Smart Monitoring of Population Health Risk Behaviour

Abstract

Monitoring population-level health-risk behaviour is integral to preventing chronic diseases (i.e., diabetes, cardiovascular disease, cancer, etc.). Physical activity and sleep are the key behaviours which influence human health. Smart technologies can be used to improve real-time monitoring of risky behaviours. The objective of this study is to explore population- and individual-level remote monitoring of sleep, indoor physical activity and sedentary behaviours in Canada using data from the Internet of Things (IoT) (ecobee smart thermostat) and fitness trackers. Method: 386 person-hours of data were collected in a pilot study (n =8) to validate the motion sensor data from ecobee smart thermostats. Then, using “Donate your Data” data from ecobee indicators of population-level health were calculated. Results: A positive Spearman correlation coefficient 0.8 (p>0.0001) was found between standard fitness tracker data and ecobee sensors validating its use for population-level analysis. Our results were similar to the Public Health Agency of Canada’s results derived from self-reported surveillance methods. Discussion: This project demonstrates the use of data from non-health sources, like ubiquitous IoT to curate population- and individual-level health indicators. We will deliver novel indicators and insights into health status through the creation of user-centered designed dashboards for individuals, researchers, and policy-makers.

Full text

Smart Monitoring of Population Health Risk Behaviour
Kirti Sundar Sahu
PhD Candidate,
School of Public Health and Health Systems
University of Waterloo
Arlene Oetomo
MSc Candidate,
School of Public Health and Health Systems
University of Waterloo
Plinio Morita
Assistant Professor,
School of Public Health and Health Systems
University of Waterloo
Monitoring population-level health-risk behaviour is integral to preventing chronic diseases (i.e., diabetes,
cardiovascular disease, cancer, etc.). Physical activity and sleep are the key behaviours which influence
human health. Smart technologies can be used to improve real-time monitoring of risky behaviours. The
objective of this study is to explore population- and individual-level remote monitoring of sleep, indoor
physical activity and sedentary behaviours in Canada using data from the Internet of Things (IoT) (ecobee
smart thermostat) and fitness trackers. Method: 386 person-hours of data were collected in a pilot study (n
=8) to validate the motion sensor data from ecobee smart thermostats. Then, using “Donate your Data” data
from ecobee indicators of population-level health were calculated. Results: A positive Spearman correlation
coefficient 0.8 (p>0.0001) was found between standard fitness tracker data and ecobee sensors validating
its use for population-level analysis. Our results were similar to the Public Health Agency of Canada’s
results derived from self-reported surveillance methods. Discussion: This project demonstrates the use of
data from non-health sources, like ubiquitous IoT to curate population- and individual-level health
indicators. We will deliver novel indicators and insights into health status through the creation of user-
centered designed dashboards for individuals, researchers, and policy-makers.
INTRODUCTION
Public health surveillance relies on surveys
and/or self-reported data collection, both of which
require human resources, time commitments, and
financial resources from public health agencies and
participants (Choi, 2012). Unfortunately, survey
results can quickly become outdated due to rapid
changes in society. The health habits of Canadians
have evolved with technology, and research shows an
increase in sedentary behaviour, highlighting the
importance of measuring indicators like the levels of
physical activity (PA) to ascertain human health at a
population level (Glavinovic et al., 2018).
In this paper, we present a novel method for
gathering detailed data in near real-time, with
minimal effort from participants. Simple thermostats
are in nearly every house, with a simple upgrade to a
smart thermostat one can make efficient temperature
adjustments and save on energy costs by adjusting
settings according to human activity. Thermostats are
ubiquitous in Canadian homes, and the current
expansion of the smart thermostat market in line with
smart home technology makes them an ideal data
source collection method over traditional methods
(Cetin & O’Neill, 2017; Dewis, 2008). Utilizing
technology that is deployable at a population level
will enable vast, granular data collection that is
beyond the capabilities of traditional methodologies.
This project is exploring the use of zero-effort
technology using sensor data collected by smart
thermostats and other associated sensors to monitor
individual-level health at the household level
(Mihailidis, Boger, Hoey, & Jiancaro, 2011). Using
the ecobee, smart, wi-fi thermostat we are capable of
reporting PA, sedentary behaviour, and sleep patterns
at the household level (“Ecobee,” 2019). The
thermostat and remote sensors collect both
temperature and motion data, which can be used to
classify activity in the home (i.e., limited movement
likely indicates sedentary behaviour and lack of
movement is indicative of sleep and sleep quality).
