Pause-related speech biomarkers measure cognitive impairment in MS

Simona Schäfer 2, Susett Garthof 1 , Julia Elmers 1, Louisa Schwed 2, Nicklas Linz 2, Tina Boggiano 3, Christopher Chatham 3, James Overell 3, Helen Hayward-Koennecke 3, Johannes Tröger 2, Anja Dillenseger 1, Christian Beste 1, Björn Tackenberg 3, Tjalf Ziemssen 1

1) Universitätsklinikum Dresden. 2) ki elements GmbH. 3)  F. Hoffmann LaRoche AG.

* Poster presented at ACTRIMS 2024, Florida, USA



Cognitive impairment is a highly prevalent symptom in people with Multiple Sclerosis (pwMS) and has a significant impact on activities of daily living and quality of life. Cognitive impairment might appear at all stages of MS and is present in all MS phenotypes. Monitoring cognitive functions amongst other symptom groups is important in pwMS given the highly heterogeneous disease trajectories of pwMS. Digital speech biomarkers represent a great opportunity to monitor cognitive impairment in pwMS remotely and at low patient burden.


To evaluate how speech biomarkers could be used to detect cognitive impairment in pwMS, especially highly-prevalent domains, such as processing speed and executive functions.


A sample of n = 169 German speaking participants (76 pwMS, 93 healthy controls (HC)) underwent a test battery, which contained several free speech tasks, and the SDMT. Temporal as well as acoustic aspects of speech such as duration of pauses or pitch were extracted from three different free speech tasks (telling a negative and positive story, as well as picture description) and correlated with the SDMT performance. Moreover, the ki:elements speech biomarker for cognition (ki:e SB-C), a composite score derived from a 15-min long speech assessment, was compared between healthy controls (HC), and pwMS.


In pwMS, in both the picture description and the negative storytelling task, the pause rate correlated significantly with the SDMT (r = -.200, p < .05, d = 0.41 & r = -.291, p < .05, d = 0.61). Additionally, in both tasks the standard deviation of pause length correlated significantly with the SDMT (negative storytelling: r = -.210, p < .05, d = 0.43, picture description: r = -.293, p < .01, d = .61). Some acoustic features, including deviations in jitter and formants all correlate with the SDMT in all three tasks. PwMS had significantly lower scores than HC in the ki:e SB-C composite score for overall cognition (H = 5.04, p = .01, d = 0.24) and for the subscore of executive functioning (H = 6.86, p = .01, d = 0.29).


Pause-related speech features significantly correlated with the SDMT in pwMS, indicating that a higher number of pauses and a higher variability of pause length is associated with reduced cognitive functioning. This might reflect a deficit in executive functioning as pauses are typically made between single content units and a reduced planning ability is likely to lead to increased pause frequency. Overall, this analysis demonstrates that low-burden speech biomarkers from naturalistic free speech tasks that can be collected remotely could be used for screening and monitoring of cognitive impairment in pwMS.

Share this article