Parametrization of Physical Activity Aggregation
Monika
ˇ
Simaityt
˙
e
1
, Andrius Petr
˙
enas
1,2
and Vaidotas Marozas
1,2
1
Biomedical Engineering Institute, Kaunas University of Technology, Kaunas, Lithuania
2
Faculty of Electrical and Electronics Engineering, Kaunas University of Technology, Kaunas, Lithuania
Keywords:
Physical Activity Distribution, Physical Activity Profile, Smart Wristband, Steps, Sedentary Behaviour,
Long-term Monitoring, Cardiovascular Disease.
Abstract:
This work introduces a novel approach to parametrization of physical activity profile. The proposed parameter,
named as physical activity aggregation, is useful for evaluating a distribution of daily or weekly physical
activity. The parameter takes a large value for a highly accumulated physical activity, whereas is much lower
for an evenly spread activity over the monitoring period. The parameter was investigated on step data obtained
using a smart wristband on a group of 71 participants with cardiovascular disease. The results of the pilot
study show that the proposed parameter is capable of discriminating among different physical activity profiles,
including sedentary behaviour, going to and from work, walking in a park and being active the entire day.
Moreover, the results demonstrate the tendency that middle-aged and older women are associated with lower
aggregation values, suggesting that they probably spend less time in sedentary behaviour compared to men of
the same age. The proposed parameter has potential to be useful for characterizing physical activity profile,
as well as, for investigating its relation to health outcomes, e.g., during ambulatory rehabilitation after major
cardiovascular events.
1 INTRODUCTION
Physical inactivity is often considered among the lea-
ding risk factors of chronic diseases, and is closely
related to all cause mortality (Biswas et al., 2015;
de Souto Barreto et al., 2017). Unfortunately, ac-
cording to the World Health Organization, more than
80% of the population are insufficiently physically
active. It is widely accepted that even low physi-
cal activity is beneficial for health (Sattelmair et al.,
2011), thus it is recommended to avoid sedentary be-
haviour as much as possible to reduce the risk of a
hazardous outcome (Biswas et al., 2015).
Based on the latest report by the Physical Acti-
vity Guidelines Advisory Committee [PAG] (2008),
there is an absence of research on optimal physical
activity for healthy and unhealthy individuals. There-
fore, optimization may potentially lead to an impro-
ved general health status and reduced number of de-
aths (Alves et al., 2016). Positive effect on health is
observed for at least of 150 min weekly moderate acti-
vity (PAG, 2008). However, there is no consensus on
whether 30 minutes in 5 days or 50 minutes in 3 days
are more beneficial for health. In most studies, a mea-
sure of the total time spent in physical activity is usu-
ally employed for investigating causal relationships
with chronic diseases and mortality (Warburton et al.,
2010; Wilmot et al., 2012). Nevertheless, such factors
as activity session frequency and duration may also be
desirable to account for in order to comprehensively
evaluate physical activity profile (PAG, 2008).
Rapid development of electronics and cloud
technology has given rise to various means of long-
term physical activity monitoring (Piwek et al., 2016).
Therefore, the conventional approach for collecting
information on physical activity via questionnaires
can now be replaced by objective evaluation. The ad-
vancements in technology, capable of tracking phy-
sical activity (e.g., smart wristbands, smart watches,
smartphones), have led to compact, user-friendly and
inexpensive devices, which are especially suitable for
monitoring for extended periods of time (months, ye-
ars). Most of them provide information about the
number of steps, sedentary time, climbed floors, tra-
velled distances, etc. It has been shown that these
devices are sufficiently accurate in tracking physi-
cal activity, therefore, they are becoming increasingly
popular for use in research and medical applicati-
ons (Leininger et al., 2016; Althoff et al., 2017; Leth
et al., 2017).
190
Šimaityte, M., Petr
˙
enas, A. and Marozas, V.
Parametrization of Physical Activity Aggregation.
DOI: 10.5220/0006632101900194
In Proceedings of the 11th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2018) - Volume 4: BIOSIGNALS, pages 190-194
ISBN: 978-989-758-279-0
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
Time
Steps
a
A = 0
Time
Steps
b
A = 0.5
Time
Steps
c
A = 0.25
Time
Steps
d
A = 0.7
Time
Accumulated steps
a
u
Time
Accumulated steps
a
u
Time
Accumulated steps
a
u
Time
Accumulated steps
a
u
Figure 1: Physical activity profiles (above) with corresponding accumulated steps (below) of actual (a) and reference uniform
(u) distributions: a) uniform distribution over the entire monitoring period, b) a continuous single episode taking half of the
total monitoring period, c) two episodes of equal intensity, d) two episodes of unequal intensity.
