With the capabilities now available to allow for digital integration procedures to be incorporated into small, lightweight ambulatory activity monitors, it may be an opportune time to re-evaluate this suggestion and open a discussion among the members of sleep specialists on the matter
Within the past 2 decades, practitioners in the field of sleep have demonstrated an increased interest in the use of activity (actigraphy) monitors as both clinical and research tools.1-5 Such devices, which analyze activity patterns by use of an acceleration sensor (accelerometer),6-8 can be used to estimate variables of sleep such as duration, latency, efficiency, quality, and fragmentation9-21; sleep and wake patterns22-26; circadian rhythms27-32; and day/nighttime activity levels.33 A primary reason for the increased use of actigraphy monitoring is this methods effectiveness as a low-cost, nonintrusive method of assessing sleep-related variables for long periods of time in the patients natural environment with a minimal amount of disturbance.11,34,35 In addition, these devices can be used to collect data in situations where standard, polysomnographic recording is not economically or logistically feasible.10,36-39
As interest in activity recording has grown, the number of such devices available in the marketplace has also increased.3 In addition, it has become conventional to refer to the output data produced by actigraphy monitors as activity counts, an ambiguous term that is becoming increasingly common in the vernacular of the sleep field. However, not only is it unclear to many professionals what, specifically, an activity count is, and what the numbers generated by actigraphy monitors actually mean, the terms indiscriminant use has also given rise to two commonly held assumptions about activity recording, both of which are incorrect. First, the general use of the term activity counts suggests that all actigraphy devices utilize similar methodologies to calculate these values and, therefore, produce comparable data. Second, this nomenclature gives the perception that an activity count is a standard unit of measurement, implying that an activity count value recorded with one brand of monitor is equivalent to that recorded by a different model. In actuality, different makes and models utilize different methods to measure activity and, hence, produce output values that are not necessarily comparable.40
There are three different methodologies that have typically been used in actigraphy monitors to measure physical activity, which are time above threshold, zero-crossing, and digital integration.8,33,34,41,42 Each technique works differently and can produce disparate output values. With the growing use of actigraphy monitoring in the field of sleep medicine, it is important that researchers be familiar with these different techniques in order to be cognizant of exactly what is being measured.
Time Above Threshold
In this method, a specific activity level is preset as the signal threshold (typically 0.l-0.2g) and the resulting activity counts reflect the amount of time that the movement signal exceeds this preset value (see Figure 1A). When this occurs, the monitor counts, at a prescribed frequency, the period of time during which the activity level remains above the threshold value.8,34,42 For a given time period, the sum of these counts is then stored into memory as activity counts for that epoch. The activity count is then reset to zero and the counting process begins for the next epoch.41
|Figure 1. Activity count values produced by three different methods for a 1-minute activity waveform.
This approach has one significant limitation in that it provides only a partial description of the wearers activity. Specifically, there is little accounting for the amplitude of the acceleration signal (or waveform) of the movements being measured. Utilizing the time above threshold method makes it possible for two activity patterns with significant differences in intensity to return the same measured value. In addition, it has been suggested that this type of output variable does not properly measure acceleration induced by muscle force,33 which may result in an incomplete assessment of the activity pattern being analyzed.
A second method used in actigraphy monitors is zero-crossing.42 This procedure is a type of threshold crossing detection34 where the threshold value is typically set to zero or some low level of activity determined to indicate the absence of activity without artifact (the zero reference).33 Monitors utilizing zero-crossing techniques record and store counts as the number of times the activity signal crosses the zero reference point within an epoch; the resulting activity count value is the number of signal zero-crossings that occur within that time period (see Figure 1B, page 42).
The zero-crossing method has limitations similar to the time above threshold procedure in that there is, again, limited sensitivity to the amplitude of the activity waveform being measured. It has also been argued that zero-crossing methods may not provide an accurate measure of movement-related acceleration33 and may provide an inadequate index of overall activity because it favors the higher frequency components of the [activity] signal [and] high-frequency components of the signal are potential sources of noise; most normal human movement is restricted to lower frequencies, especially during periods of sleep or rest.34
The third, and most recently implemented, method utilized in actigraphy monitoring is a procedure known as digital integration. In this technique, the acceleration signal is sampled at a high rate, typically up to 40 times per second, with analog-to-digital (A/D) conversion procedures used to virtually reconstruct the activity waveform.43 The area under the activity curve is then calculated by numeric integration procedures,11,34,44 and the activity counts produced represent the average level of activity within an epoch as determined by the total area under the curve45 during that period of time (see Figure 1C, page 42).
