Zampeta-Sofia Alexopoulou, Louisa Schwed, Johannes Tröger, Nicklas Linz, Qingyue Li, Stefanie Köhler, Josef Priller, Annika Spottke, Bjoern Falkenburger, Gabor C. Petzold, Jens Wiltfang, Inga Zerr, Frank Jessen, Anja Schneider, Emrah Düzel, Michael Wagner, Christoph Laske, Stefan J. Teipel, Valeria Manera, Alexandra König
*Poster presented at AAIC 2026
Background: Digital speech-based assessments demonstrate cross-sectional sensitivity to cognitive impairment in Alzheimer’s Disease. Evaluating their ability to capture longitudinal trajectories over short follow-up intervals is crucial. Here, we examined longitudinal change captured by a speech-based composite of global cognition (SB-C) and domain-specific
scores (memory/executive function/processing speed) compared to gold-standard paper-
pencil measures.
Methods: Within PROSPECT-AD, data were drawn from the German DELCODE/DESCRIBE cohorts (baseline: N = 29 healthy (HC), N = 112 subjective cognitive decline (SCD), N = 67 mild cognitive impairment (MCI)). Participants completed the same automated remote phone- based speech assessment every three months over 15 months, alongside in-clinic evaluations at baseline and 12 months. SB-C scores were derived using a validated pipeline (ki:elements). Longitudinal SB-C trajectories were examined using linear mixed-effects models (LMMs) with time as primary predictor and diagnostic group/clinical severity (CDR-Global) as stratification variables, adjusting for covariates and participant-specific random effects. Estimated slopes and least-squares (LS) mean changes from baseline were derived. Longitudinal change in clinical measures was assessed using Wilcoxon tests and LMMs.
Results: At baseline, SB-C was significantly lower in MCI compared to HC (β =−0.24, p<0.001). HC and SCD did not differ. Longitudinally, SB-C showed a significant time effect (β=0.066, p=0.022), with improvement in HC and SCD, but not in MCI. LS-mean changes confirmed sustained improvement in HC and SCD (15-month Δ=0.08, p<0.001) (Fig. 1). Domain-specific SB-C scores exhibited similar patterns. Beyond diagnosis, less impaired participants at baseline (CDR = 0) exhibited significant longitudinal SB-C improvement (Fig.2), reflecting potential cognitive gains. In contrast, paper-pencil measures showed limited longitudinal sensitivity, with small effect sizes (Fig. 3), potentially due to lower temporal resolution.
Conclusion: SB-C captured longitudinal cognitive dynamics, including early improvement in less impaired individuals, that were less consistently detected by clinical measures. We propose speech-based assessments as sensitive, low-burden tools for cognitive monitoring in early-stage impairment and for contexts requiring fast/frequent assessments.