, the low content‐low fluency phenotype showed the most global atrophy. There were no differences in concurrent estimated amyloid accumulation (Table 2, Figure 2).ConclusionThree phenotypes of narrative discourse were identified and were associated with multiple neuropsychological tests, overall cognitive status and global brain atrophy. While connected speech metrics are generally viewed as a promising ADRD digital biomarker, the underlying taxonomy identified here suggests a more precision‐based approach is possible.