Nicklas Linz1, Elisa Mallick1, Mario Mina1, Nina Possemis2, Daphne B.G. ter Huurne3, Alexandra König4, Inez H.G.B. Ramakers5 and Johannes Tröger1
1ki elements, Germany. 2Maastricht University Medical Center (MUMC+), Netherlands. 3Alzheimer Center Limburg, School for Mental Health and Neuroscience, Maastricht University, Netherlands. 4National Institute for Research in Computer Science and Automation (INRIA), France. 5School for Mental Health & Neuroscience, Maastricht, Netherlands.
Poster presented at AAIC 2022.
Introduction: Progressive cognitive decline is the cardinal behavioral symptom in most dementia-causing diseases such as Alzheimer’s disease. Whereas classic neuropsychological tests often have excellent psychometric properties to measure cognitive decline in dementia, there are scenarios in which they are less suitable or not applicable at all. Some assessments are not perfectly suitable for decentralized remote clinical trials as they require physical presence of clinicians. Moreover most traditional assessments are unsuited for automated patient-administered screening or pre-screening at low cost to accelerate onboarding. Also maximizing outreach to draw unbiased and representative trial populations beyond established clinical site and hospital networks requires adapted procedures. Speech-based digital biomarkers are less vulnerable to the aforementioned shortcomings, as they can be deployed remotely and extracted in a highly automated fashion allowing solutions to scale. We present the validation of a novel digital Speech Biomarker for Cognition (SB-C) in a Dutch memory clinic population containing patients with Mild Cognitive Impairment (MCI) and age- as well as education-matched participants with Subjective Cognitive Decline (SCD).
Methods: The ki:e SB-C is a hierarchical composite score that combines explainable speech features extracted from the Semantic Verbal Fluency (SVF) and the 15 word Auditory Verbal Learning Test (AVLT; Rey, 1958) and combines them into subdomain scores for episodic memory, semantic memory, executive function and processing speed. Subsequently, subdomain scores are merged into one overall cognition score. The presented validation results are based on a sample that was not used when developing the SB-C. Validation data has been collected from two age- and education-matched groups: Normal Cognition but subjective cognitive decline (SCD; n = 48, 16 F), MCI (n = 48, 18 F); for more specifics in the population see Table 1. We automatically extract the SB-C from SVFand AVLT speech recordings using our proprietary speech analysis pipeline including automatic speech recognition and feature extraction (Tröger et al., 2018). Recordings were collected from a Dutch memory clinic population containing MCI and SCD participants (ter Huurne et al., 2021). We performed (1) analytical and (2) clinical validation. For (1) we performed Spearman rank correlation between SB-C score and anchor score Mini Mental State Examination (MMSE; Folstein et al., 1975) to show that the algorithm is correctly measuring the concept of interest cognition. For (2) we performed a non-parametric Kruskal-Wallis test to compare SB-C scores of both NC and MCI groups.
Results: The biomarker score SB-C and MMSE were strongly correlated (r = 0.55, p < 0.001), such that lower MMSE scores are reflected in lower biomarker scores (compare also figure 1). Additionally, there was a very significant group difference for the SB-C biomarker score between the NC and MCI group (NC > MCI; χ2 = 28.301 (1), p < 0.001).
Conclusion: The ki:e SB-C is a reliable score for cognition, showing convergent validity with the MMSE and separating well between MCI and SCD populations. The availability of valid digital speech biomarkers for cognition has great potential for decentralized clinical trials in dementia aetiologies and beyond.
References:
- Rey, A. (1958). L’examen clinique en psychologie. Paris, France: Presses Universitaires de France.
- Tröger, J., Linz, N., König, A., Robert, P., & Alexandersson, J. (2018, May). Telephone-based dementia screening I: automated semantic verbal fluency assessment. In Proceedings of the 12th EAI International Conference on Pervasive Computing Technologies for Healthcare (pp. 59-66).
- Folstein, M.F., Folstein, S.E., & McHugh, P.R. (1975). Mini- Mental State: a practical method for grading the cognitive state of patients for the clinician. Journal of Psychiatric Research, 12, 189–198.
- ter Huurne, D., Ramakers, I., Possemis, N., Linz, N., Truong, D., König, A., … de Vugt, M. (2021). Remote telephone-based assessment and the additional clinical use of speech features in the Semantic Verbal Fluency task. In Proceedings of the 31st Alzheimer Europe conference—Session P.17 New ways of diagnosing dementia and of recruiting and assessing research participants.