Key points are not available for this paper at this time.
BackgroundAlzheimer's disease (AD) lacks effective disease-modifying therapies and scalable, ecologically valid biomarkers to monitor treatment response. Transcranial pulse stimulation (TPS) is an emerging non-invasive neuromodulation technique with potential to attenuate cognitive decline. Sensitive digital endpoints are needed to quantify intervention-related changes. ObjectiveTo develop and validate connected-speech-derived digital biomarkers as a longitudinal framework for monitoring TPS treatment response in AD. MethodsIn this open-label, single-arm pilot study, 32 patients with AD were compared to cognitively healthy controls. A three-stage framework was implemented: (1) machine-learning classification using linguistic features to derive a parsimonious biomarker panel; (2) construction of a Speech Composite Index (SCI) calibrated against the CERAD total score (CTS) ; and (3) longitudinal SCI tracking in a sub-cohort receiving TPS. ResultsThe classifier discriminated AD from controls with an AUROC of 0. 879 and an F1-score of 0. 825. The SCI showed strong convergent validity with global cognition (CTS: r = 0. 76, p < 0. 001; MMSE: r = 0. 76, p < 0. 001) and executive function (Stroop interference: r = -0. 51, p = 0. 015). Longitudinal modeling demonstrated a significant positive deviation from a CERAD-based progression reference (βₜime = 0. 057 z-units/month, p = 0. 013), indicating relative stabilization of speech performance. Individual trajectories were heterogeneous (range -0. 053 to +0. 336) without significant demographic associations. ConclusionsConnected-speech-derived digital biomarkers can serve as scalable longitudinal endpoints for neuromodulatory interventions in AD. The SCI captures treatment-related dynamics and may support response stratification. Further validation in larger, sham-controlled multicenter studies is needed to establish clinical utility and specificity to TPS.
Wiechmann et al. (2026) studied this question.