This method has a significant advantage over the time above threshold and zero-crossing methods in that it yields a more comprehensive measure of physical acceleration, which allows for the amplitude, or the intensity of motion, to be taken into account. Early proponents of the use of actigraphy for the assessment of sleep variables had suggested that digital integration would be optimal for the measurement of physical activity34 but, until recently, technological limitations have prevented the efficient utilization of this methodology.
While all three of these techniques for measuring physical activity by means of an acceleration sensor have been discussed in the literature, and while specific monitors have been compared to one another in terms of their utility,40 researchers have yet to directly compare these recording methodologies, at the mechanical level, on a specific, controlled set of data. The purpose of the current investigation was to conduct such a procedure in order to provide an illustrative comparison of these three techniques.
A series of analog waveforms were generated from an accelerometer assembly and subjected to random motions at frequencies between 0.5 Hz and 10 Hz with accelerations ranging from 0.05-1.0g and digitized at 40 Hz to provide a database of model physical activity. Ten 1-minute samples of activity data were randomly selected from this database, and each sample was then analyzed by a computer program written to calculate activity counts using each of the three methods of activity assessment. The resulting activity counts produced by these three procedures were then compared with one another.
Each of the selected samples was then modified by significantly changing the amplitude of the digitized activity waveform in order to assess the sensitivity of each method to changes in movement intensity. The magnitude of the original waveforms was first increased by 100% from baseline and then reduced by 50% from baseline. Activity count values derived from the altered waveform samples were then compared with the counts derived from the original waveform samples for each of the three activity count methods.
It was observed that for any given activity waveform, each of the calculation methods produced a widely divergent activity count value (Table 1 and Figure 1, page 42). The three procedures demonstrated only a modest degree of correlation, with the time above threshold and zero-crossing procedures demonstrating a stronger relation with one another than with the digital integration method (Table 2).
|Table 1. Activity count values as calculated by the three methods for 10 1-minute epochs of activity.
|Table 1. Correlations between the three activity count calculation methods for all 30 activity waveforms.
Comparisons of the activity count values calculated for the original baseline samples of activity to those calculated for the modified waveforms yielded the observation that the counts produced by both the time above threshold and the zero-crossing methods demonstrated no significant changes as a result of these changes in movement intensity. This was observed for both a 100% increase in waveform amplitude compared to baseline (tTAT=0.03, P>0.05; tZC=0.02, P>0.05) and a 50% decrease in waveform amplitude compared to baseline (tTAT = 0.03, P > 0.05; tZC < 0.01, P>0.05). However, for the digital integration method, it was observed that a change in signal amplitude was accurately and proportionally registered in the resulting activity count, with significant changes occurring in the activity count values when the baseline amplitudes were increased by 100% (t =2.30, P<0.05) and decreased by 50% (t = 2.26, P<0.05).
The use of actigraphy monitoring in the sleep field has grown rapidly in recent years. Related to the increased utilization of this technology, the term activity counts is being used with greater frequency. However, there is still much confusion among sleep professionals as to exactly what actigraphy monitors measure and what an activity count actually represents. The three methods typically used by actigraphy monitors for measuring physical activity can each produce widely different activity values, which are not equivalent to one another, as is illustrated in the comparison of the activity counts produced by these procedures when calculated from the same activity waveforms. This results in an inherent difficulty in comparing data obtained from actigraphy monitors that utilize these different methodologies.
Furthermore, the first two methods are limited by their inability to account for the magnitude (or strength) of physical movements. The digital integration technique, however, overcomes this limitation by providing a more comprehensive description of physical activity in which the intensity of the limb movements is quantified and taken into account. Thus, the authors believe that digital integration should be the preferred technique in actigraphic recording. It is not, however, being argued that the time above threshold and zero-crossing techniques are inadequate for the measurement of physical activity, but researchers should be aware of this inherent limitation in the data collected by actigraphy monitors that utilize these calculation methods.
In addition, another, and perhaps more important, problem arises from the fact that an activity count is not a standardized unit of measurement. Even monitors that use the same basic recording method may be calibrated in a way so as to assign disparate values when subjected to the same activity pattern, which can further hamper the ability to compare recordings made with different monitors.6 For this reason, the authors recommend that activity counts should be defined using standardized measures of acceleration, such as g-force units. This would not only make the numbers produced by actigraphy monitors more meaningful, it would also allow for comparisons to other measures of acceleration, which would serve to further advance the development of actigraphy practice and technology in our field.