The considerable benefit of this method is that it
requires no action from participants or intrusion into
their lives for data collection. It also enables extended
data collection periods without the associated costs

and resources required for regular survey data
collections.
The rising tide of our ageing population
necessitates innovative solutions to meet the
challenges our society will face. The healthcare
system especially will need to adapt to the eminent
demographic transition. Experts have stated that in
Canada:
1. 80% of adults over age of 65 have a chronic
disease (Statistics Canada, 2016)
2. Chronic diseases consume more than 50% of
total healthcare expenditures (Ministry of
Health and Long-Term Care, 2017)
3. 60% of all hospitalizations are due to
chronic diseases (Health Council of Canada,
2011)
4. Shortage of skilled human and financial
resources in healthcare to support aging
population (Fontaine, 2018)
This combination of challenges presents an
opportunity to provide care to individuals aging in
place and delay the need for institutionalized care
until it is truly needed.
With these challenges, it is essential to design
innovative solutions that will prevent chronic
diseases, help people living with chronic diseases
manage them, reduce the burden on the healthcare
system, and train individuals to fill the gaps in
healthcare.
Smart Home Technologies experienced rapid
growth in the last few years, with experts estimating
this space to be worth USD 30 billion in 2018
(Ricquebourg et al., 2006; Zion Market Research,
2018). The most popular brands are Samsung’s
SmartThings, Google, Amazon Alexa, Apple’s Home
Kit, Nest, and ecobee (Tom’s Guide, 2019). Each
example, listed above, is designed to increase
convenience in and around the home by automating
everyday tasks like turning lights on and off,
controlling the temperature in the home or providing
voice-activated commands.
Figure 1: Ecobee 3 smart wi-fi thermostats
Smart Home Technology refers to home
automation which offers home dwellers comfort,
convenience, energy efficiency, and security by
enabling control of various connected smart devices
in their home. Bluetooth is most commonly used to
enabled devices, mobile apps, or other similar
connected devices. A Canadian company
manufacturing smart thermostats since 2007 is ,
ecobee. ecobee is a Canadian smart home automation
company (Figure 1). Their thermostats are usable in
residential and commercial settings. The main
thermostat is connected to the internet via Wi-Fi and
controlled via the built-in touchscreen or remotely via
the desktop or iOS and Android-based mobile
application. ecobee’s primary objective is to cut
energy costs for its customers by improving heating
and cooling according to actual need/usage based on
presence detected in the room.
OBJECTIVES
The main goal of this study is to explore
individual- and household-level health indicators
collected in the home via smart thermostats.
Objectives of this study were to (a) identify if it is
possible to isolate specific user behaviours using the
motion and thermostat sensor data, and (b) develop
remote monitoring of healthy behaviours at the
population level. Furthermore, this study is interested
in identifying whether observed patterns will suffer
variations and if, as a result, it would be possible to
understand human behaviours and consequently,
understand lifestyle habits of an individual or a group
of people. The UbiLab plans to build a prototype
system with machine learning algorithms that will
mine the data to find patterns to achieve these goals.
Based on the trends and patterns, personalized health
behaviour recommendations will be delivered to the
user using an interactive user interface. Lastly, data
will be presented to public health officials in a
dashboard designed using user-centered design
methods presenting big data in a dashboard format.
METHODS
Zero-effort-technologies (ZETs) represent the
future of Ambient Assisted Living (AAL) in which
sensors gather data generated by the person without a
conscious effort on the part of the user (Mihailidis et
al., 2011; Vimarlund & Wass, 2014). Such data could
be integrated with other technologies to give the
healthcare system the ability to tackle unsolved
remote monitoring issues to overcome the limitations
of traditional data collection methods. For example,
when remote motion sensors are in the bedroom, they
can provide insight into sleep duration and quality.

Passive data collection addresses the challenges of
declining participant engagement, low response rates
in surveys and focus groups, rising costs, and
technical barriers to wearable technology.