In this paper, we propose a novel parameter for
an objective evaluation of physical activity aggrega-
tion, allowing to express the distribution of physical
activity over time in terms of a single number. The
acquired new information can potentially be useful
for long-term tracking of changes in the physical acti-
vity profile, as well as, for investigating its relations-
hip to health status.
2 METHODS
2.1 Physical Activity Aggregation
In this study, we define physical activity as a number
of steps in a time interval. Physical activity aggrega-
tion is given by
A =
2
SN
N
i=1
|a
i
u
i
|, (1)
where a
i
and u
i
are accumulated steps of actual and
uniform distributions. The latter is used as a reference
for computing the aggregation of actual distribution.
S the total number of steps in a time period under
analysis (e.g., day or week), N the total number of
time intervals.
The function of accumulated steps of actual distri-
bution a
i
is expressed by
a
i, j+1
=
i+ j
k=i
s
k
, i = 1, ..., N j, j = 0, ..., N 1, (2)
a
i
= (a
i, j
)
max
, i, j = 1, ..., N, (3)
where s – the number of steps in a time interval k.
The function of accumulated steps of reference
uniform distribution u
i
is expressed by
u
i
=
i
N
N
k=1
s
k
, i = 1, ..., N. (4)
Physical activity aggregation A takes values bet-
ween 0 and 1. Values close to 0 indicate low physical
activity aggregation. This applies for physical activity
profiles with steps evenly spread over the monitoring
period (Fig. 1 a). In contrast, values close to 1 indi-
cate maximal temporal aggregation, which is inherent
for physical activity profiles with a single continuous
activity episode (Fig. 1 d). It should be noted, that
the aggregation parameter depends on the duration of
an episode, i.e., A becomes larger for shorter physical
activity episodes.
2.2 Study Population
Seventy-one participants (37 women), 51 ± 13 ye-
ars old, with a body mass index of 27.1 ± 4.7 kg/m
2
,
were enrolled in the study. Most of the participants
were diagnosed with serious cardiovascular disea-
ses, namely, congestive heart failure, angina pectoris,
myocardial ischemia, atrial fibrillation, hypertension,
etc. A signed, written consent to participate in a study
was obtained from all the participants.
In order to investigate the differences in physical
activity profiles among different age women and men,
the participants were assigned to three age groups (see
Table 1).
Parametrization of Physical Activity Aggregation
191
Figure 2: Acquisition of step data and evaluation of daily physical activity aggregation.
Table 1: The number of the participants in each age group.
Age Women Men
<50 years 11 16
50-60 years 13 12
>60 years 13 6
2.3 Data Acquisition
Minute-by-minute step data was obtained using a Fit-
bit Charge 2 (Fitbit, San Francisco, CA, the US) smart
wristband. Then, the data was processed and used for
estimating the physical activity aggregation, see block
diagram in Fig. 2. Since no major activity is expected
during the night, this time period was excluded from
the computation of A . The onset of the night was set
when the number of steps per hour decreased to less
than 20. Similarly, the end of the night was set when
the number of steps per hour exceeded 20.
3 RESULTS
The obtained A values for various physical activity
profiles clearly show that the aggregation parameter
is sensitive to various step distributions (see Fig. 3).
For example, a physical activity profile, dominated by
a low intensity physical activity, results in a very low
aggregation value (Fig. 3 a). Another physical acti-
vity profile (Fig. 3 b) demonstrates the case when the
largest number of steps is aggregated during time pe-
riods corresponding to going to and from work. Such
a profile is especially common among those working
in an office. Similarly, the profile in Fig. 3 c repre-
sents an ordinary workday, however, with additional
physical activity in the evening due to a 2 h walk in
a park. Since half of the total daily activity is aggre-
gated in the evening, the parameter value approaches
0.5. The example in Fig. 3 d stands for the profile
when the participant is only active for a short period
of time, therefore, almost all activity is aggregated in
a continuous episode starting from 12 pm to 4 pm.
For this reason, a highly aggregated physical activity
results in a large A value.
