Novel metrics of semantic fluency as preclinical markers of cognitive impairment

Abstract

AbstractBackgroundAccurate identification of the preclinical phase of Alzheimer’s disease (AD) is important for intervention and timely diagnosis. Some researchers suggest that this phase is only detectable using biomarkers, but these are expensive and/or invasive. Because not all biomarker‐positive individuals develop AD‐dementia and cognitive impairment is at the core of this diagnosis, there is a critical need for sensitive, low‐cost, and high‐access cognitive markers. This study investigated how novel item‐level metrics of semantic fluency—naming as many animals in one minute—uniquely relate to 1) memory decline and 2) mild cognitive impairment (MCI).MethodTo investigate the effect of traditional and novel metrics of semantic fluency on both outcomes in a diverse community‐based cohort followed up to 24 years (mean age=75), we used 1) growth curve models of memory decline (n=534, modeled up to 4 visits), and 2) Cox regression models of time to incident MCI (n=349, free of MCI at baseline and >1 visit). Each word’s lexical frequency value—how often a word occurs in daily language—was derived from SUBTLEXus. Models with total number score were compared to two novel metrics: ratio of mean frequency:number and mean frequency of the lowest 10 frequency words. Covariates included age, sex/gender, education, race/ethnicity, recruitment wave, and all other neuropsychological scores.ResultsBoth novel metrics of ratio (standardized B=‐.049, p=.003) and lowest 10 (B=‐.024, p=.003) predicted rate of memory decline over and above other neuropsychological tasks, while total number did not (B=.015, p=.055). The novel metrics were also predictive of time to incident MCI over and above other neuropsychological tasks (HR ratio =1.43, HR lowest 10=1.54, both p<.05), while total number was not (HR=.74, p=.141).ConclusionNovel item‐level metrics of semantic fluency predicted rate of memory decline and incident MCI over and above other neuropsychological tasks, while the traditional fluency score did not. Item‐level data hold a wealth of information that could detect cognitive symptoms in the preclinical phase of dementia beyond our current neuropsychological measures. Future research should investigate more item‐level data to identify optimal metrics that can be used as clinical tools and aid selection of high‐risk individuals for intervention.