Additionally, this eliminates recall bias, which is
common when participants quantify the amount of
PA and recount their behaviours or activities. Using
motion-data, we can quantify the amount of PA in the
home to determine individual levels of PA. The
UbiLab partnered with ecobee and leveraged data
from over 10,000 households in North-America
collected through the Donate Your Data (DYD)
program (UbiLab, 2019). This is an opt-in research
program that all thermostat owners can volunteer
(must consent) to be a part of and donate their
thermostat and sensor data. The data is anonymized
and coded before the researchers acquire access. The
methodology for this study is described in Figure-2.
Figure 2: Conceptual framework
A small pilot study (n = 8, ~386 person hours)
was done to validate the use of motion sensor
readings of movement between rooms through cross-
comparison with Fitbit Charge 2 HR (Fitbit, 2019)
step data. Then the DYD dataset containing data from
21,311 participants across 10,250 North American
households were analyzed for patterns using Python,
pandas, and Elasticstack (Elastic, 2019; Pandas,
2018; Python, 2019). Using the larger dataset will
reveal insights at the individual and population level
and can be delivered to relevant stakeholders.
Future work will involve analyzing the latest
DYD dataset from ecobee which contains data form
76,003 households globally, 99% of which are in
North America. Data is recorded at 5-minute
intervals, and the roughly one Terabyte worth of data
spans 36 months from 2015 to 2018.
RESULTS
PASS (Physical Activity, Sleep and Sedentary
Behaviours) Indicators Framework, produced by
Public Health Agency of Canada, developed using
several survey instruments, including the Canadian
Health Measures Survey and Canadian Community
Housing Survey which are collected by Statistics
Canada (Public Health Agency of Canada, 2017).
Using this technology will enable public health
agencies like the Public Health Agency of Canada to
collect novel health indicators, monitor health in real-
time and deliver health insights to Canadians to
increase health literacy. Other benefits include
reducing research to action gap, as well as prudent
time management and financial savings.
The pilot study identified a positive association
between the Fitbit steps data and ecobee motion
sensors data (Spearman’s Correlation coefficient =
0.8, p > 0.001) seen in Figure 3.
Figure 3: Correlation between Fitbit and ecobee
Indicators based on the current methods
developed by PHAC (from PASS) on sleep,
interrupted sleep, daily indoor activity, sedentary
behaviour) were calculated using DYD data. Single
occupant ecobee households in Canada averaged 7.2
hours of sleep in 24-hours, 2.1 hours of interrupted
sleep, were active for 85 minutes daily and spent 4.44
hours being sedentary. A set of rules was created to
best fit the current indicator measures as specified by
PHAC (i.e., similar definitions were used).
We used Fitbit Charge 2 HRs to capture sleep
and heart rate. Adding more sensor functionality is
crucial for algorithm improvement; this includes
collecting additional data via the Samsung
SmartThings Hub; presence, light usage, and
luminance. ecobee is sharing participants and data
from their study, increasing variability within data.
We have improved our data storage and analysis
process, moving the big data architecture from
python to Elasticstack for real-time data streaming
and analysis. We are also actively collaborating with
the Public Health Agency of Canada (PHAC) and
improving our algorithm and analysis process using
their feedback (PHAC, 2019).
DISCUSSION
This project is innovative at the population level:
(i) novel data source(s) used for population-level
surveillance, (ii) new indicators measured, (iii) novel

tools/solutions used to capture the data, and (iv)
creating results from unconventional linked datasets
combining a variety of sources. Data collection on a
granular level in real-time could be a reality soon.
The methods proposed here would enable access to a
significantly increased sample size and increase the
generalizability of results. This awareness and
improvement can help to better PA, sleep and
sedentary behaviour which may result in
improvements in health literacy, and overall health
and wellbeing.
With this project, we are not merely trying to
create a new mobile app or product; instead, the goal
is to leverage existing data already collected for
another purpose. Traditional, nationwide-surveys
collect data from households using smaller sample
sizes and generalize the findings through statistical
methods. However, this method enables ongoing,
dynamic data collection from an increased number of
households which is more representative and
realistic, as it helps to monitor health indicators in
near real-time, including key risk factors for chronic
diseases, such as dementia, diabetes, hypertension,
obesity, mental health, and even some cancers. In the
long term, it will not only help to monitor the rising
disease burden but also will help to reduce it by
enabling early policy level prevention efforts to be
implemented rapidly at the population level.
