Semantic fluency predicts survival of memory clinic patients

Semantic fluency is commonly assessed in the diagnostic work-up of individuals with suspected cognitive impairment due to neurodegenerative disease. Semantic fluency has a predictive value for survival in clinical AD and in healthy elderly individuals, but it is unknown if this also applies to biomarker-confirmed AD and non-AD memory clinic patients. Potential associations with and added value of imaging biomarkers of amyloid pathology and neurodegeneration have yet to be explored.

From our clinical registry, we included patients who were assessed at a memory clinic with the neuropsychological assessment battery of the Consortium to Establish a Registry for Alzheimer's Disease (CERAD-NAB) and whose vital status could be retrieved in 07/2024. We tested the association of semantic fluency performance at first assessment and over time (and, for comparison, further CERAD-NAB subtest scores) with mortality risk using age-adjusted single-predictor Cox proportional hazard models. Amyloid status (positive vs negative on clinical PET reads) and global cognitive impairment (MMSE) were included as covariates. In addition, we explored associations between semantic fluency performance and both regional cortical glucose metabolism (FDG PET) and global amyloid load (centiloids; PiB PET). Finally, the predictive value of global amyloid load and glucose metabolism in comparison to and in combination with semantic fluency performance was assessed.

583 patients were included (age 68.9 ± 8.7, 45% female). 280 patients (48%) had died and 303 (52%) were alive after a median of 8.0 years [95% C.I. 7.7-8.4]. Better semantic fluency (age-, sex- and education adjusted Z score, based on normative data) was significantly associated with lower mortality risk (HR = 0.71 [0.63 - 0.80], Bonferroni-corrected p < 0.001). Its predictive value was higher than that of all other CERAD-NAB subscores (e.g., memory, visuospatial abilities). Semantic fluency remained a significant predictor when amyloid status and global cognitive impairment were accounted for (HR = 0.75 [0.61 - 0.91], p = 0.0035). In patients with more than one assessment (N = 163), longitudinal change of semantic fluency (derived from a linear mixed effects model) was also a significant predictor (HR = 0.66 [0.53-0.83], p < 0.001). In 218/583 patients who had received amyloid and FDG PET, worse semantic fluency was associated with decreased FDG uptake of left inferior and middle temporal, dorsolateral frontal, and posterior parietal cortical regions (Bonferroni-corrected p < 0.05), but not with amyloid load. FDG uptake of the left IFG (pars opercularis) was itself a predictor of survival (HR = 0.68 [0.55 - 0.84], Bonferroni-corrected p < 0.05), but to a lesser degree than semantic fluency (and MMSE, naming and figure drawing). Predictive accuracy of semantic fluency was further improved by including FDG uptake of the right anterior cingulate (Bonferroni-corrected p = 0.078). Global amyloid load was not associated with survival.

Survival of memory clinic patients can be predicted by semantic fluency, independently from amyloid status and global cognitive impairment. Anterior cortical glucose metabolism is itself a significant predictor of survival and slightly improves prediction by semantic fluency.