Actigraphy in the digital health revolution: still asleep?

The introduction of actigraphy in the early 1990s initiated a new era in sleep monitoring. Actigraphy established itself as the accepted alternative to the gold-standard polysomnography (PSG) method for sleep assessment for non-laboratory settings, allowing “prolonged” recordings of sleep in the participant’s natural environment. Actigraphy devices are small accelerometers, usually housed in watches, which assess sleep by analyzing the individuals’ pattern of motion (with the assumption that periods of no motion indicate sleep). Certainly, their introduction led to several advancements, allowing typical sleep to be recorded for weeks at a time in a variety of populations (e.g. patient groups, children, older adults) in an ecologically valid environment, which would not be feasible at the same scale using PSG. Furthermore, actigraphy has made sleep research more accessible for the wider scientific community, allowing sleep to be investigated more broadly, raising the profile of sleep research and the vital role sleep plays in health, developmental, and psychological outcomes. Actigraphy also allows clinicians to gain valuable insights into their patients sleeping patterns across time, particularly when needing to identify the presence of a circadian rhythm disorder. However, despite its wide use, several known pitfalls still exist for this technology, which are often overlooked or underestimated. After more than 20 years of actigraphy, the majority of validation studies, including both healthy participants or a variety of patient groups, still report that actigraphy’s ability to classify PSG wake (specificity) is less than 50%. Consequently, “motionless wake” and “sleep with motion” cannot be reliably estimated, a particular concern when using actigraphy to study patient populations (e.g. insomnia, apnea, restless legs syndrome) in which periods of wakefulness are higher and motion abnormalities may be present during sleep. This is particularly troublesome when one considers that although relatively few studies have validated actigraphy in clinical populations, actigraphy is commonly used in clinical research and practice. However, low specificity is often minimized or overlooked, and actigraphy is assumed to be a reliable estimate of PSG. This assumption is partly due to findings that PSG and actigraphic derived measures, such as total sleep time (TST) and composed indices like sleep efficiency (SE), are often similar despite epoch-by-epoch analyses showing low specificity. Actigraphy is particularly good at detecting sleep epochs, and since the percentage of sleep is relatively high within a normal night of sleep, with few wake epochs to detect, global parameters are minimally affected by low specificity. Indeed, several studies have identified that as sleep periods contain more wake, not only is specificity poorer, but actigraphic derived measures, such as TST, also become less accurate [1], e.g. overestimating TST by 1–1.5 hours on average [2]. Interestingly, Marino, Li [3] reports that when WASO is greater than 30 minutes, for every minute increase in WASO, actigraphy underestimates PSG by 0.93 minutes. Unreliable estimation of wake during the rest period affects other global sleep parameters, e.g. SE, TST, SOL. Therefore, actigraphy’s low specificity continues to be an ongoing issue for studies utilizing the technology as a measure of sleep, especially in populations with motion abnormality. Actigraphy’s specificity issue is further exacerbated by the absence of comprehensive guidelines for choosing set recording modalities (e.g. thresholds for immobility, sensitivity). This becomes more troublesome when one considers how appropriate settings may differ across different devices, algorithms and participant groups and results in an ongoing trade-off between sensitivity (ability to detect true sleep) and specificity. More sensitive thresholds require smaller activity counts to deem an epoch as wake, which increases specificity but at the cost of sensitivity, conversely, less sensitive thresholds increase sensitivity at the cost of specificity, due to the greater activity count threshold required for wake. There is also no consensus as to whether a discrepancy is clinically-significant: despite often non-significant differences between PSG and actigraphy measures, the absolute differences, e.g. a PSG-actigraphy WASO discrepancy of ~40 minutes [1], may be deemed clinically significant. The question remains whether researchers should aim for high overall accuracy and sensitivity and acknowledge that sleep is overestimated, or, aim to more accurately detect wake at the cost of sleep. Some of the issues surrounding actigraphy are likely driven by the fact that actigraphy is limited by an intrinsic hardware constraint, i.e. the use of a single channel of information (motion). In contrast, PSG characterizes sleep using multiple data sources (electro-cortical, -cardiac, -ocular, and -myographic activity). The categorization of wake and sleep by actigraphy is determined by the specific algorithm applied to the actigraph data, which is typically provided by the manufacturer. Alternatively, publicly available algorithms can be applied to the data (e.g. Cole–Kripke and Sadeh). However, public algorithms are not integrated into existing software and even when algorithms have been shown to be less affected by wake (e.g. regression algorithms [2]) they have not been widely evaluated or adopted. Instead, researchers apply settings recommended by the manufacturer, despite them not necessarily being appropriate for their sample. However, even with further algorithm development, given the current actigraphy devices, accuracy and specificity are unlikely to be improved significantly without hardware development (i.e. multiple channels of data). Therefore, despite a clear need for standardization in methodology and use of objective criteria in validating actigraphy, an important question remains: will actigraphy ever be able to achieve satisfactory performance in wake detection? Although the limitations of actigraphy have long been acknowledged, the digital health revolution has emphasized them, as new consumer wearable devices offer alternative and technologically advanced methods for detecting sleep (e.g. integrating multiple bio-signals with motion). Although wearable devices still require validation, researchers and clinicians are increasingly adopting them for measuring sleep. Given recent advances, it seems timely to question whether current actigraphy devices can maintain their superiority and dominance within the sleep field or whether technological advances are beginning to render actigraphy outdated. Actigraphy’s usefulness in the field of sleep is unquestionable and has the advantage of collecting data for weeks without requiring any user involvement. However, over the past twenty years, there have been very few advances in improving its major flaw (low specificity). It seems that the field is struggling to keep pace with the digital health revolution and availability of increasingly more sophisticated commercial multisensory wearables at a lower price, able to measure sleep/wake patterns and now claiming to differentiate sleep stages. Many researchers have acknowledged that poor specificity is actigraphy’s major flaw. However, this seems to have become a standard statement to include in the study limitations, rather than a real concern to be addressed. Given the advances in the burgeoning field of wearable technology, it begs the question “what is the future for actigraphy”? M.dZ. and F.C.B. have received research funding unrelated to this work from Ebb Therapeutics Inc., Fitbit Inc., and International Flavors & Fragrances Inc. Conflict of interest statement. None declared.