Currently, we are in the phase of building a proof
of concept, but in the future data will stream directly
to the servers at the Ubiquitous Health Technology
Lab at the University of Waterloo. Real-time analyses
will run on the live data — reducing subjective bias
and human error 24/7. We will build a user-centred
design dashboard for data visualization for the public
health official that will enable decision making which
is timely, specific and actionable. User-centred
design has been shown to enhance human
understanding which enriches the value of the data,
and in doing so, public health practice can be
improved and bring population-level benefits to an
aging society (Cheng et al., 2011; Silva da Silva,
Martin, Maurer, & Silveira, 2011).
STRENGTHS AND CHALLENGES
Unobtrusive data collection, coupled with our
novel methods and analysis, are essential to innovate
and improve upon traditional data collection methods
to improve healthcare and policy decision making.
This will give public health officials and politicians
access to relevant, real-time data analysis to address
program and system level deficiencies and
weaknesses. It is possible to incorporate the use of
human factor design to monitor health risk
behaviours at a population level, as well as develop
personalized health solutions with individualized
content. The ubiquity of the platform makes this the
optimal choice to encourage widespread behaviour
changes based on user-generated activity data.
The proposed method also uses technologies that
are zero-effort/ unobtrusive for participants and users,
enabling naturalistic data collection. With well-
planned infrastructure, real-time data collection and
analysis will change how population-level data is
collected, shifting initiatives to prevention rather than
reaction.
People are interested in individual-level
personalized feedback, and this will enable the
creation of tailored health messaging for individuals
based on their data.
Perhaps the greatest, future challenge lies in the
interoperability of said proposed system. Currently,
sensors and their platforms are proprietary
technology owned by companies who create data
silos. We cannot overlook the ever-present concern of
data privacy and security, which is part of the larger
issue surrounding data governance and access to data.
Healthcare is a complex field with many players, and
this requires governments and organizations to work
together to meet the challenges outlined at the
beginning of this paper.
CONCLUSION
This project builds upon current knowledge in
the public health surveillance landscape. The data,
while not intended for these outcomes is available to
be used for the greater good. The primary benefit is
that without any additional effort it is possible to
answer questions at the population health level. This
project is unique in terms of data source, usage, size,
and analytic methods. New indicators for physical
activity at the population level can be formulated to
represent community level. Data collection has never
been achieved on such a granular level and in near
real-time. The technologies available today can be
customized to collect information at the minute-level
and expand the range of data used for population
health analysis. Using this outcome, we can measure
the changes due to the impact of a specific
intervention at both the individual level and the
population level. For example, the success individual
level health behaviour changes, driven by public
health campaigns or programs can be evaluated via
mobile apps or population level surveillance systems.
Nonetheless, using this platform, we can understand
human behaviour at a detailed level in near real-time,
in terms of physical activity and sleep. Leveraging
such technology will result in a minimal burden to
the user while enabling remote monitoring,
effectively reducing study costs and saving time.

Acknowledgements: We would like to thank our
UbiLab colleagues at the University of Waterloo and
ecobee for their support.
REFERENCES
Cetin, K. S., & O’Neill, Z. (2017). Smart Meters and Smart
Devices in Buildings: a Review of Recent Progress and
Influence on Electricity Use and Peak Demand. Current
Sustainable/Renewable Energy Reports, 4(1), 1–7.
https://doi.org/10.1007/s40518-017-0063-7
Cheng, C. K. Y., Ip, D. K. M., Cowling, B. J., Ho, L. M., Leung,
G. M., & Lau, E. H. Y. (2011). Digital dashboard design
using multiple data streams for disease surveillance with
influenza surveillance as an example. Journal of Medical
Internet Research, 13(4), e85.
https://doi.org/10.2196/jmir.1658
Choi, B. C. K. (2012). The Past, Present, and Future of Public
Health Surveillance. Scientifica, 2012, 1–26.
https://doi.org/10.6064/2012/875253
Dewis, G. (2008). Thermostat use in Canadian homes. EnviroStats,
2(3), 3–22. https://doi.org/Catalogue no.16-002-X
Ecobee. (2019). Retrieved April 21, 2019, from
https://www.ecobee.com/
Elastic. (2019). Elasticsearch. Retrieved April 22, 2019, from
https://www.elastic.co/
Fitbit. (2019). Fitbit Official Site for Activity Trackers & More.