12 am 6 am 12 pm 6 pm 12 am
Steps
0
60
120
a
A = 0.05
12 am 6 am 12 pm 6 pm 12 am
Steps
0
60
120
b
A = 0.34
12 am 6 am 12 pm 6 pm 12 am
Steps
0
60
120
c
A = 0.48
12 am 6 am 12 pm 6 pm 12 am
Steps
0
60
120
d
A = 0.70
Figure 3: Different physical activity profiles with computed
physical activity aggregation: a) uniformly distributed acti-
vity, b) aggregated during the time periods corresponding
to going to and from work, c) aggregated during walking in
the park at the evening, d) most of the activity aggregated
in the afternoon.
Figure 4 displays physical activity aggregation
among different age women and men. The re-
sults show that physical activity aggregation is larger
among participants over 60 years compared to those
under 50 years, which is obvious since younger indi-
viduals are more likely to be physically active, spre-
ading physical activity over the day. Moreover, the
results demonstrate the tendency of physical activity
aggregation being lower for women than men, sug-
gesting that women spend less time in sedentary be-
haviour.
BIOSIGNALS 2018 - 11th International Conference on Bio-inspired Systems and Signal Processing
192
<50 years 50-60 years >60 years
A
0
0.25
0.5
Women Men
Figure 4: Physical activity aggregation among different age
groups. Results are expressed as mean ± two-sided 95%
confidence interval.
4 DISCUSSION
The goal of this work is to propose a parameter for
an objective evaluation of a physical activity distribu-
tion. With such a parameter daily or weekly physical
activity profile is characterized by a single number.
The results of the pilot study show that the aggrega-
tion parameter is capable of differentiating among dif-
ferent physical activity profiles. Therefore, such in-
formation, collected over an extended period of time
(months, years), could be useful for answering the
core question what physical activity profile is optimal
for different patient groups.
It is widely agreed that regularly performed phy-
sical activity extends life expectancy after major car-
diovascular events, such as myocardial infarction. A
study on leisure time physical activity influence on
survival after myocardial infarction has shown that
different survival rates are associated with regular and
irregular physical activity, suggesting that a physi-
cal activity pattern may play a significant role in life
expectancy (Gerber et al., 2011). Therefore, long-
term monitoring of physical activity, and required cor-
rections of an activity profile may improve ambula-
tory rehabilitation. Since a conventional approach
using questionnaires is neither sufficiently accurate,
nor convenient for both the patient and the physician,
objective monitoring of aggregation using smart wris-
tbands may be considered as a promising replacement
for conventional methods.
The approach for quantifying temporal aggre-
gation of specific events was first introduced for
the purpose to characterize the distribution of self-
terminating atrial arrhythmia episodes (Charitos et al.,
2012; Charitos et al., 2013). Differently from the ori-
ginal approach, in which a single arrhythmia episode
shorter than the total monitored time is always assig-
ned to maximal aggregation, our approach is flexible
and duration-dependent. That is, aggregation increa-
ses when the duration of a continuous physical acti-
vity episode decreases. This parameter update is mo-
tivated by the rationale that there is a major difference
between a single very short episode (e.g., 5 min) and
a long one (e.g., 2 h). Therefore, it is incorrect to as-
sign such diverse physical activity profiles to the same
aggregation value.
Large aggregation values represent physical acti-
vity profiles dominated by the sedentary behaviour.
However, since the aggregation parameter is not af-
fected by physical activity intensity, but rather by dis-
tribution itself, the value is the same both for low and
high intensities. Based on the current agreement that
“some activity is good, but more is better” (Sattelmair
et al., 2011; PAG, 2008), the aggregation parameter
can be even more valuable if studied with respect to
other parameters, such as, physical activity intensity.
4.1 Limitations
The major limitation of this study is the small num-
ber of women and men participants assigned to diffe-
rent age groups. In addition, a study cohort preferably
should cover a larger span of age to draw a more reli-
able insights on gender-related physical activity pro-
file.
5 CONCLUSIONS
This study shows that the physical activity aggrega-
tion parameter is useful for an objective evaluation of
a physical activity profile. The proposed parameter is
especially suitable for implementation in devices, ca-
pable of tracking physical activity (smart wristbands,
smartphones), therefore, can provide additional infor-
mation on physical activity relationship with health
status.
