ki:elements

Tracking Cognitive Change over Time Using a Remote Telephone-Based AI-Driven Digital Cognitive Assessment

Elisa Malick, Johannes Tröger, Stefanie Köhler, Alexandra König, Anika Wuestefeld, Erik Stomrud, Pontus Tideman, Oskar Hansson, Sebastian Palmqvist, Stefan Teipel, and Nicklas Linz

*Presented at CTAD 2025

Background: Remote digital cognitive assessments allow more frequent monitoring and may detect changes earlier than traditional in-clinic evaluations. The ki:elements Speech Biomarker for Cognition (SB-C) is a low-burden, remote, and cost-efficient tool for tracking cognitive changes across the Alzheimer’s disease (AD) spectrum. It combines an automated landline telephone protocol with automated analysis of two established speech-based tasks: the Rey Auditory Verbal Learning Test (RAVLT) and Semantic Verbal Fluency (SVF).

In this study, we investigated whether remote longitudinal administration of the SB-C over landline telephone calls could track cognitive changes over time in a multi-site Swedish-German memory clinic population.

Methods: We examined whether longitudinal SB-C assessments track cognitive change over time in two PROSPECT-AD memory clinic cohorts: the Swedish BioFINDER-Primary Care (N=103, mean age=76.6 ± 6.4) and German DESCRIBE-AD registry (N=177, mean age=72.9 ± 7.4). Participants completed automated telephone-based assessments at baseline (t0) and 3, 6, and 12 months (t3, t6, t12). SB-C scores were derived from RAVLT and SVF speech recordings using ki:elements’ proprietary pipeline with automatic speech recognition and feature extraction. The resulting global cognition z-scores were corrected for age and education.

Participants were classified at t0 as healthy/subjective cognitive impairment (HC/SCI) or mild cognitive impairment (MCI). Across timepoints: t0 HC/SCI=169, MCI=111; t3 HC/SCI=115, MCI=52; t6 HC/SCI=147, MCI=75; t12 HC/SCI=109, MCI=50. Change in SB-C was computed as t0 minus later timepoints, and linear mixed-effects models tested effects of group, time, and their interaction, with separate group comparisons per timepoint.

Results: We observed a significant group effect (F=11.13; p<.001), with the MCI group showing greater decline in cognitive performance compared to HC/SCI across timepoints. Group differences in absolute SB-C scores were significant at all timepoints except t3. A significant group-by-time interaction (F=4.78; p<.01) indicated that cognitive decline in the MCI group was more pronounced over time. No significant main effect of time was detected independently of the group.

Conclusion: These findings demonstrate that the SB-C, administered remotely via automated telephone calls, is sensitive to longitudinal cognitive changes, particularly in individuals with MCI. The SB-C differentiates between clinical groups and captures progression over time, supporting its potential as a scalable, low-burden tool for remote cognitive monitoring in both research and clinical settings. This approach could facilitate future AD clinical trials by improving the efficiency of longitudinal tracking of cognitive changes.

Share this article

Discover more articles