=Paper=
{{Paper
|id=Vol-1389/paper4
|storemode=property
|title=Ecologically valid trials of elderly unobtrusive monitoring: analysis and first results
|pdfUrl=https://ceur-ws.org/Vol-1389/paper4.pdf
|volume=Vol-1389
|dblpUrl=https://dblp.org/rec/conf/aime/BillisKGTKB15
}}
==Ecologically valid trials of elderly unobtrusive monitoring: analysis and first results==
https://ceur-ws.org/Vol-1389/paper4.pdf
Ecologically valid trials of elderly unobtrusive
monitoring: analysis and first results
Antonis S. Billis1, Panagiotis Kartsidis1, Dimitris-Konstantinos G. Garyfallos1, Mari-
anna S. Tsatali1, Maria Karagianni1 and Panagiotis D. Bamidis1
1
Medical Physics Laboratory, Medical School, Faculty of Health Sciences, Aristotle
University of Thessaloniki
{ampillis, bamidis}@med.auth.gr
{panos.kartsidis, mkaragianni.psy}@gmail.com
dgaryfal@physics.auth.gr
mtsatali@yahoo.gr
Abstract. Intelligent health monitoring systems of elderly have been around for
several years now. Evaluation of sensor measurements and intelligent pro-
cessing algorithms has been performed mainly in lab settings, prohibiting the
collection of datasets that reflect real behavior of seniors. As a result, when
technology migrates to real-life settings, fails to achieve similar monitoring ac-
curacy. Our approach tackles this problem, by piloting the USEFIL intelligent
monitoring system, to elderly people both at lab and home settings. Fifteen (15)
seniors were recruited to follow a number of predefined activities in a free-form
manner for 2 weeks. Five (5) of them were also recruited for piloting the system
in their own homes for a period of two months. Statistical analysis of sensor ob-
servations and clinical assessment tools revealed the monitoring added value of
the sensors in an ecological valid environment. In addition, trend analysis based
on lab findings, showed – by means of a single case study- the potential of the
system to continuously assess health indicators and detect health deterioration
signs.
Keywords: ecological validity; continuous in-home health assessment; active
and healthy ageing; statistical process control; living lab; ambient assisted liv-
ing
1 Introduction
Ambient Assisted Living (AAL) systems have widely developed and evaluated to-
wards their capacity to monitor pathological patterns in elderly people, so to promote
early risk identification, related to chronic diseases [1], [2]. However, most approach-
es followed have severe limitations in their prospect to be applied in real-life settings
[3], since evaluation of algorithms is done either by recruiting young adults [4] or by
strict lab experiments [3] or short-term trials at home with small amount of trial
homes [5]. Our approach provides evaluation of the USEFIL intelligent monitoring
system [6], both in an ecologically valid lab environment and at seniors’ residencies.
Analysis of low level events, derived by sensors, are correlated to clinical assessment
32
AI-AM/NetMed 2015 4th International Workshop on Artificial Intelligence and Assistive Medicine
batteries, providing evidence for the clinical added value of the USEFIL intelligent
monitoring system. Contrary to existing work in the field [3], free-form activities
have been introduced to alleviate strict execution of tasks, resulting to a free-form,
ecological valid dataset. Long-term, trend analysis has been subsequently applied to
low-level events that have been found statistical significantly correlated to clinical
assessment tests. Statistical process control modeling [7] has allowed for retrospective
visualization of seniors health patterns, while leaving at their own homes.
2 Materials & Methods
2.1 Lab pilots
Lab pilots ran in Thessaloniki, in the Active & Healthy Aging Living Lab (AHA
LL). There, a living room environment and a kitchen environment were set up in the
same room. The initial layout of the AHA LL is visualized in Fig. 1. In order to look
more realistic, AHA LL was equipped with home appliances and furniture so as to
better resemble a senior’s home. There, the necessary technological infrastructure and
the USEFIL hardware were installed.