In 1985, Redmond and Hegge34 indicated that a digital integration method of activity recording would be a highly desirable choice, but could not be seriously considered because of the limitations of the then-available technology, which led to the adoption of time above threshold and zero-crossing techniques. With the capabilities now available to allow for digital integration procedures to be incorporated into small, lightweight ambulatory activity monitors10,11 it may be an opportune time to re-evaluate this suggestion and open a discussion among the members of our profession on the matter. The American Academy of Sleep Medicine (AASM) recently presented a set of guidelines for the use of actigraphy in the assessment of sleep disorders,36 which indicated that serious consideration needs to be given to the adoption of standards regarding the use of actigraphy in the sleep field. The AASM has suggested how these types of devices should be used, but these recommendations are, in effect, trivial if no standards are set for the mechanics of the devices themselves.
Stephen W. Gorny and Jennifer R. Spiro are both graduate students in the Developmental Psychology Program at Johns Hopkins University, Baltimore.
1. Hauri PJ, Wisbey J. Wrist actigraphy in insomnia. Sleep. 1992;15:293-301.
2. Montoye HJ, Washburn R, Servais S, Erti A, Webster JG, Nagle FJ. Estimation of energy expenditure by a portable accelerometer. Med Sci Sports Exerc. 1983;15:403-407.
3. Sadeh A, Hauri PJ, Kripke DF, Lavie P. The role of actigraphy in the evaluation of sleep disorders. Sleep. 1995;18:288-302.
4. Webster JB, Kripke DF, Messin S, Mullaney DJ, Wyborney G. An activity-based sleep monitor system for ambulatory use. Sleep. 1982;5:389-399.
5. Sadeh A, Sharkey KM, Carskadon MA. Activity-based sleep-wake identification: an empirical test of methodological issues. Sleep. 1994;17:201-207.
6. Kloeck B, de Rooij NF. Mechanical sensors. In: Sze SM, ed. Semiconductor Sensors. New York: Wiley; 1994:153-204.
7. Bassey EJ, Fentem P. Monitoring physical activity. In: Littler WA, ed. Clinical Ambulatory Monitoring. London: Chapman Hall; 1980:97-124.
8. Kripke D, Mullaney D, Messin S, Wyborney J. Wrist actigraphic measures of sleep and rhythms. Electroencephalogr Clin Neurophysiol. 1978;44:674-676.
9. Mullaney DJ, Kripke DF, Messin S. Wrist-actigraphic estimation of sleep time. Sleep. 1980;3:83-92.
10. Gorny SW, Allen RP, Krausman DT, Earley CJ. Parametric analyses of factors affecting accuracy for detection of wake epochs after sleep onset based on wrist activity data. Sleep Res. 1996;25:490.
11. Gorny SW, Allen RP, Krausman DT, Cammarata J, Earley CJ. A parametric and sleep hysteresis approach to assessing sleep and wake from a wrist activity meter with enhanced frequency range. Sleep Res. 1997;26:662.
12. Al-shajlawi A, Bootzin RR, Lack L, Wright H. Interrelationships between sleep diaries, actigraphy, and polysomnography in measuring improvement in sleep onset during treatment. Sleep. 1999;22:S355.
13. Middelkoop HAM, Neven AK, van Hilten JJ, Ruwhof CW, Kamphuisen HAC. Wrist actigraphic assessment of sleep in 116 community based subjects suspected of obstructive sleep apnoea syndrome. Thorax. 1995;50:284-289.
14. Ludwig C, Valdiserri M, Baldwin D, Rider S, Wyatt JK, Bootzin RR. Individual difference correlates of the extent for agreement between wrist actigraphy and sleep diaries for insomniacs. Sleep Res. 1994;23:451.
15. Lavie P, Epstein R, Tzischinsky O, Gilad D, Nahir M, Lorber M, Scharf Y. Actigraphic measurements of sleep in rheumatoid arthritis: comparison of patients with low back pain and healthy controls. J Rheumatol. 1992;19:362-365.
16. Cole RJ, Kripke DF, Gruen W, Mullaney DJ, Gillin JC. Automatic sleep/wake identification from wrist actigraphy. Sleep. 1992;15:461-469.
17. Monk TH, Buysse DJ, Rose LR. Wrist actigraphic measures of sleep in space. Sleep. 1999;22:948-954.
18. Matsumoto M, Miyagishi T, Sack RL, Hughes RJ, Blood ML, Lewy AJ. Evaluation of the ACtillume wrist actigraphy monitor in the detection of sleeping and waking. Psychiatry and Clinical Neuroscience. 1998;52:160-164.
19. Shinkoda H, Matsumoto K, Hamasaki J, Seo YJ, Park YM, Park KP. Evaluation of human activities and sleep-wake identification using wrist actigraphy. Psychiatry and Clinical Neuroscience. 1998;52:157-159.
20. Friedman L, Benson K, Noda A, et al. An actigraphic comparison of sleep restriction and sleep hygiene treatments for insomnia in older adults. Geriatr Psychiatry Neurol. 2000;13:17-27.