ACKNOWLEDGEMENTS
This work was partially supported by the Lithuanian
Business Support Agency (LBSA) under the Intellect
LT measure (Agreement No. J05-LVPA-K-01-0254),
and the Research Council of Lithuania (agreement
No. MIP088/15).
REFERENCES
Althoff, T., Hicks, J. L., King, A. C., Delp, S. L., Le-
skovec, J., et al. (2017). Large-scale physical acti-
Parametrization of Physical Activity Aggregation
193
vity data reveal worldwide activity inequality. Nature,
547(7663):336–339.
Alves, A. J., Viana, J. L., Cavalcante, S. L., Oliveira, N. L.,
Duarte, J. A., Mota, J., Oliveira, J., and Ribeiro, F.
(2016). Physical activity in primary and secondary
prevention of cardiovascular disease: overview upda-
ted. World Journal of Cardiology, 8(10):575–583.
Biswas, A., Oh, P. I., Faulkner, G. E., Bajaj, R. R., Silver,
M. A., Mitchell, M. S., and Alter, D. A. (2015). Se-
dentary time and its association with risk for disease
incidence, mortality, and hospitalization in adults: a
systematic review and meta-analysis. Annals of Inter-
nal Medicine, 162(2):123–132.
Charitos, E. I., Stierle, U., Ziegler, P. D., Baldewig, M., Ro-
binson, D. R., Sievers, H.-H., and Hanke, T. (2012).
A comprehensive evaluation of rhythm monitoring
strategies for the detection of atrial fibrillation recur-
rence: insights from 647 continuously monitored pa-
tients and implications for monitoring after therapeu-
tic interventions. Circulation, 126(7):806–814.
Charitos, E. I., Ziegler, P. D., Stierle, U., Sievers, H.-H.,
Paarmann, H., and Hanke, T. (2013). Atrial fibril-
lation density: A novel measure of atrial fibrillation
temporal aggregation for the characterization of atrial
fibrillation recurrence pattern. Applied Cardiopulmo-
nary Pathophysiology, 17(1):3–10.
de Souto Barreto, P., Cesari, M., Andrieu, S., Vellas, B.,
and Rolland, Y. (2017). Physical activity and incident
chronic diseases: a longitudinal observational study in
16 European countries. American Journal of Preven-
tive Medicine, 52(3):373–378.
Gerber, Y., Myers, V., Goldbourt, U., Benyamini, Y., Schei-
nowitz, M., and Drory, Y. (2011). Long-term trajec-
tory of leisure time physical activity and survival after
first myocardial infarction: a population-based cohort
study. European Journal of Epidemiology, 26(2):109–
116.
Leininger, L. J., Cook, B. J., and Adams, K. J. (2016). Va-
lidation and accuracy of Fitbit Charge: A pilot study
in a university worksite walking program. Journal of
Fitness Research, 5(2):3–9.
Leth, S., Hansen, J., Nielsen, O. W., and Dinesen, B. (2017).
Evaluation of commercial self-monitoring devices for
clinical purposes: results from the future patient trial,
phase I. Sensors, 17(1):211.
Physical Activity Guidelines Advisory Committee Report
(2008). Washington, DC: US Department of Health
and Human Services, 2008:A1–H14.
Piwek, L., Ellis, D. A., Andrews, S., and Joinson, A. (2016).
The rise of consumer health wearables: promises and
barriers. PLoS Medicine, 13(2):e1001953.
Sattelmair, J., Pertman, J., Ding, E. L., Kohl, H. W., Has-
kell, W., and Lee, I.-M. (2011). Dose response bet-
ween physical activity and risk of coronary heart dise-
ase. Circulation, 124(7):784–791.
Warburton, D. E., Charlesworth, S., Ivey, A., Nettlefold,
L., and Bredin, S. S. (2010). A systematic review of
the evidence for Canada’s physical activity guidelines
for adults. The International Journal of Behavioral
Nutrition and Physical Activity, 7:39.
Wilmot, E. G., Edwardson, C. L., Achana, F. A., Davies,
M. J., Gorely, T., Gray, L. J., Khunti, K., Yates, T.,
and Biddle, S. J. (2012). Sedentary time in adults and
the association with diabetes, cardiovascular disease
and death: systematic review and meta-analysis. Dia-
betologia, 55(11):2895–2905.
BIOSIGNALS 2018 - 11th International Conference on Bio-inspired Systems and Signal Processing
194