Fig. 1. AHA LL spaces & monitoring system unobtrusive set up
The methodology that was followed towards the execution and evaluation of the trials
at the lab was: i) recruitment, ii) baseline assessment & follow up, iii) protocol of
directed activities definition, iv) trial execution – ongoing period of trials, v) end users
feedback and vi) data analysis.
As a first step seniors’ demographic data and medical history were obtained. Global
cognitive functioning was assessed using the MMSE. Depression levels were evaluat-
ed with the PHQ-9 scale, Quality of life index was measured by SF12, ICECAP and
ASCOT INT 4, whereas the ability of independent living was assessed by the Barthel
index. Fullerton test was used to assess participant’s physical performance. After a
two weeks period, participants were assessed to the previous assessment battery for
follow up purposes.
33
AI-AM/NetMed 2015 4th International Workshop on Artificial Intelligence and Assistive Medicine
The real testing and use of the environment took place for 8 days maximum for each
participant. Each session lasted approximately 60-90 minutes. The participants were
asked to conduct a series of specific tasks as independently as possible.
Fig. 2. Recorded activities in AHA LL
2.2 Home pilots
Technical setup of the USEFIL system took place in five (5) seniors’ homes. USEFIL
software and hardware was installed and setup a-priori at lab premises. Typical instal-
lation example is shown in Fig. 3.
WWU
charger
Smart TV
Kinect
Body Scale
Fig. 3. USEFIL system setup at senior’s home
Five (5) elderly, lone-living women aged 75,6±4,72 years and 14.8±6.57 years of
education were recruited. Four out of five seniors (4/5) had memory problems, while
34
AI-AM/NetMed 2015 4th International Workshop on Artificial Intelligence and Assistive Medicine
two (2) of them had depressive symptomatology. All five seniors had participated in
the lab pilots. So, recruitment took place after they had completed the testing of the
system and were asked of their intention to use the USEFIL system at their own
homes, in the realm of a focus group discussion. Participants, that declared interest,
were explained about the purposes of the home study and upon acceptance, they
signed an informed consent, declaring their voluntary participation. Seniors were
examined by two (2) neuropsychologists at baseline, one-month follow up and at the
end of the two-month period. Global cognitive functioning was assessed using the
MMSE. Depression levels were evaluated with the PHQ-9 scale, Quality of life index
was measured by SF12, ICECAP and ASCOT INT 4, whereas the ability of inde-
pendent living was assessed by the Barthel index. Fullerton test was used to assess
participant’s physical performance.
After, the initial training period neuropsychologists either visited in person seniors
twice per week or they contacted them via telephone. Seniors were encouraged to
perform a list of minimum optional daily tasks related to their interaction with
USEFIL system’s devices and apps.
3 Results
3.1 Sensors vs Clinical assessment
In order to evaluate the clinical added value of the USEFIL system, sensor meas-
urements (Low-Level Events) were correlated to the battery tests that were performed
at the baseline and the follow up. In particular, correlation analyses were performed
between the neuropsychological, physical test results and sensors’ observations. The
correlation coefficient used was Pearson's r. The statistical significant findings of the
analyses are shown in Fig. 4. PHQ results – which refer to the assessment (existence
and severity) of the depressive symptomatology - were correlated either negative or
positive to mobility or gait parameters as measured by sensors, e.g. StepCount (num-
ber of Steps per minute), WalkingSpeed (cm/sec), feetElevation (height of feet while
walking in front of the Kinect) speech and facial expression characteristics, e.g.