Retrieved April 22, 2019, from https://www.fitbit.com/en-
ca/home
Fontaine, D. (2018). Opinion: There’s a human resources crisis in
seniors care - Canadian Association for Long Term Care.
Retrieved April 29, 2019, from
https://caltc.ca/2018/09/opinion-theres-a-human-resources-
crisis-in-seniors-care/
Glavinovic, T., Ferguson, T., Komenda, P., Rigatto, C., Duhamel,
T. A., Tangri, N., & Bohm, C. (2018). CKD and Sedentary
Time: Results From the Canadian Health Measures Survey.
American Journal of Kidney Diseases, 72(4), 529–537.
https://doi.org/10.1053/J.AJKD.2018.03.031
Health Council of Canada. (2011). How Do Sicker Canadians with
Chronic Disease Rate the Health Care System? Results
from the 2011 Commonwealth Fund International Health
Policy Survey of Sicker Adults. In Canadian Health Care
Matters. Toronto,ON. Retrieved from
www.healthcouncilcanada.ca.
Mihailidis, A., Boger, J., Hoey, J., & Jiancaro, T. (2011). Zero
Effort Technologies: Considerations, Challenges, and Use
in Health, Wellness, and Rehabilitation. Synthesis Lectures
on Assistive, Rehabilitative, and Health-Preserving
Technologies, 1(2), 1–94.
https://doi.org/10.2200/S00380ED1V01Y201108ARH002
Ministry of Health and Long-Term Care. (2017). 3.10 Public
Health: Chronic Disease Prevention. Retrieved from
http://www.auditor.on.ca/en/content/annualreports/arreports
/en17/v1_310en17.pdf
Pandas. (2018). Python Data Analysis Library — pandas: Python
Data Analysis Library. Retrieved January 14, 2018, from
https://pandas.pydata.org/
PHAC. (2019). Public Health Agency of Canada - Canada.ca.
Retrieved April 21, 2019, from
https://www.canada.ca/en/public-health.html
Public Health Agency of Canada. (2017). Physical Activity,
Sedentary Behaviour and Sleep (PASS) Indicator
Framework for surveillance. Retrieved January 14, 2018,
from https://www.canada.ca/en/services/health/monitoring-
surveillance/physical-activity-sedentary-behaviour-
sleep.html
Python. (2019). Welcome to Python.org. Retrieved April 22, 2019,
from https://www.python.org/
Ricquebourg, V., Menga, D., Durand, D., Marhic, B., Delahoche,
L., & Logé, C. (2006). The smart home concept: Our
immediate future. In 2006 1st IEEE International
Conference on E-Learning in Industrial Electronics,
ICELIE (pp. 23–28). IEEE.
https://doi.org/10.1109/ICELIE.2006.347206
Silva da Silva, T., Martin, A., Maurer, F., & Silveira, M. (2011).
User-Centered Design and Agile Methods: A Systematic
Review. In 2011 AGILE Conference (pp. 77–86). IEEE.
https://doi.org/10.1109/AGILE.2011.24
Statistics Canada. (2016). Research Highlights on Health and
Aging. Ottawa. Retrieved from
https://www150.statcan.gc.ca/n1/pub/11-631-x/11-631-
x2016001-eng.htm
Tom’s Guide. (2019). Best Smart Home Technology of 2019 -
Systems, Devices and Products. Retrieved April 15, 2019,
from https://www.tomsguide.com/us/best-smart-home-
devices,review-2008.html
UbiLab. (2019). Ubilab. Retrieved April 22, 2019, from
https://uwaterloo.ca/ubiquitous-health-technology-lab/
Vimarlund, V., & Wass, S. (2014). Big Data, Smart Homes and
Ambient Assisted Living. Yearbook of Medical
Informatics, 23(01), 143–149. https://doi.org/10.15265/IY-
2014-0011
Zion Market Research. (2018). Market Size of Global Smart Home
Industry Predicted to Reach USD 53.45 Billion by 2022.
Retrieved April 15, 2019, from
https://www.globenewswire.com/news-
release/2018/08/03/1547019/0/en/Market-Size-of-Global-
Smart-Home-Industry-Predicted-to-Reach-USD-53-45-
Billion-by-2022.html