Dollars to donuts..>8 patients in P2a had sleep issues.
Poster from CTAD 2018:
P146: COMPARISON OF SLEEP MEASUREMENTS FROM ACTIGRAPHY TO SELF-REPORTED SLEEP DIARIES. Kirsi Kinnunen1 , Richard Joules1 , Janet Munro1 , Iain Simpson1 , Robin Wolz1,2, Yves Dauvilliers3 ((1) IXICO Plc, London, UK; (2) Imperial College London, London - UK; (3) Sleep Unit, Department Neurology, Centre Hospitalier Universitaire, Montpellier, INSERM 1061 - France) Background: Sleep disturbances are common in people living with dementia. Reduced and fragmented sleep can impact pathology, worsen other symptoms, and significantly affect daytime functioning. Additionally, poor sleep quality may be a risk factor of many neurological conditions, including Alzheimer’s disease (Ju et al. 2014). While polysomnography (PSG) remains the gold standard for sleep assessment, it is costly, often recorded on an isolated night with no daytime assessment, and may not be feasible in many clinical trials. The increasing availability of wearable sensor technology offers a practical, non-intrusive means of collecting continuous sleep-wake data, over several days. This provides “digital biomarkers” from real world settings that can be used to predict risk, detect symptoms and monitor changes in sleep and activity. Objectives: The aim of this study was to examine relationships between actigraphy-derived and self-reported sleep diary measurements within an epidemiology study cohort of elderly subjects. Methods: We analyzed actigraphy and sleep diary data, available for 22 subjects (Age: Mean=80.6, SD=9.9; 68% female) from a study of sleep difficulties, lifestyle factors and general health in the Centre Hospitalier Universitaire, Montpellier - France. All subjects were asked to wear an Axivity 3-axis accelerometry biosensor device (http://axivity.com/) on their non-dominant wrist for 14 nights and to complete sleep diaries for the same 2-week period. IXICO’s fully integrated wearables work-flow (Figure 1) was employed for data analysis. This incorporated quality control (QC) for periods of perceived device “non-wear”. In-bed periods were estimated with the McVeigh algorithm (McVeigh et al. 2016) and sleep and wake periods with the Cole-Kripke (CK) algorithm (Cole et al. 1992). Correlations between the actigraphy- and self-reported sleep measurements were assessed sample-wise, for all nights with both types of data available. As part of a quantitative QC, unrealistic night-time in-bed periods of <2.5 and >13 hours (N=17) were excluded from subsequent analysis, as well as self-reported nights with <1 hour of sleep (N=2). This resulted in 141 nights’ data for the statistical analysis. Correlations (Pearson’s R/Spearman’s Rho) were calculated between four variables available from both actigraphy and the diaries: sleep efficiency/quality, total time in bed, night-time in-bed sleep and number of awakenings. Results: Actigraphy-derived sleep efficiency correlated with self-reported sleep quality at r = .16 (p = .06; Figure 2). McVeigh-estimated times in bed at night and self-reported times from into-bed to out-of-bed correlated at r = .32 (p = .0001). CK-estimated in-bed sleep times and selfreported times from lights-out to awakening (minus sleep onset latency) correlated at r = .15 (p = .07). As expected, the number of estimated night-time awakenings was considerably higher from actigraphy (median=13, IQR=11) than from the diaries (median=2, IQR=2), with these measurements correlated at r = .06 (p = .51). Conclusions: The relatively low correlations between sleep diaries and actigraphy measures reported here may implicate poor reliability and inaccuracy of self-reported sleep measures (see e.g. Goelema et al. 2017). However, as selfreported awakenings may include lying still in bed, the higher number of actigraphy-derived awakenings may be partially explained by interpretation of the minute-wise sleep/no-sleep labeling from the algorithm: a period of wakefulness being split into several if quiet wakefulness is labelled as asleep. With sleep efficiency, we have previously shown in the same cohort that measures derived from 1-night accelerometry recordings using the CK algorithm (as above) correlated at r = .61 (p = .003) with simultaneous PSG (Wolz et al. 2017). Using a welldesigned work-flow, actigraphy can provide objective realworld measures of sleep and activity from data collected over 14 days or longer, with minimal patient discomfort and acceptable QC failure rates. Although PSG can be used to assess sleep stages, apnea/hypopnea indexes and periodic leg movements, the advantages of actigraphy over traditional PSG include the well-tolerated continuous recording in real life settings and relative cost-effectiveness. The current results suggest that compared to sleep diaries, actigraphy can offer an attractive and more reliable alternative for the measurement of signs and symptoms of disease, or the evaluation of therapeutic effects. Customized data analytics, including disease-specific models, have the potential to detect sleep disturbance more accurately than widely used algorithms such as the CK, which may incorrectly interpret periods of quiet wakefulness as sleep. In the Wolz et al. (2017) 1-night study, we found the IXICO Deep Learning Sleep (DLS) algorithm to outperform the CK in estimating PSG-derived sleep efficiency (DLS: r = .84, p < .0001 vs. CK: r = .61, p = .003). Our approach of combining actigraphy with advanced data analytics shows promise for providing improved biomarkers of sleep, circadian rhythm and activity outcomes in clinical trials. In future work, we will employ the DLS algorithm to estimate sleep efficiency in elderly populations and neurodegenerative disease cohorts and will extend the presented statistical analysis to include daytime naps and an assessment of variability between night and day activity/sleep over the 14 day period. References:Cole et al. (1992). Automatic sleep/wake identification from wrist activity. Sleep, 12(5), 461–9. Goelema et al. (2017). Determinants of perceived sleep quality in normal sleepers. Behavioural Sleep Medicine, (20):1-10. Ju, et al. (2014). Sleep and Alzheimer disease pathology--a bidirectional relationship. Nature Reviews Neurology, 10(2), 115–9. McVeigh et al. (2016). Validity of an automated algorithm to identify waking and in-bed wear time in hip-worn accelerometer data collected with a 24 h wear protocol in young adults. Physiological Measurement, 37(10), 1636–52. Wolz et al. (2017). Extracting digital biomarkers of sleep from 3-axis accelerometry using deep learning. The Journal of Prevention of Alzheimer’s Disease, 4(4), P81.
AI