speech arousal, eyes’ blinking rate and facial skin color redness level. Most of the
above findings are in line with medical literature [8][9][10]. SF12 mental component
is a subjective feeling of a senior about his/her mental ability/ies. This subjective
measure of quality of life was negatively correlated to feet elevation. However no
data are available, supporting the fact that someone has increased levels of quality of
life, while their feet elevation decreases. A statistically significant relationship was
found between ICECAP (sum score) and walking speed (p=.046). This evidence is in
line with previous studies [11], where walking speed is considered as a predictor of
quality of life. Additionally, the item of ‘thinking about the future’ from ICECAP is
related to speech arousal (p=.05), which means that participants who expressed wor-
ries and were anxious regarding the future, were more likely to have higher speech
arousal scores, compared to those who felt more safe about the future. The variable of
independence, measured by ICECAP too, is related to the sitting speed and the walk-
ing speed, which means that those who feel independent in their daily life, had a bet-
ter mobility status. Furthermore, ASCOT INT 4 scale, which also assess quality of
life, was found to be positively related to speech arousal (p=.015), and negatively
35
AI-AM/NetMed 2015 4th International Workshop on Artificial Intelligence and Assistive Medicine
related to sitting speed (p=.034). Although there are no data supporting this evidence,
it is a quite important evidence to be studied in following studies. Parameters of in-
dependent living, specifically, bowels control, toilet use and transfer activity, are sig-
nificantly related to sitting speed, while toilet use (p=.025) and transfer activity
(p=.025) are correlated with number of steps. Chair stand test (measures lower body
strength in terms of number of completed chair stands in 30 seconds) was negatively
related to feet elevation, step count and sitting speed. Lower sitting speed time de-
notes better balance and lower body strength. Therefore more repetitions executed by
participants show their good balance ability and lower body strength. 2-minute step
test (measures seniors’ aerobic endurance and dynamic balance) was negatively relat-
ed to sitting speed and walking speed. The latter seems to be inconsistent and it needs
more data to be confirmed. Finally, Foot up & go test (measures speed, agility and
balance while moving) had statistical significant relationship to step count (p=.044),
while it is negatively related to walking speed (p=.036). The latter was an unexpected
result and needs to be studied with more participants.
Fig. 4. Clinical Assessment vs Sensor Measurements correlation (1st line – Pearson
correlation, 2nd line - significance)
36
AI-AM/NetMed 2015 4th International Workshop on Artificial Intelligence and Assistive Medicine
3.2 Long-term follow up
In order to demonstrate the monitoring capabilities of the USEFIL system within
home settings, long-term trends relative to health parameters such as mobility, gait,
emotion and cognition. Taking into account lab findings, i.e. correlations found
among sensor observations and clinical assessment batteries, three time periods were
recognized (baseline period, intermediate period and follow up period) and modeled
as statistical processes with respect to sensor observations, calculating their mean and
control limits. Based on these control limits, days that do not lie within process con-
trol limits are candidate abnormal points.
Analysis of long-term sensor observations is presented in one case study, which re-
fers to recognition of depressive symptoms’ deterioration.
Participant #5.
Participant #5 is 71 years old and lives alone. She presents with symptoms of de-
pression of which the most eminent are her lack of interest in activities and her fre-
quently expressed sadness. Her mood fluctuates throughout the day from happy and
energetic to pretty sad and tired. Loneliness and bad quality of sleep are important
factors of her symptoms of depression. Other important factors are her poor capability
to concentrate on activities and her fear of having memory losses. Also her mobility is
limited because of her arthritis. Her knees are a source of severe and persistent pain
which also affects negatively her mood.
Participant’s clinical assessment of depressive symptomatology is provided for all
three assessment periods: baseline, 1-month interim and 2-months follow up.
Table 1. Participant #5 depressive symptoms. Red cells indicate symptoms’ deterioration.
