Digitally Assessed Sleep Disturbances for Early Diagnosis of Alzheimer’s Disease

• Published in: Alzheimer's & dementia Vol. 20; no. S2
• Main Authors: Krix, Sophia, Odeh, Joud, Lentzen, Manuel, Val‐Guardiola, Andrea, Martínez, Neus Falgàs, Atreya, Alankar, Conde, Pauline, Doherty, Aiden, Vairavan, Srinivasan, Muurling, Marijn, de Boer, Casper, Curcic, Jelena, Grammatikopoulou, Margarita, Nikolopoulos, Spiros, Brem, Anna‐Katharine, Coello, Neva, Narayan, Vaibhav, Wittenberg, Gayle, Duijn, Cornelia M Van, Hinds, Chris, Aarsland, Dag, Sanchez‐Valle, Raquel, Fröhlich, Holger
• Format: Journal Article
• Language: English
• Published: Hoboken John Wiley and Sons Inc 01.12.2024
• Subjects: Biomarkers
• ISSN: 1552-5260; 1552-5279
• DOI: 10.1002/alz.088269

Background Alzheimer’s Disease (AD) is associated with sleep disturbances. Moreover, individuals with sleep disturbances have been reported to have a higher risk for developing AD. The measurement of sleep behavior therefore opens the opportunity for a potential digital biomarker of AD. Method We modeled sleep patterns coming from the RADAR‐AD cohort from two sleep monitoring devices (Table 1). We applied a stochastic modeling approach, multi‐state models, and analyzed the times spent in each sleep state before transitioning and transition probabilities in and between sleep states. We further applied statistical analysis of sleep monitoring data and sleep questionnaires (ESS, PSQI) from the RADAR‐AD study (Fitbit Charge 3, DREEM) and preliminary data from the ADIS study (MotionWatch8) (Table 1), using a likelihood ratio test with the aim to assess the diagnostic potential of sleep monitoring devices compared to traditional sleep questionnaires. Result Modeling of digital device data showed that preclinial (preAD), prodromal (proAD) and mild‐to‐moderate (mildAD) AD patients spent more time in the light sleep and awake state, and less time in the REM sleep states compared to healthy controls (HC) before transitioning to the next state, showing non‐linear associations between diagnostic stage and sojourn times (Figure 1A). ProAD and mildAD patients had a higher probability to transit to a light sleep phase compared to HC and to subsequently wake up (Figure 1B). Based on our current and partially still preliminary data, only digital sleep monitoring via Fitbit allowed for a separation between HC and preclinical AD at nominal significance (preAD) but findings were not significant when adjusting for multiple testing (Table 2A). A significant distinction between HC and proAD (p < 0.001) as well as between HC and mildAD (p < 0.01) was possible using Fitbit sleep monitoring data, whereas traditional sleep questionnaires were only able to distinguish HC from mildAD (p < 0.05) (Table 2). Conclusion Sleep patterns assessed via tested digital devices were able to separate proAD from HC – in case of Fitbit ‐ which was not possible by traditional sleep questionnaires. Digital sleep monitoring has thus the potential to support the early diagnosis of dementia.