Clinical Manifestations

BACKGROUND: With increasing life expectancy, aging-related neurocognitive challenges are becoming more prevalent worldwide. A continuum of severity, from subjective cognitive decline (SCD) to mild cognitive impairment (MCI) and Alzheimer s disease dementia (ADD), is measurable through cognitive and neuroimaging assessments. However, these approaches are costly, subject to scheduling delays, and reliant on specialized personnel or resources, which are scarce in many regions. Automated speech and language analysis (ASLA) offers an affordable, scalable solution for detecting neurocognitive compromise in underserved areas. Yet, no prior study has employed ASLA to predict neuropsychological and brain measures across this continuum in Spanish-speaking Latinos. Our study addresses this gap. METHOD: We recruited 150 Chilean individuals with diverse cognitive profiles: 17 healthy controls, 55 with SCD, 57 with MCI, and 21 with ADD. Participants completed 1-minute phonemic and semantic fluency tasks, alongside cognitive (Addenbrooke s Cognitive Examination-III [ACE-III], Montreal Cognitive Assessment [MoCA]) and executive function (INECO Frontal Screening [IFS]) tests and MRI scans. Machine learning models were trained with word-property and speech-timing features from fluency responses (both separate and combined), derived via the TELL app, to predict cognitive test outcomes, total gray matter (GM) and white matter volumes, hippocampal GM volume, and a mask encompassing ADD-sensitive regions. The best regressors were selected based on the 95% confidence intervals of R2 scores. Pearson s partial correlations between actual and predicted values were computed, controlling for age, sex, and education (and MoCA scores for brain-related measures). Analyses were repeated for each fluency task, their combination, and their average. RESULTS: Significant partial Pearson correlations were obtained for ACE-III (r = .55, p < 0.000001), with no significant effect from digit count alone after adjusting for interaction. Conversely, BDS performance showed a significant negative influence from digit count (p = 0.00858), with numerical magnitude and syllable count nearing significance (p = 0.083 and p = 0.066, respectively). CONCLUSION: Variations in digit span performance across languages illustrate the role of linguistic and numerical factors in cognitive assessments, even with tests targeting non-language domains using digit stimuli. These findings underscore the critical value of language diversity in cognitive research. © 2025 The Alzheimer s Association. Alzheimer s & Dementia published by Wiley Periodicals LLC on behalf of Alzheimer s Association.