PHQ-1 PHQ-2 PHQ-7 (con-
(loss of (depressive centration defi-
interest) mood) cits)
Baseline
1 2 3
27/1/2015
Interim
3 2 1
27/2/2015
Follow up
3 3 3
17/3/2015
Based on correlations that were found in the AHA LL data between sensors and di-
agnostic tools, concentration deficits severity is inversely proportional to number of
steps, walking speed and speech arousal. Therefore, all three parameters are modeled
and their statistical properties – the three parameters are modeled as statistical pro-
cesses as described in [7] - are calculated for time periods where state deterioration is
annotated according to PHQ-9 (c.f. Table 1). The whole monitoring period is divided
in three time periods: the baseline period, - which accounts for a 2-week period, start-
ing from the date that the baseline assessment was performed -, the interim period,-
which accounts for the period starting right after the end of the baseline period and
ending at the time of the interim visit and assessment was performed-, and the follow
37
AI-AM/NetMed 2015 4th International Workshop on Artificial Intelligence and Assistive Medicine
up period, which accounts for the period starting right after the end of the interim
period and ending at the time that follow up assessment is performed, at the end of the
trial period.
Connected line represents the parameter’s value fluctuation during the reference peri-
od, while the dots represent parameter’s values during the period under investigation.
Horizontal lines represent the statistical properties of the reference period, namely the
mean process value, the lower and upper control limits (green, yellow and red color
lines respectively). Values out of reference period’s control limits may be considered
as “abnormal” values and need to be interpreted according to the given context.
Fig. 5. Participant #5 step count modelling. Horizontal axis represents day number. Vertical
axis represents the total number of steps per day.
Fig. 6. Participant #5 walking speed modelling. Horizontal axis represents day number. Verti-
cal axis represents daily average walking speed measured in meters/second (m/s).
38
AI-AM/NetMed 2015 4th International Workshop on Artificial Intelligence and Assistive Medicine
Fig. 7. Participant #5 speech arousal modelling. Horizontal axis represents day number. Verti-
cal axis represents daily average speech arousal measured in abstract units (1-10).
All three figures show a decreasing trend in number of total steps per day, daily aver-
age walking speed and speech arousal. The decreasing trend is expressed in terms of
follow up period days that lie below the lower control limit (yellow horizontal line) of
the interim period. However, this is more apparent to the modeling representation of
the mobility and gait parameters, rather than in the speech modeling. Concentration
deficits of the participant seemed to got worsen according to the ground truth provid-
ed by the neuropsychological assessment. Therefore, there exists a correlation with
the decreasing trends of the three parameters and the seniors’ cognitive status.
4 Discussion
Three clinical scenarios were piloted in the AHA LL: monitoring of emotional dis-
turbances, cognitive decline and functional ability. According to sensor analysis, qual-
ity of life and depressive symptomatology are related to mobility quantified as walk-
ing speed, step count and feet elevation. Through this kind of identification specific
directions can be followed for both early diagnosis and accurate treatment. Elderly
people quality of life is strongly related to physical performance [12], and therefore,
there is a need to early detect any decreasing trends.
Robust measurement of health parameters in ecologically valid environments is a
very important step, towards integration of intelligent monitoring systems in seniors’
homes. We need to stress the fact that the protocol of activities that was used in the
lab pilots, led seniors to behave in a free-form manner, being themselves and not hav-
ing the belief and the anxiety they were assessed or monitored. This fact strengthens
the results that have been obtained and is obviously along the lines of the overall sys-
tem objectives which call for unobtrusiveness.
Pilots at home focused on the potential of using the technology developed within
the project in real-life settings and provide evidence regarding its efficacy as a daily
39
AI-AM/NetMed 2015 4th International Workshop on Artificial Intelligence and Assistive Medicine
assistive tool for the elderly. Long term monitoring of seniors, based on lab evidence
allow for deriving safer conclusions about the intelligent monitoring aspects of the
monitoring system. Trend analysis presented preliminary evidence on decreasing
health patterns, as sensor measurements were tested against changes annotated by
neuropsychologists with clinical assessment tests. However, a two month monitoring
period is considered as a limitation of our study, since it does not allow to check for
slow varying disease trends, such as cognitive decline. This way the reason that just
one case study was presented, since no significant health changes were observed to
the rest of the participants, during the two-month period. However, since equipment is
already in place in a limited number of homes, we plan -for those individuals that will
accept the system to continue to be in their homes, - to allow for its existence for an-
other six months or 1 year period. In this way, more validated data may be gathered
and multiple follow up measurements may be obtained. The latter will provide useful
insights not only for the health and quality of life of the involved individuals but for
the entire health care system per se.
