Item-Level Analysis of Category Fluency Test Performance: A Systematic Review and Meta-Analysis of Studies of Normal and Neurologically Abnormal Ageing

Abstract

While Category Fluency (CF) is widely used to help profile semantic memory, item-level scoring (ILS) approaches to this test have been proposed to obtain indices that are less influenced by non-semantic supportive functions. We systematically reviewed the literature to test the hypotheses that (1) compared with healthy adults, individuals with a clinical diagnosis suggestive of neurodegeneration generate words of lower semantic complexity; (2) compared with young adults, older adults generate words of higher semantic complexity. We searched six databases (date of search: 8 December 2023) for studies that relied on CF and ILS methods, in normal ageing and in age-associated neurodegeneration. Thirty-four studies were shortlisted: 27 on neurodegenerative conditions; 7 on normal ageing. Risk of bias was evaluated via a published checklist. Data were presented via qualitative synthesis. Most studies reported words of lower semantic complexity in relation to at least one item-level feature in individuals with mild cognitive impairment (MCI), Alzheimer's dementia (AD), and other neurodegenerative diseases. Post-hoc meta-analyses focussing on the MCI/AD continuum confirmed an effect on words' frequency (385 MCI/AD individuals and 350 controls; Hedges's G = 0.59) and age-of-acquisition (193 MCI/AD individuals and 161 controls; Hedges's G = - 1.51). Studies on normal ageing, conversely, failed to demonstrate any overall effect. Most studies on MCI and AD have not relied on neurobiological diagnostic criteria. Moreover, only a small number of studies analysed ILS controlling for quantitative CF performance. Despite these two limitations, this study suggests that ILS can contribute to an in-depth characterisation of semantic memory in neurological ageing.

Keywords: Dementia; Item-based; Qualitative scoring; Semantic complexity; Semantic fluency; Semantic memory.

Fig. 1

Literature search flowchart

Fig. 2

Count of studies that have investigated each item-level semantic feature and each clinical diagnosis

Fig. 3

Effect of a clinical MCI-AD diagnosis on average item-level CFT words’ features. Studies based on MCI/AD vs. controls between-group differences only are reported. While significant and non-significant effects are reported in green and red, respectively, yellow cells indicate “incomplete” significance, as follows: * the group difference emerges in relation to the principal component on which the feature loads; ** the group difference emerges in relation to rating-based scoring, not test-based scoring; *** the group difference emerges only when the feature is scored for the CFT and Letter Fluency Test combined; **** the group difference emerges when the feature is scored in relation to two of the three CFT categories (but not in relation to the third one). Abbreviations: AOA, age of acquisition; CON, concreteness; DIV, semantic diversity; FAM, familiarity; FRQ, frequency; IMG, imageability; OCC, relative occurrence; TYP, typicality; VAL, valence

Fig. 4

Forest plots summarising the effect of clinical diagnosis (i.e. AD/MCI vs. controls) on item-level scores. Effect sizes calculated from between-group comparisons of frequency scores are positive as MCI/AD participants tend to generate words of higher frequency than controls. Effect sizes calculated from between-group comparisons of age-of-acquisition score are negative, as MCI/AD participants tend to generate words of the earlier age of acquisition than controls. W indicates the proportional weight of each study

Fig. 5

Linear association between the average educational attainment calculated in the group of participants with MCI/AD (i.e. moderator in the meta-analysis of age of acquisition values) and study effect size. Individual studies are numbered: (1) Venneri et al. (2011); (2) Forbes-McKay et al. (2005); (3) Sailor et al. (2011); (4) Wakefield et al. (2018); (5) Venneri et al. (2008); (6) Henderson et al. (2023); (7) Paek and Murray (2021); (8) Won et al. (2021)