Table 1 Scoring for each algorithm

Count-scaled

Count-scaled algorithm [6]

Uses ‘average’ estimated bedtime and waketime for population under study. To detect the bedtime sleep “event” the algorithm first moves 3 h forward to detect the first sleep onset event. If sleep is not detected in this 3 h it moves 2 h backwards to identify the last sleep onset event. If a sleep event is not detected within the 3 h after or 2 h before the chosen bedtime, the chosen bedtime (e.g. 7:30 pm) is used. Sleep onset = first of 15 continuous minutes of sleep preceded by 5 min of awake. Sleep offset: last of 15 continuous minutes of sleep followed by 5 min of awake. Awakening: 5 continuous minutes of awake preceded and followed by 15 min of sleep.

Sadeh 1

Sadeh algorithm [13] used in Actilife that changes the algorithm sleep wake thresholds based on newer model sensitivities

This uses the original Sadeh algorithm with changes to incorporate the sensitivity differences between the original accelerometer used to develop the algorithm and newer models. The Sadeh algorithm uses an 11-min window that includes the five previous and five future epochs. Note: any missing epochs are considered ZERO. This happens if the current epoch is at the beginning or end of a dataset. The Sadeh algorithm uses the y-axis epoch data. If any of the epoch counts are over 300, it reduces them to 300. The algorithm requires the following information about the window that you’re looking at: Arithmetic mean (average) of the activity counts for the window (AVG) Number of epochs that have counts ≥ 50 and < 100 (NATS) Standard deviation for the first 6 epochs of the window (SD) Natural (base e) logarithm of a current epoch. Note: If the epoch count is 0, we make this value 0 to avoid infinity problems (LG). Those calculations are put through the following algorithm: (7.601—(0.065 * AVG)—(1.08 * NATS)—(0.056 * SD)—(0.703 * LG)). If the result of that algorithm is GREATER than -4, then current epoch is considered sleep. The original algorithm used a > 0 threshold and Actigraph changed this to -4 because of the differences between old and new models. Requires diary times for TIB and TOB.

Sadeh 2

Sadeh algorithm [13] used in Actilife that changes the algorithm sleep wake thresholds based on newer model sensitivities

Same as above, except entirely automated so diaries are not used

Cole-Kripke 1

Original Cole-Kripke algorithm [5]

The Cole-Kripke algorithm computes a weighted sum of the activity in the current minute, the preceding 4 min, and the following 2 min as follows:

S = 0.0033(1.06an4 + 0.54an3 + 0.58an2 + 0.76an1 + 2.3a0 + 0.74al + 0.67a2) where an4–an1 are activity counts from the prior 4 min, a0 is the current minute, and a1 and a2 are the following 2 min. The current minute is scored as sleep when S < 1. Includes the Webster rescoring rules: (a) After at least 4 min scored as wake, the next 1 min scored as sleep is rescored as wake; (b) after at least 10 min scored as wake, the next 3 min scored as sleep are rescored as wake; (c) after at least 15 min scored as wake, the next 4 min scored as sleep are rescored as wake; (d) 6 min or less scored as sleep surrounded by at least 10 min (before and after) scored as wake are rescored as wake.

Cole-Kripke 2

Based on Cole-Kripke algorithm that is used in Actilife software and uses diary sleep and wake times

Same as above but the epoch data are adjusted to help reduce the variation of the counts from the original devices compared to newer actigraphs. Using the y-axis epoch data, counts are divided by 100. If any of those scaled values are over 300, set them to 300. Also uses diary times to constrain the algorithm.

Cole-Kripke 3

Based on Cole-Kripke algorithm that is used in Actilife software, but does not use the diary sleep and wake times

Same as above but does not use diary times to constrain the algorithm.

Tudor-Locke 1

Based on Cole-Kripke algorithm that is used in Actilife software and uses diary sleep and wake times

Original Tudor-Locke automated method [8] that rescores the sleep/wake states using inclinometer data. Bedtime is identified as the first 5 consecutive minutes defined as sleep. Similarly, wake time is identified as the first 10 consecutive minutes defined as wake after a period of sleep. Bedtime and wake time are only identified when at least 160 min has elapsed between these 2 time points. An unlimited number of nonconsecutive wake minutes are allowed between bedtime and wake time, in keeping with the definition of sleep-period time that includes all sleep epochs and wakefulness after onset. Multiple sleep periods (≥ 160 min) are allowed during each 24-h day. The algorithm was constructed to output the beginning and ending minutes for each sleep period identified, but ultimately retains only the beginning minute of the first period (bedtime) in the time block studied and the final minute of the last period (wake time). Sleep-period time is ultimately calculated as the number of minutes between bedtime and wake time. Does not account for the possibility of nocturnal nonwear or extended episodes of wakefulness separating the SPT into multiple sleep episodes. Also does not consider the potential for misclassifying daytime periods of nonwear or other sedentary behaviors as sleep episodes (i.e., “naps”).

