Abstract
Changes in sleep during mid-to-late life are associated with risk for Alzheimer's disease (AD). Mechanistic understanding of this association necessitates measurement tools able to quantify these sleep changes longitudinally and accurately. We conducted a systematic review with meta-analysis of validity studies of non-invasive sleep-measuring devices published since 2015 that record sleep metrics associated with AD in adults over 40 (mean 52.9, SD 6.1 years). We reviewed 52 studies, including 32 wearable and ten non-wearable single or multi-sensor devices validated against polysomnography (minimum one night). The apnoea hypopnoea index and oxygen desaturation index were accurately measured across devices. Total sleep time and sleep efficiency were significantly overestimated (p < 0.001) by mean 33.2 minutes and 7.6%, respectively. Slow wave sleep duration was inaccurately measured except by a headband device with electroencephalography. There was no significant difference in accuracy between participants with and without sleep disorders. Studies were undermined by high risk of bias from closed-access algorithms and classification thresholds, and incomplete reporting of accuracy data. Only one study investigated slow wave activity, and none investigated sleep spindles. Nonetheless, we have identified devices that could be used in future studies of sleep and AD risk and discuss some of the limitations of available research.