Acknowledgements.
Part of the research leading to these results has received funding from the Europe-
an Union's Seventh Framework Programme (FP7/2007-2013) under grant agreement
no 288532. For more details, please see http://www.usefil.eu. A.S. Billis is supported
by a scholarship from Fanourakis Foundation (http://www.fanourakisfoundation.org/).
References
1. M. Popescu, G. Chronis, R. Ohol, M. Skubic, and M. Rantz, “An eldercare electronic
health record system for predictive health assessment.”, In e-Health Networking Applica-
tions and Services (Healthcom), 2011 13th IEEE International Conference on, pp. 193-
196, 2011.
2. N. Noury, T. Hadidi, M. Laila, A. Fleury, C. Villemazet, V. Rialle, and A. Franco, “Level
of activity, night and day alternation, and well being measured in a smart hospital suite”,
Proc. 30th Annual Int. Conference of the IEEE-EMBS, pp. 3328–3331, 2008.
3. Riboni, D., Bettini, C., Civitarese, G., Janjua, Z. H., & Bulgari, V. (2015). From lab to life:
Fine-grained behavior monitoring in the elderly’s home. In Proc. of PerCom Workshops.
IEEE Comp. Soc.
4. Alemdar, H., Tunca, C., & Ersoy, C. (2014). Daily life behaviour monitoring for health as-
sessment using machine learning: bridging the gap between domains. Personal and Ubiqui-
tous Computing, 1-13.
5. Cheng, H., Liu, Z., Zhao, Y., Ye, G., & Sun, X. (2014). Real world activity summary for
senior home monitoring. Multimedia Tools and Applications, 70(1), 177-197.
6. Artikis, A., Bamidis, P.D., Billis, A., Bratsas, C., Frantzidis, C., Karkaletsis, V., Klados,
M., Konstantinidis, E., Konstantopoulos, S., Kosmopoulos, D., Papadopoulos, H.,
Perantonis, S., Petridis, S. & Spyropoulos, C.S., (2012, August). Supporting tele-health
and AI-based clinical decision making with sensor data fusion and semantic interpretation:
The USEFIL case study. Paper presented at the 1st International Workshop on Artificial
Intelligence and NetMedicine, Montpellier, France.
7. Billis AS, Papageorgiou EI, Frantzidis CA, Tsatali MS, Tsolaki AC, Bamidis PD. A Deci-
sion-Support Framework for Promoting Independent Living and Ageing Well. IEEE J Bi-
omed Heal Informatics 2015;19:199–209.
8. Mackintosh, J. H., Kumar, R., & Kitamura, T. (1983). Blink rate in psychiatric illness. The
British Journal of Psychiatry, 143(1), 55-57.
40
AI-AM/NetMed 2015 4th International Workshop on Artificial Intelligence and Assistive Medicine
9. Drummond, P. D. (1994). The effect of anger and pleasure on facial blood flow. Australian
Journal of Psychology, 46(2), 95-99.
10. Siegman, A. W., & Boyle, S. (1993). Voices of fear and anxiety and sadness and depres-
sion: the effects of speech rate and loudness on fear and anxiety and sadness and depres-
sion. Journal of Abnormal Psychology, 102(3), 430
11. Purser, J. L., Weinberger, M., Cohen, H. J., Pieper, C. F., Morey, M. C., Li, T., ... &
Lapuerta, P. (2005). Walking speed predicts health status and hospital costs for frail elder-
ly male veterans. Journal of rehabilitation research and development, 42(4), 535.
12. Rejeski, W. J., & Mihalko, S. L. (2001). Physical activity and quality of life in older
adults. The Journals of Gerontology Series A: Biological Sciences and Medical Scienc-es,
56(suppl 2), 23-35.
41