Tudor-Locke 2

Based on Cole-Kripke algorithm that is used in Actilife software and uses diary sleep and wake times

Refined Tudor Locke method [4] rescores sleep/wake states using the inclinometer to identify the probability of sleep and define parameters. Constrains algorithm to nocturnal sleep, by rule that only allows sleep onset between 7:00 p.m. and 5:59 a.m. Sleep offset refined and identified as the first of 10 or 20 consecutive inclinometer revised scored wake minutes, depending on the time of day (10 min—5:00 a.m. to 11:58 a.m.; 20 min—9:40 p.m. to 4:59 a.m.). The RSA allows identification of extended episodes of wakefulness that separate the sleep period time into distinct sleep episodes with multiple sleep onsets and offsets. If two sleep episodes were separated by less than 20 min of inclinometer re-scored wake minutes, then they were combined into a single sleep episode starting with the first minute of the first sleep episode and ending with the final minute of the last sleep episode. Sleep episodes that were separated by at least 20 min of inclinometer re-scored wake minutes were distinct within the sleep period time and were not combined. Total sleep episode time (TSET) represented the total minutes from all sleep episodes; in cases where there was a single sleep episode, or all sleep episodes were separated by less than 20 min of inclinometer re-scored wake, the TSET was equal to RSA sleep period time. Nonwear was identified when 90 consecutive minutes of 0 activity counts were encountered while allowing for up to 2 min of nonzero activity counts. The nonwear period ended when a third minute of nonzero activity counts was detected. If at least 90% of a sleep episode was categorized as nonwear, then all minutes within that sleep episode were redefined as nonwear and not included in the calculation of TSET.

Tudor-Locke 3

Sadeh algorithm [13] used in Actilife that changes the algorithm sleep wake thresholds based on newer model sensitivities

Original Tudor-Locke automated method [8] that rescores the sleep/wake states using inclinometer data. Bedtime is identified as the first 5 consecutive minutes defined as sleep. Similarly, wake time is identified as the first 10 consecutive minutes defined as wake after a period of sleep. Bedtime and wake time are only identified when at least 160 min has elapsed between these 2 time points. An unlimited number of nonconsecutive wake minutes are allowed between bedtime and wake time, in keeping with the definition of sleep-period time that includes all sleep epochs and wakefulness after onset. Multiple sleep periods (≥ 160 min) are allowed during each 24-h day. The algorithm was constructed to output the beginning and ending minutes for each sleep period identified, but ultimately retains only the beginning minute of the first period (bedtime) in the time block studied and the final minute of the last period (wake time). Sleep-period time is ultimately calculated as the number of minutes between bedtime and wake time. Does not account for the possibility of nocturnal nonwear or extended episodes of wakefulness separating the SPT into multiple sleep episodes. Also does not consider the potential for misclassifying daytime periods of nonwear or other sedentary behaviors as sleep episodes (i.e., “naps”).

Tudor-Locke 4

Sadeh algorithm [13] used in Actilife that changes the algorithm sleep wake thresholds based on newer model sensitivities

Refined Tudor Locke method [4] rescores sleep/wake states using the inclinometer to identify the probability of sleep and define parameters. Constrains algorithm to nocturnal sleep, by rule that only allows sleep onset between 7:00 p.m. and 5:59 a.m. Sleep offset refined and identified as the first of 10 or 20 consecutive inclinometer revised scored wake minutes, depending on the time of day (10 min—5:00 a.m. to 11:58 a.m.; 20 min—9:40 p.m. to 4:59 a.m.). The RSA allows identification of extended episodes of wakefulness that separate the sleep period time into distinct sleep episodes with multiple sleep onsets and offsets. If two sleep episodes were separated by less than 20 min of inclinometer re-scored wake minutes, then they were combined into a single sleep episode starting with the first minute of the first sleep episode and ending with the final minute of the last sleep episode. Sleep episodes that were separated by at least 20 min of inclinometer re-scored wake minutes were distinct within the sleep period time and were not combined. Total sleep episode time (TSET) represented the total minutes from all sleep episodes; in cases where there was a single sleep episode, or all sleep episodes were separated by less than 20 min of inclinometer re-scored wake, the TSET was equal to RSA sleep period time. Nonwear was identified when 90 consecutive minutes of 0 activity counts were encountered while allowing for up to 2 min of nonzero activity counts. The nonwear period ended when a third minute of nonzero activity counts was detected. If at least 90% of a sleep episode was categorized as nonwear, then all minutes within that sleep episode were redefined as nonwear and not included in the calculation of TSET.

HDCZA

Calculates wrist rotation (changes in the z-angle) for each 5-min rolling window and values under the 10th percentile over an individual day (noon-to-noon). The algorithm then detects blocks lasting > 30 min, with gaps < 60 min counted towards the identified blocks. The longest block in the day between noon–noon represents the sleep period window. Sleep episodes were defined as the sustained periods of inactivity within the sleep period window. From this, the number of sleep episodes within each sleep period window detected is calculated as well as sleep efficiency within the sleep period window calculated as the percentage of time asleep within the sleep period. Note, newer versions of this algorithm can use values under the 13th, 20th and 50th percentile.