Measuring sleep has not been easy for those in the sleep medicine community. Current methods include polysomnography (PSG), sleep diaries and questionnaires. PSG is expensive, as it measures brain wave rhythms, muscle activity and eye movements simultaneously. Sleep diaries are inconvenient, while questionnaires may be inaccurate. Actigraphy has been proposed as the solution to these downsides. This review explores how actigraphy works, and how it compares with the abovementioned methods when used to measure sleep characteristics. An actigraphy device tracks movements to determine if the patient is awake or asleep through algorithms. Mathematical relationships of various strengths between values from actigraphy and those from PSG, diaries and questionnaires, but the two sets of values were generally close to each other. Differences between actigraphical values and PSG could be due to the patient being awake but not moving, which can be corrected for; while differences between actigraphical values and subjective measures could be due to human error. Although actigraphy does have the potential to replace current methods of measuring sleep, the choice to use it has to be after careful consideration of its practical benefits versus its slight inaccuracies.
Sleep measurement is historically challenging. Polysomnography (PSG) is expensive, as it is composed of simultaneous electrooculography, electromyography and electroencephalography at the minimum. Subjective measures like sleep logs are labour-intensive and questionnaires are prone to bias and human error. Actigraphy has been proposed as the solution to these downsides. This review discusses the technology behind actigraphy, its estimation of sleep phases, and summarises literature on the agreement between actigraphy and the abovementioned methods for the measurement of sleep parameters, namely total sleep time, sleep efficiency, wake after sleep onset and sleep onset. Agreement is considered in 2 ways: correlation coefficients and measurement errors from PSG and subjective measures. In general, correlation coefficients were variable, but the magnitudes of error had debatable clinical significance. Measurement errors between actigraphy and PSG may be attributed to motionless wakefulness, a systematic error which can be corrected for, after extensive research on the mathematical discrepancy. Measurement errors between actigraphy and subjective measures, however, are attributed to human error and hence not always predictable. Therefore, the use of actigraphy as a replacement for either PSG or subjective measures needs to be after careful consideration of the balance between slight technological inaccuracy and practical non-clinical benefits.
Sleep in itself is not well understood, and the attempts to investigate the normal physiology of sleep have been limited by this same lack of understanding. The 3 main methods of studying sleep are polysomnography (PSG), subjective measures (i.e. sleep diaries and questionnaires) and actigraphy. PSG consists of simultaneous electroencephalogram (EEG), electrooculogram (EOG) and electromyogram (EMG) measurements, and is considered the “gold standard” for the measurement of sleep parameters1. Actigraphy is an emerging technology, which estimates sleep phases using motion detection2.
Actigraphy has been proposed as a cost-efficient and convenient alternative, overcoming the downsides of the expensive and uncomfortable PSG, the labour-intensive sleep diaries and unreliable sleep questionnaires. However, since sleep has generally been studied based on its neurophysiological characteristics, actigraphy, which purely measures movement, is merely an estimation. Therefore, this review will discuss actigraphy and its validity in the measurement of sleep parameters, as compared to PSG and subjective sleep methods, in order to confirm its suitability.
2.0 Technology of actigraphy
Actigraphy is the study of physical activity using movement detectors (e.g. pedometers and accelerometers). Current actigraphy devices contain multiple accelerometers that measure motion in multiple axes, taking data samples several times per second3. These accelerometers generally detect motion within 0.25Hz to 2-3Hz, in line with studies that show that voluntary human movement is rarely above 3-4Hz3. The period of time between data samples is known as an epoch. An epoch is classed as “activity” or “inactivity” by an algorithm based on 3 methods for deriving activity: Time Above Threshold, Zero Crossings and Digital Integration. These 3 methods translate analogue data into digital data for interpretation, and are described in table 14, 5. Algorithms vary between devices, hence all need to be tested for their validity.
Table 1: Table explaining the 3 methods used to derive activity from analogue actigraphical data: time above threshold, zero crossing and digital integration5
Time above threshold Method Zero Crossings Method Digital Integration Method
All data above a certain frequency is determined to be “activity”, and the sum of the durations in each epoch The number of times motion crosses 0Hz Raw data is graphed, and the coloured area under the curve represents the duration of activity
Various characteristics of movement are considered before an epoch can be classed as “activity”: frequency, amplitude, acceleration and duration. None of the methods are able to measure all characteristics. The Time Above Threshold method can only measure duration of movement; the Zero Crossings method can only measure the number of movements; the Digital Integration method can measure both amplitude and acceleration of movement5. Digital integration has been determined to be the best method for measuring activity; followed by time above threshold and zero crossings4. However, most algorithms incorporate one or more methods, allowing most factors to be recorded.
2.1 Algorithms in actigraphy
Efforts have been made to develop a standardised algorithm, most notably Actigraphy Data Analysis Software (ADAS), the Sadeh algorithm and the Cole-Kripke algorithm 6-8; although universal application of one algorithm has not been realised, due to heavy competition amongst manufacturers and developers of actigraphy devices. Hence, the same raw data can give very different results when a different algorithm is used for scoring5. Meltzer et al.9 showed that for total sleep time (TST), wake after sleep onset (WASO) and sleep efficiency (SE), the Cole-Kripke algorithm generally yielded values closer to PSG values, compared to the same device model using the Sadeh algorithm.