21. Pollack CP, Nagaraja H, Tryon WT, Dzwonczyk R, Rahn K, Loxterman M. How accurately does wrist actigraphy (circa 1999) identify the states of sleep and wakefulness? Sleep. 1999;22:S109-S110.
22. Borbely AA. New techniques for the analysis of the human sleep-wake cycle. Brain Devel. 1986;8:482-488.
23. Sadeh A, Lavie P, Scher A, Tirosh E, Epstein R. Actigraphic home monitoring of sleep-disturbed and control infants and young children: a new method for assessment of sleep wake patterns. Pediatrics. 1991;87:494-499.
24. Matty MK, Dahl RE, Al-Shabbout M. Actigraphy as a method to document adolescent sleep patterns. Sleep Res. 1994;23:452.
25. Sadeh A. Evaluating night wakings in sleep-disturbed infants: a methodological study of parental reports and actigraphy. Sleep. 1996;19:757-762.
26. Acebo C, Sadeh A, Seifer R, Tzischinski O, Carskadon MA. Sleep/wake patterns in one to five year old children from activity monitoring and maternal reports. Sleep. 2000;23:A30.
27. Brown AC, Smolensky MH, DAlonzo GE, Redmond DP. Actigraphy: a means of assessing circadian patterns in human activity. Chronobiol Int. 1990;7:125-133.
28. Lieberman HR, Wurtman JJ, Teicher MH. Circadian rhythms of activity in healthy young and elderly humans. Neurobiol Aging. 1989;10:259-265.
29. Pollak CP, Perlick D, Linsner JP. Daily sleep reports and circadian rest-activity cycles of elderly community residents with insomnia. Biol Psychiatry. 1992;32:1019-1027.
30. Tryon WW. Activity Measurement in Psychology and Medicine. New York: Plenum Press; 1991.
31. Brown AC, Smolensky MH, DAlonzo GE, Redman DP. Actigraphy: a means of assessing circadian patterns in human activity. Chronobiol Int. 1990;7:125-133.
32. Mullington JM, Mantzoros C, Samaras J, et al. Circadian rhythm amplitude of Leptin is reduced by chronic sleep restriction to 4 hours per night. Sleep. 2000;23:A71.
33. Van Someren EJW, Lazeron RHC, Vonk BFM, Mirmiran M, Swaab DF. Gravitational artefact in frequency spectra of movement acceleration: implications for actigraphy in young and elderly subjects. J Neurosci Methods. 1996;65:55-62.
34. Redmond DP, Hegge FW. Observations on the design and specification of a wrist-worn human activity monitoring system. Behavior Research Methods, Instruments and Computers. 1985;17:659-669.
35. Sadeh A, Alister J, Urbach D, Lavie P. Actigraphically based automatic bedtime sleep-wake scoring: validity and clinical applications. Journal of Ambulatory Monitoring. 1989;2:209-216.
36. American Sleep Disorders Association. Practice parameters for the use of actigraphy in the clinical assessment of sleep disorders. Sleep. 1995;18:285-287.
37. Haberger R, Geisler P, Tracik F, Klein HE. High resolution actigraphy: a simple new method to detect periodic limb movements in sleep. Paper presented at: 14th Congress of the European Sleep Research Society; September 9-12, 1998; Madrid.
38. van Hilten JJ, Braat EAM, van der Velde EA, Middelkoop HAM, Kerkhof GA, Kamphuisen HAC. Ambulatory activity monitoring during sleep: an evaluation of internight and intrasubject variability in healthy persons aged 50-98 years. Sleep. 1993;16:146-150.
39. Krahn LE, Lin SC, Wisbey J, Rummans TA, OConnor MK. Assessing sleep in psychiatric inpatients: nurse and patient reports versus wrist actigraphy. Ann Clin Psychiatry. 1997;9:203-210.
40. Pollak CP, Stokes PE, Wagner DR. Direct comparison of two widely used activity recorders. Sleep. 1998;21:207-212.
41. Colburn TR, Smith BM. Activity monitor for ambulatory subjects. US Patent #4,353,375; 1982.
42. Leidy NK, Abbot RD, Fedenko KM. Sensitivity and reproducibility of the dual-mode actigraphy under controlled levels of activity intensity. Nurs Res. 1997;46:5-11.
43. Cogdell JR. Foundations of Electrical Engineering. Englewood Cliffs, NJ: Prentice-Hall; 1990.
44. Rorabaugh CB. DSP Primer. New York: McGraw-Hill; 1999.
45. Korn GA, Korn TM, Sakk E. Mathematics, formulas, definitions, and theorems. In: Fink DG, Jurgen RK, Torrero EA, eds. Electronics Engineers Handbook. 4th ed. New York: McGraw-Hill; 1997.