| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH |
|
|
|
|||||||
|
|||||||||
Electronic Letters to:
|
|
Electronic letters published:
![[Read eLetter]](/icons/misc/sectionDOWN.gif){kind=icon}
Inter-generation accelerometer differences and correction for on-board frequency-based filtering
|
|
|||
|
Soren Brage, Investigator Scientist MRC Epidemiology Unit, Institue of Metabolic Science, Addenbrookes Hospital, Cambridge, UK, Vincent van Hees and Niels Brage
Send letter to journal:
soren.brage{at}mrc-epid.cam.ac.uk Soren Brage, et al.
|
Dear Sir, We read with great interest the recent article by Rothney and colleagues (3), describing a well-designed extensive evaluation of inter- generation differences between three different Actigraph models. Besides the obvious implication for achieving consistency in large epidemiological studies with longitudinal designs where the monitor pool is gradually replaced, this evaluation also has implications for researchers with newer generation monitors who wish to estimate intensity using equations derived with older generations of monitors. We note that in the “Radius Experiment” (figure 2), solving the two regression equations for Actigraph models GT1M and 7164 for a radius of 0.06 m (~9.4m∙s-2), one obtains a ratio between the model responses as (60*205-590.5)/(60*225-889.6) = 0.93. This result is remarkably similar to the values 0.89 and 0.91 obtained by Corder et al, comparing these two models during sinusoidal accelerations (3 Hz, +/-1 g) and in a side-by- side comparison in free-living adolescents, respectively (2). Although we acknowledge the complexity imposed by the non-linear differences, it would still be useful if Rothney and colleagues could provide the pragmatic no- intercept regression equation for the newer Actigraph models, using the 7164 model as a reference, since their data represent the most elaborate mechanical evaluation of this particular issue to date. It also highlights, however, the importance of accompanying field-based comparisons which are likely to more realistically reflect the common distribution of acceleration magnitudes and movement frequencies (2). In general, the inter-generation and even inter-monitor-type differences would be more manageable for the field if the accelerometric information was expressed in SI units of m∙s-2, a process which is greatly facilitated by a high degree of linearity between acceleration and recorded information (counts) in the human range of movement. It is therefore somewhat concerning that the admirable attempt made by these investigators to achieve this goal seemed unsuccessful (figure 6a). We would, however, question the implementation of the correction for frequency-based filtering; this is described as done via a second-order approximation (1), derived on the basis of the voltage-frequency response (4). For example, the average responses from the GT1M and 7164 models spun at 250 rpm (4.17 Hz) at radius 0.0466 m (~20.3 m∙s-2) in figure 3 were approximately 6000 and 7000 counts∙min-1, respectively. At this frequency, the acceleration signal is attenuated by a factor of approximately 0.06118*(250/60)^2 - 0.5573*(250/60) + 1.453 = 0.193 in Actigraph monitors, a factor otherwise known as the filter weight. Division of the raw counts by this filter weight, a procedure which should only be applied for frequencies between 0.9 and 4.6 Hz with this quadratic filter weight equation, will yield corrected counts for the two models around 31,077 and 36,256 c-counts∙min-1, respectively. However, reading off figure 6a, these ‘corrected’ values appear to be of the order 20-25,000 c-counts∙min-1. Further, application of the previously published conversion equation for estimating acceleration from model 7164 yields 0.0003024 * [7000 / (0.0576*(250/60)^2 - 0.559*(250/60) + 1.45)] + 0.346 = 17.9 m∙s-2, a value within the ‘true’ 20.3 m∙s-2, considering the error margins caused by potential differences in the two monitor pools (1). Despite the differences between figure 6a and the figures in our paper, the general shape of the response curves are very similar (1; 3), presumably reflecting the manufacturer’s aim in providing comparable measurements across monitor generations. However, subtle differences do remain and perhaps even more than what is revealed by results published so far. If indeed this could be the case, it may be necessary to more fully evaluate these responses in a wide acceleration range across a relatively dense representation of frequencies. This would not only reveal differences, if any, in filter characteristics between monitor types but would also explain the potential impact of differences in the dynamic range which would result in differential clipping (saturation) of the signal. The upper bound of the dynamic range of the GT1M is reported to be 2.5 g and about 2.13 g in the older Actigraph generations. The highest average acceleration in both experiments was about 20 m∙s-2 (1; 3), which represents the average of a rectified sine-wave with amplitude of about 32 m∙s-2 (3.3 g), suggesting some degree of signal saturation could have occurred. Sincerely Yours, Soren Brage, Vincent van Hees, and Niels Brage References 1. Brage S, Brage N, Wedderkopp N and Froberg K. Reliability and validity of the Computer Science and Applications accelerometer in a mechanical setting. Meas Phys Edu Exerc Sci 7: 101-119, 2003. 2. Corder K, Brage S, Ramachandran A, Snehalatha C, Wareham N and Ekelund U. Comparison of two Actigraph models for assessing free-living physical activity in Indian adolescents. Journal of Sports Sciences 25: 1607-1611, 2007. 3. Rothney MP, Apker GA, Song Y and Chen KY. Comparing the performance of three generations of ActiGraph accelerometers. J Appl Physiol 105: 1091-1097, 2008. 4. Tryon WW and Williams R. Fully proportional actigraphy: A new instrument. Behavior Research Methods, Instruments & Computers 28: 392 -403, 1996. |
|||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH |
| Visit Other APS Journals Online |
eLetters: