1st Project - Health Systems.
As a result of ageing population, increasing demand and evolving technology on healthcare systems, the progress in the Internet of Things (IoT) has a key role in suppressing all these needs, in particular, redesigning modern health care with promising technological, economic and social prospects. This paper attempts to comprehensively review the current research and development on the impact of IoT in Healthcare. Relying on a comprehensive literature review, this paper analyses the architecture of an IoT-based systems, focusing on the main components and their value to the overall system. In addition, a perspective on electronic health records and on privacy and security issues are overviewed, along with the review of clinical cases of IoT-based systems. Given IoT clear acceptability and affordability among youngers and elders, combined to a broad range of devices and machine learning techniques, it’s expected these devices will facilitate in many ways health providers’ job, as long as other topics like data protection keep side-by-side.
IST - 4th Year - 2nd Semester - Biomedical Engineering.
2. Motivation
Under the course “Health Systems”, we
thought it would be interesting to study the
impact of technological advances on Health,
namely, the role of Internet of Thing (IoT) in
Healthcare. IoT is an emergent technology that
is playing a key role on every industry, specially
in Health. With the skillset acquired as
Biomedical Engineers studying Medicine and
Technology, great development in IoT can be
achieved.
Introduction
With the eradication of many diseases and an
increasingly higher life expectancy (“In USA the
number of adults age 65–84 is expected to
double from 35 million to nearly 70 million by
2025 when the youngest Baby Boomers retire,
this trend is global”. [33]), the prevalence of
chronic and age-associated diseases is rising.
And, consequently, the costs in health are also
higher than ever (“Overall health care
expenditures in the United States reached $1.8
trillion in 2004 with almost 45 million
Americans uninsured. It is projected that health
care expenditures will reach almost 20% of the
Gross Domestic Product (GDP) in less then 10
years, threatening the wellbeing of the entire
economy” [33]). Important factors like, cheaper
medicines, more effective ones, preventive
care with monitoring patient’s health, play an
important role in cost containment.
The development of Internet (we
forecast there will be 34 billion devices
connected to the internet by 2020, up from 10
billion in 2015. [34]), namely Internet of Things
(IoT) (Nearly $6 trillion will be spent on IoT
solutions over the next five years. [34]), i.e. any
device connected to any network, had and has
a great impact on our daily life. With this study,
we want to understand the role of IoT in
Healthcare.
To begin with, we propose ourselves to
study the architecture of a generic residential
IoT-based System in Health Care. These
systems require the biological sensors to be
connected to an IoT-network. Thus, we will
index the types of sensors for physiological data
acquisition that are currently available, and its
wireless communication standards.
The emergence of IoT-network and the
consequent increase of data accessibility,
provides ground for the establishment of an
Electronic Health Record (EHR).
With technological advances in
sensors’ wireless communication and
miniaturization, the manufacturing of wearable
devices became possible. We will mention
some examples of commercially available IoT
wearable devices.
The devices’ interface is often
mediated through smartphone application. We
will study briefly the types of application
available related to IoT devices.
Based on the concepts introduced, real
life trials were performed in nursing homes,
where the IoT network sensors are
experimented. We will comment these results.
The implementation of these IoT
systems concerns about data security and
privacy are raised. The fundamental basis of
these concepts are discussed.
The integration of these IoT-based
systems on healthcare has proven to be
beneficial, nonetheless this technology still
needs improvement in order to fulfil healthcare
demands, and to enhance patient’s health
monitoring.
3. IoT-based Multi-tier residential System Architecture
IoT-Based Multi-tier telemedicine, based on
individual wireless entities interconnected
through Internet (IoT), delineates the principal
topic of this study. In this concept of IoT-based
multi-tier telemedicine, each tier will have its
own purpose, which, when in operation, will
allow for physicians to receive and analyse real-
time data from a patient which is outside of the
medical centre.
Based on this description of separate entities
composing an interconnected system, three
tiers can be identified considering their nature
and purpose.
Tier 1 consist on wireless sensor
devices that are integrated into a wearable
wireless body area network (WWBAN). Each
sensor node performs data collection and pre-
processing, transferring the data to Tier 2.
According to the variety of biosignals for
analysis, Tier 1 is the most customizable from
all. Each WWBAN is designed in order to meet
the physicians’ requisitions. The process of
designing a WWBAN is based on numerous
factors, where the success of each individual
component will dictate the viability of the
system.
The key factors to consider are:
wearability, aesthetic issues, data encryption
and security, operational lifetime,
computational & storage requirements,
components’ reliability, cost, interference
robustness, decision support, fault tolerance
[1]. In further detail, interference robustness is
“the reliability of wireless transmitted
physiological measurements” [1]; decision
support is the ability to interpolate the pre-
processed data pattern (recognition algorithm)
for context aware parameters; fault tolerance
is the reliability in which the system will provide
correct measurements.
In accordance with the system’s duty
and design parameters, the sensor nodes will
be selected from a wide variety, Table 1. Each
sensor will be then connected to a Personal
Server (PS) application operating on a Personal
Digital Assistant (PDA), constituting the
connection between Tier 1 and Tier 2.
Tier 2 consist on the PS and PDA. Tier 2
may also integrate wireless ambient sensor
nodes for ambient parameters managing and
analysis, e.g., motion or temperature. The
interconnection between Tier 1 and the
ambient network structures a Personal Area
Network (PAN). With the establishment of a
PAN, patients have portability inside the
network boundaries (IoT-based network
establishes connection to PDA through
Wireless technology).
In Fig. 1, PS is represented by the blue
block, which is responsible for wireless network
configuration & management, sensors’
communication & control signals and graphical
user interface. Through the system’s interface,
PS interconnects Tier 1 (Wireless sensors), Tier
2 (User’s interface and Processing) and Tier 3
(Medical server).
With modern technology,
smartphones are emerging as leading
platforms for PDA’s applications, “for
implementing personal health gateways” [2].
Among a list of factors, Smartphone have
increasing computing power and storage space,
Tier 1
Tier 2
Tier 3
Fig. 1 - Architecture of a IoT-based Multi-tier system. [1]
4. incorporated sensor capabilities e.g.,
accelerometer, support for third-party health
apps by mobile operating systems (e.g., iOS and
Android) and faster internet connection. The
increasing smartphone’s capacities have to be
supported by the battery capacity, which has
been outpaced when in comparison with other
smartphone’s constituents. This is an important
factor when considering smartphones as PDA
because battery autonomy limits the system
continuous operation.
In order to establish a connection
between Tier 1 and Tier 2, there’s a process
which PS is responsible for: sensor node
registration; initialization (e.g., specify sampling
frequency and mode of operation); secure
communication setup. Once the network is
configured and established, Tier 2’s processing
is responsible for data retrieval, time
synchronization and data fusion. Note that
these processes are assured by wireless
connection, e.g., Bluetooth, 4G, GPRS, WLAN.
As data is processed, a connection to
the medical server is established through a
secure channel, enabling an exchange of data
between Tier 2 and Tier 3 (bidirectional data
transmission). Tier 2 is able to locally store
information in case of malfunction/inaccessible
medical server, and to initiate data upload once
the connection is possible.
Before describing Tier 3, it’s important
to identify and categorize each device (e.g.,
ECG sensor, smartphone) based on its
interconnection capability due to the wide
variety of devices with increasing complexity
(Fig. 2):
1. Devices with incorporated mobile module:
these devices have the capability to establish a
wide area of connectivity via the mobile radio
access network (e.g., LTE, 4G). Smartphones
are the main devices in this category;
2. Devices without incorporated mobile
module: these devices are dependent on an
external device (e.g., health hub) to establish
wide area of connectivity. These devices are
only able to establish small-scale area of
connectivity via short-range radio (e.g., LAN);
3. Health hubs: these devices receive data
from devices without incorporated mobile
module and establish wide area of connectivity
(e.g., Samsung SmartThings Hub [3]).
Tier 3 consist on the medical server
(main receiver of data) that is connected to the
IoT-system, as mentioned before. The primary
Tier 3 data receiver are Hospital/health centres
(medical servers), but data may also be used by
e.g., emergency care providers, insurance
companies or health data repositories.
A fundamental breakthrough in health
care systems technology is the use of Electronic
Health Record (EHR). With this Multi-tier
telemedicine system, users can automatically
and continuously update, access, review their
EHR, while health care providers receive real-
time data. The conjugation of medical expertise
and computational power allows for data to be
analysed in order for patter recognition, health
condition recognition and emergency situation
detection.
5. Physiological Signals and Biosensors
As components of Tier 1 of IoT-Based Multi-
tier systems, these types of sensors are
responsible for the acquisition of real-time
physiological parameters, which evaluate
the user’s health condition at any time.
The following table (Table 1) indexes
the several types of sensors for biological
data acquisitions.
Fig. 2 – Representative of IoT-based residential system. [2]
Table 1 - Type of Bio-signals and Sensors. Description of data acquisition. [1]
6. Wireless Communication Standards for WHM
As an attempt to overcome the problems
with current healthcare systems, previously
mentioned, the development of Wearable
Health-Monitoring Systems (WHMS)
represent a key role for continuous patient
monitoring, which “aim at providing real-
time feedback information” [1] about a
patient’s health. The data is handled not
only by the user but can also go straight to a
health professional or a medical centre.
The biosensors systems (Tier 1)
have the purpose of measuring physiological
parameters. The collected information is
connected to a central node, through a
wireless or a wired link, and may be
displayed on a user interface, like a
smartphone (Tier 2), or be transferred to a
healthcare facility. These systems integrate
several components: sensors, wearable
materials, actuators, power supplies,
wireless communication modules and links,
control and processing units, interface for
the user, software, and advanced algorithms
for data extracting and decision making.
The transmission of the collected
data is performed in 2 steps. The first one,
for the communication between biosensors
and the central node (communication
between Tier 1 and Tier 2) and the second
one for sending the data from the wearable
system to, for instance, a medical centre
(communication between Tier 2 and Tier 3).
Regarding the first step of the data
transmission, it is managed by wires or by
multiples wireless links. Using wires
seriously decreases the user’s mobility and
comfort and there’s a higher chance of
failure. Autonomous sensor nodes can form
a body area network (BAN) or body sensor
network (BSN), transmitting the data to the
central node of BAN central node (e.g., a
smartphone).
We will only discuss the short-range
wireless communication technologies used
for intra-BAN communication (Table 2). The
most used ones are IEEE 802.15.1
(Bluetooth).
Bluetooth is usable for a short-range
RF-based connectivity between portable
and fixed devices. Among the several
characteristics, it is a low-power and low-
cost RF standard. “Bluetooth operates in 2.4
GHz spectrum with a frequency hopping
technique over 79 channels, that may
support up to 3 Mb/s in the enhanced data
rate mode and maximum transmission
distance of 100 m” [1]. It supports
encryption; however, it is an optional
feature. In addition, the Bluetooth
framework may be susceptible to attacks
and risks. Recently, there are another two
protocol stacks: Bluetooth low energy
(ultralow power technology for devices with
limited battery capacity), which focus on
sports, wellbeing and medical devices, and
Bluetooth 4.0 (for higher data transfer rate).
Medical Implant Communication Service
(MICS) is another technology for short-range
intra-BAN communication for transmitting
low-rate data to help therapeutic function in
medical devices, like pacemakers and
defibrillators. It is ultralow and uses 402–
Table 2 - Wireless Communication Standards for WBAN. [1]
7. 405 MHz frequency band, with 300 kHz
channels. It isn’t commonly used because of
the lack of commercially available MICS
solutions. Ultra Wideband (UWB), the same
type of technology, works in a frequency of
3.1–10.6 GHz. Nonetheless, it isn’t adequate
for BANs by cause of its high complexity and
unsuitable wide bandwidth modulation.
Finally, another alternative is infrared (IrDA),
also a low-cost technology, that uses
infrared light for the short-range
communication. It deals up to 16 Mb/s
speed, but it needs line-of sight
communication, which isn’t efficient in
WHMS.
From the previous technology, we
conclude that there is still improvement
needed to be done, e.g., interference
problems, security issues or power lifetime
in order to entirely fulfil the BANs’
requirements.
EHR – Electronic Health Record
An electronic health record (EHR) is “a
repository of electronically maintained
information about an individual’s lifetime
health status and health care, stored such
that it can serve the multiple legitimate users
of the record” [23]. This system “adds
information management tools to provide
clinical reminders and alerts, linkages with
knowledge sources for health care decision
support, and analysis of aggregate data both
for care management and for research” [23].
Such system allows a health care from
wellness to illness and recovery – continuum
and managed care. It presents linkages and
tools to help in communication and making
decisions.
Unlike paper-based record, EHR is
flexible and adaptable, since it keeps data in
a single copy and in a single format, which
allows the communication between
electronic devices/systems. As its first
purpose, EHR gives to all authorized
personnel an integrated access to the
patient data, at different medical centres,
emergency rooms, offices or even at home.
Apart from numbers and text, it can store
data such as graphs, videos and images, that
were not possible using paper record. These
systems introduce many more benefits,
including legible and coherent information
(instead of hand writing) and reusable data.
However, we can also identify some
drawbacks about EHR when comparing to
the use of paper, such as the initial
investment, the time required to learn how
to deal with the system, the security issues
or the requirement of contingency plans in
case of failure, among others.
We can distinguish five different
functional components that integrate such
systems: integrated view of patient data,
clinical decision support, clinician order
entry, access to knowledge resources,
integrated communication and reporting
support. Regarding the first one, the
collection and organization of data establish
very difficult tasks due to the complexity and
variety of data and the abundant number of
patient data sources (clinical laboratories,
radiology departments, etc.). In addition,
each one of these source systems use their
own identifier, terminology and data format.
Thus, each administrator of an EHR system
must adapt the data coming from various
sources to their format and codes.
Furthermore, physicians need not only
integrated access to data but also to be able
to deal with it in different views (for
example, chronologic order).
About the clinical decision support,
the EHR systems provide support through
8. the access to a “theory/hypothesis” with
recommendation, but still allows the
physician to make the final decision. It
consists in a guidance/suggestion of action
(reminders and alerts) and, at the same
time, educate the user. At the end, the
feedback given by the physician upgrades
the program and clinical protocol. Today,
there’s an easy access to a vast knowledge
sources available in these systems, that
helps a clinician at decision making and
avoid spending time searching on
literatures.
As mentioned above,
communication tools are an important
function of such systems – they can improve
the regulation of care and the disease
management. Nowadays, the health care of
a patient is assigned to several professionals
(sometimes not all of them at the same
facility), and how they communicate with
each other and how they have access to
other’s reports affects the care given to the
patient. “Connectivity to the patient’s home
will provide an important vehicle for
monitoring health (e.g., home blood-glucose
monitoring, health status indicators) and for
enabling routine communication” [23].
Regarding data capture, we can
distinguish two methods: electronic
interfaces and manual data entry. Enforcing
electronic interfaces between EHR and
electronic data sources is the main method
for capturing data. They grant an almost
immediate availability of the data and
prevent labour costs and manual errors.
Database interface engines “not only
provides message-handling capability but
can also automatically translate codes from
the source system to the preferred codes of
the receiving EHR” [23]. The data-entry step
is a time consuming process: besides the
actual part of enter the data, it requires its
interpretation or translation. Data is
introduced in a text form, coded form or the
combination of both. Using codes, the data
is standardized and classified, which
facilitates computer processing. Physicians’
notes are entered using one of the three
mechanisms: transcription of dictated or
written notes (most common due to its
comfort, according to the physicians),
structured encounter forms, or direct data
entry.
In order to avoid errors (for
instance, of transcriptional matter), EHR
exhibit several types of checks to analyse the
medical data: range checks (detect and
prevent values out of range), pattern checks
(recognize if data is introduced according to
a certain pattern), computed checks
(validate correct mathematical
relationships), consistency checks (identify
errors by comparing data), delta checks
(identify unlikely variations between the
new and previous values) and spelling
checks.
To conclude, it is important to
acknowledge that the concept EHR isn’t
consolidated or stationary, it is still evolving
and expanding, in both hardware and
software technologies. Concerning the
purpose of this paper, data should be
recorded into an EHR system in a way that
allows the access and communication to
every authorized people, anywhere,
recorded by any device (IoT concept). It also
should enable the access of the patient in
order to record every single measure and
vital sign, either at home or in a medical
centre (by a health care professional).
The following application integrates the
concept of EHR in a medical device for
patient’s health monitoring.
SiOne Smartinjector (Fig. 3), released in
2017 (by QuiO, CEO Alexander Dahmani) is
the new smart injection administration
device, consisting on the Tier 1 of a IoT-
9. Based Multi-tier system, that provides real-
time monitoring remotely (e.g., at home),
designed to overcome self-administration
issues related to the application of incorrect
doses.
One of the big novelty of this
technology it the fact that it supports any
type of syringe, unlike its competing devices.
The administered dose, as well as the time
at it was taken or any eventual errors, is
being constant monitored by a medical
clinician and recorded through the
connected QuiO Cloud HIPAA-compliant
software platform. This cloud depicts the
Tier 3 of a IoT-Based Multi-tier system. Its
(cellular) connection to the cloud requires
no setup, syncing or extra components. To
follow up the patient health situation, the
program is composed by dashboards, used
to track the administration and to “monitor
drug performance, adherence, and
outcomes” [24].
Wearable Devices
This type of devices has already been
discussed, as a component of a IoT-based
Multi-tier system. However, due to its
characteristics, “these devices can be viewed
as IoT innovations that can lead to various
healthcare solutions” [6], and will be further
analyzed. Regarding the architecture and
design of the Wearable devices, there are
some medical criteria that need to be
fulfilled, as well as ergonomic constraints
and hardware limitations. In particular,
radiation concerns; esthetical issues; high
operational lifetime; security and privacy of
the collected data; affordable; and should be
light-weighted and with small size in order to
avoid blocking any user’s movement. As we
can understand, there is no specific design
for such systems due to all the medical
requirements – should be according to the
area of application.
Data transmission via wires
seriously decrease the user’s mobility and
comfort, increasing the chance of failure.
Due to this, for these types of devices,
wireless transmission should be considered
as the main procedure for data transmission.
When considering IoT-based wearable
devices, the number of applications and
prototypes being developed is numerous,
being impossible to mention every type.
According to this, some of the devices that
already have commercial applications and
represent IoT innovations will be
mentioned.
Fig. 2 - QuiO SiOne Smartinjector.
Fig. 4 - IoT system evidencing wearable applications and EHR
technology.
10. Artificial Pancreas is an application
that targets diabetes’ patients. The
equipment automatically controls blood
glucose level by mimicking pancreatic
endocrine functionality. The main endocrine
function of the pancreas is insulin
production. The device has as constituents a
blood glucose sensor, insulin and amylin
pump and a connective module. Blood
glucose control reveals patterns which can
improve insulin therapy in order to give
more comfort to the patient. The insulin and
amylin pump delivers the hormones to
bloodstream automatically according to the
Blood glucose sensor readings. The
connection module establishes the
connection between the device and the
multi-tier system, as illustrated in the
following picture:
The device presented in Fig. 6 is
already commercially available. MINIMED
670G SYSTEM operates in a close loop
providing continuous glucose blood control,
with a user-friendly interface.
Electrocardiogram monitoring is an
application that is based on the “electrical
activity of the heart recorded by
electrocardiography, includes the
measurement of the simple heart rate and
the determination of the basic rhythm as
well as the diagnosis of multifaceted
arrhythmias, myocardial ischemia, and
prolonged QT intervals” [8]. The device
continuously monitors the heart’s activity
transmitting the data to a medical server
(IoT-based multi-tier system). With real-time
data, patient’s healthcare can improve
greatly, allowing for doctors and patter
recognition software’s to foresee heart
conditions.
The device presented in Fig. 7 is
already commercially available. Quardiocore
is a multifunction device, monitoring,
besides continuous wireless ECG, heart rate,
heart rate variability, skin temperature,
respiratory rate and activity tracking. This
device also includes a connection module in
order to establish connection to a PDA (only
IOS operating system supported),
constituting the first tier on a IoT-based
multi-tier system. The Qardio company has
established a free medical server, in which
the patient’s doctor signs-up and has access
automatically to the patient’s device data.
Oxygen saturation monitoring is a
suitable application for IoT systems. “Pulse
oximetry is a noninvasive method for
monitoring a person's oxygen
Fig. 6 - MINIMED 670G System. [12]
Fig. 7 – Quardio Quardiocore. [9]
11. saturation (SO2) [10]”. The device estimates
oxygen saturation with the emission of light
through the body part to a photodetector,
by measuring absorbance variances. This
application operates similarly to other IoT
applications, such as, it operates
continuously, measuring and transferring
real-time data to a medical server through
an IoT network.
The device present in Fig. 8 is
already commercially available. OXITONE
1000 is wrist pulse oximeter without
fingertip probe, which provides SpO2 and
pulse rate readings. This device includes a
communication module, in order to
establish connection to a PDA (IOS and
Android operating system are supported
through an App). The App allows Patient
symptoms self-assessment, Real-time digital
biomarkers and activity tracker, Patient
Dynamic Health Status and generating and
transmitting health reports to a medical
server established by the OXITONE
company. The medical server implemented
enables the following tools for physicians:
Real-time data analytical tools, EMR
integration and APIs, Data delivery on
demand and communication tools with
patients. Many companies are developing
Smart watches with incorporated medical
sensors, e.g., temperature monitoring and
movement tracking. We chose not to
mention this technology due to not meeting
medical accuracy standards for data
acquisition.
The employment of sensor-type
applications in IoT network, as described in
the three examples above, shall result in an
increase in healthcare quality.
Apps
Another important strand of IoT are the
smartphone applications.
According to ANACOM [12], the smartphone
penetration in Portugal was 68.8 percent in
September 2016 and the tendency is for it to
increase.
The easiness of access to applications
represents an advantage for network
establishing between tiers of an IoT-Based
Multi-tier telemedicine system.
It is possible to distinguish the health
related Apps into two groups: on one hand the
apps connected to home sensors or wearable
devices which store daily measures and that
can regularly upload the information to a
doctor; On the other hand, the apps whose
main goal is to detect anomalies and alarm the
family members and/or the medical facility. We
will present a specific example for each group,
for the first group, Withings Health Mate and,
for the second, Nursy.
Withings is a company of electronic
devices, part of Nokia, that provides several
home sensors for health parameters. One
example is the Body Cardio, a weight balance
that besides weight, body mass index, body fat,
water percentage, muscle mass and bone mass,
can also measures standing hearth rate and
pulse wave velocity (a cardiovascular health
parameter). Body Cardio stores the measured
Fig. 8 - OXITONE 1000. [11]
12. information via Wifi or Bluetooth in Health
Mate App allowing the user to monitor his
health parameters. The user has the possibility
to send the data to his doctor (Tier 3) by e-mail
(this data transfer is not done automatically).
Most wearable devices (Tier 1) can be
connected to similar apps that allow the user to
store, treat, transmit his data and update his
EHR. These apps require some familiarity of the
user with IT interfaces, which can be an
obstacle for elderly users.
Directed to the elder population are
the apps on the second group, like Nursy. This
app is composed of three components: the
patient application, the family member or care
supervisor application and the doctor backend
application. The patient component reminds
him of taking his medications and of his
doctor’s appointments and allows him to notify
the care supervisor he is carrying out the
reminders. Nursy can also be connected to a fall
detection system and alarm the supervisor
whenever a fall is detected. The doctor is able
to upload the patient records (EHR), request
him an appointment and add new
prescriptions. To overcome the unfamiliarity of
the elder with IT, the patient component has a
very simple user interface. This application was
developed by Accenture using SAP technology
but is still in a finishing period and is not yet
available in App Stores.
Health related App development holds
great potential as a strategy to continuously
manage data while overcoming portability of
systems, and, overall, increase on health care
quality.
13. Perceived need and preferences for smart home tech [+ Clinical Case]
Most studies regarding the usefulness of the
introduction of monitoring sensors in human
life were done in a senior housing site, named
TigerPlace (Fig. 9). It comprises independent
living apartments (studio, one or two bedroom
apartments) located in a restricted area of
Columbia, USA. This is much related with the
concept of aging in place. Where residents can
preserve their own independency, having
assured, at the same time, permanent
assistance. Meals and hygiene related issues
are some of the key services provided by health
carers in TigerPlace, in a daily routine.
Demiris et al. [26], to guaranty the
effectiveness of the interaction between users
and all kind of sensors, determined how
TigerPlace senior residents perceive these
embedded sensors in their own homes.
Each group session started with the
facilitator explaining the objective of the study
and that each session would take about 1 hour.
Participants were asked to touch/interact with
the sensors, after a brief introduction to its
function and exampling, provided by the
facilitator. To ensure the reliability of the
method used, details like the duration of each
session of questions and the type of wording
included in the protocol were previously
studied in a different senior residence group of
people. Similar 2 groups of people were
essential to provide an accurate insight about
the protocol’s validity.
In-home monitoring systems, event-
driven anonymized video-sensor and activity
analysis were the central topics approached to
assess usefulness/privacy related issues.
Questions pertained the advantages and
disadvantages associated to these systems,
participants’ willingness to install it in their own
homes and opinions about who should have
access to the data obtained were the core of
the discussion. Participants were also warned
that audiotapes would be used to record the
sessions and for data analysis purposes.
After joining 14 adults, 3 different group
sessions were organized. Each lasted an
average of 64 minutes. Five participants were
male and 9 were female. All of them with more
than 65 years old.
Bed sensors were, generally, perceived as
useful. Although, one participant stated that
she relies on her spouse to detect some
restlessness problems. Most of the participants
considered the stove sensor as secondary
because they don’t cook. Being the meals
provided by the care centre. Gait monitor was
perceived, overall, as very useful. Most
participants expressed concerns about being
alone and helpless after dangerous fall. Motion
sensor was considered to be more useful to
detect intruders’ activity than to monitor
participant’s activity level. Finally, video sensors
were the one that arose more privacy concerns.
10 out of 14 wouldn’t want to have such
devices in their homes, although they only
capture a silhouette.
Findings suggested that most of the
elders perceived these devices as useful to
detect emergency situations rather than
prevention. Participants suggested that
sensors’ shape and size were important
features to take in account before installation.
Moreover, one addressed the issue of
stigmatization, saying: “as long as it is installed
in the others’ [apartments], as long
Fig. 9 - TigerPlace, at the University of Missouri –
Columbia. [31]
14. as it would be something they were going to use
all over and I would not be different…” [26].
Many subjects found the technology
useful for people in more advanced frailty
conditions. One said: “I don’t need this now, but
perhaps at a later point—I have friends who’d
benefit from this a great deal, I am not there yet
...”. Another patient who had experienced fall
in the past stated that: “If you had told me 2
months ago [about these technologies] I’d say
who needs it, but after what I have been
through, I see the benefits.” [26].
When asked about who should be able
to access data, they mentioned healthcare
providers, as well as their families. One patient
added that would like to have control over the
amount of data that is shared and to have
access to it, before anyone else.
2 participants expressed concerns
about the accuracy of each sensor. Considering
false alarms, a relevant point to consider as
they could lead to cumbersome situations for
patients and staff.
Overall, participants showed a positive
attitude towards the adoption of these
technologies. Fall detection sensors were the
ones perceived with more advantages by the
patients. Half of them, clearly said that they
would adopt these smart sensors in their own
homes. The others evidenced concerns about
privacy intrusion, although most were
receptive to find an equilibrium point between
privacy and the level of need.
15. Clinician Assistance [+ Clinical Case]
After assessing how elders (majority of users in
a near future) perceive in-home sensors and
approve them, it was important to test them in
real situations. Namely, to understand how do
they work in detecting various pathologies.
Motoi et al. [27] contributed with some
techniques to detect, in an early stage, some
pathogenic/uncontrolled conditions (life-style
related diseases: adiposis, diabetes,
cardiovascular). Non-conscious physiological
monitors were installed in a toilet-bowl and a
bed to simultaneously record parameters
directly from the sensors in contact with the
body surface. 3 patients with a history of
cardiac infarct (1) and sleep apnoea syndrome
(2) carried out the study for 7 days.
To proceed the study, a proper
prototype health-care monitoring room was
developed in Imizu City Hospital (Japan). Where
they embedded a respiration and pulse
monitor, using air mattress sensors; a
thermistor, also in the bed, to measure body
(nasal) temperature; body weight and
excretion weight balance (difference between
initial and final weight, Fig. 10) around the
toilet-bowl (very accurate device) and to
measure blood pressure, a toilet-seat sensor
was also installed.
A 65-year-old patient (male), who had
a recent cardiac infarct, and 2 patients with SAS
(45 and 56 years, male and female,
respectively) were monitored during the
experiment. First one had its own weight
registered during a week, as well as the
excretion weight. The control of these 2
parameters plus the traditional info already
available in every care unit (heartbeat, oxygen
saturation…), allowed a more precise control of
his evolution and, consequently, stabilization of
his health status.
1 out of 2 patients (45 years old), who
suffered from apnoea/hypopnea had his breath
rate registered, in addition to his nasal
temperature (Fig. 11). Dashed rectangle
corresponds to the time interval when breath
partially cessed for more than 10s (definition of
apnoea).
A compilation of the AHI (Apnea–
Hypopnea Index) for both patients is shown in
Fig. 10 - Monitored excretion and body weight. [27]
Fig. 11 - Breath rate (top) and nasal temperature
(bottom) record. [27]
Fig. 12 - Number of apnea/hypopnea events. [27]
registered for 1 night. [27]
16. Fig. 12, which summarizes the number of
episodes occurred during the whole night (~8
hours).
In relation to this setup, the users could
live in a regular way and without consciously
feeling that some measurements are being
captured regarding his health. After that, data
could be gathered and further sent to
specialized health carers.
Fig. 13 - Hospital room spatial overview. [27]
Fig. 14 - Toilet-bowl installed sensors. [27]
17. Supervised & Unsupervised Learning [+2 Clinical Cases]
Detecting Physical Impairment
Apart from gathering data from a set of
sensors, process it using machine learning
techniques will greatly reduce the time of
diagnose. Skubic et al. [28] started by displacing
a set of sensors through one bedroom
apartment (Fig. 15). 11 motion sensors, bed
sensor and a temperature sensor to capture
stove and oven activity were installed after
elder’s approval. Features extracted from the
motion sensors were: activity in bedroom,
bathroom, living room, kitchen, time out of
home, with visitors and total level of activity
estimated from motion density. Additionally,
bed sensors displayed information related to
restlessness in bed, heart and respiratory
events.
When events took place (sensor’s
unusual signal), a clinician was alerted and
asked to rate it in a scale from 1 to 5 the
severity of the situation. Then, in a
retrospective analysis, after looking to
hospitalizations, emergency room visits and
falls happened through the same period,
clinical researchers developed potential
algorithms to correlate sensors data
peculiarities to life threatening or injury related
accidents. Several machine learning
approaches were used. Furthermore, manual
classification based on clinician and researchers
experience was developed and later compared
to these supervised learning methods.
Using the feature selection method,
they determined which set of events better
identified some health impairment. Every
feature was tested individually and the best
one was chosen. After that, additional feature
combinations were tested until the
performance stabilizes or decreases. Table 3
shows that bathroom visits, visitor activity and
sleep patterns should be preferentially used to
detect health emergencies for these 2 cases.
The alert algorithm was developed
using retrospectives analysis, as well as, clinical
collaboration. A researcher manually reviewed
sensor data leading to health events. Then, a
set of algorithms were tested until consensus
was reached.
Fig. 15 - Sensors’ location in the apartment. [28]
Table 3 - Best set of features used to identify
abnormal events (for 2 different cases). [28]
18. Table 4 shows the alert parameters
agreed to monitor health after a collaborative
analysis.
Each resident has a personalized
normal activity, so different baselines were
established to guarantee that proper deviations
to mean values were detected. Each feature
had a mean value and a standard deviation
calculated after 15 days gathering data. Once
this is a 1D approach, if one feature had its
value increased or decreased more than a
predefined number of standard deviations, an
alert was generated. Using this strategy half of
the events were false alarms.
Clustering analysis were performed
using 4 of the 6 features described in Table 4 as
some of them, typically, didn’t generate
enough alerts for supervised learning. 4 alert
parameters: bathroom activity, living room
activity, bed restlessness and kitchen activity.
Considering increasing and decreasing changes,
as well as for 3 different time periods (daytime,
night time and full day), a 24D space was
generated (Fig. 16). Space represented
considering only increased values at the same
figure and another one narrowed to 6 different
features: increasing nighttime activity in the
living room, kitchen, and bathroom, increasing
full day activity in the bathroom, and increasing
bed restlessness at both nighttime and during
the full day. To represent such high dimensional
spaces in 3D a PCA reduction was performed.
Normal days (blue crosses) tend to
cluster and abnormal days (red circles) to
appear as outliers. All features were normalized
before including in classifiers.
Furthermore, 4 different classifiers
were used to create an algorithm which
predicts the severity of each situation: fuzzy
pattern tree (FPT) exclusively based on clinical
judgment. The remaining are fuzzy K-nearest
neighbor (FKNN), the neural network (NN), and
the support vector machine (SVM). Table 5
summarizes the accuracy and the percentage
of false positives and negatives for 6 and 12
dimensions. 6-D FPT and 12-D FKNN were the
tests that returned better results.
At the end, patients suffering of urinary
tract infections, pneumonia, upper respiratory
infections, heart failure, post-hospitalization
pain, delirium, and hypoglycemia saw their
health condition enhanced.
Table 4 - Set of 6 alerts that best detect uncommon
events. [28]
Table 5 - Comparison values for the 4 different
techniques used to generate alerts. [28]
Fig. 16 - Cluster analysis performed using 6/12/24 different variables. [28]
19. Detecting Mental Impairment
Dawadi et al. [29] proposed an additional
application to the smart home health equipped
sensors. Since mobility patterns are influenced
by cognitive abilities, motion sensors on the
ceiling, door magnetic sensors on cabinets and
doors, item sensors on selected kitchen items,
temperature sensors in each room, sensors to
monitor water, burner use and power meter
sensors to determine the electricity
consumption, were installed in a different
environment: Washington State University
CASAS, to predict cognitive debilities (dementia
or MCI, which stands for Mild Cognitive
Impairment) based on certain behaviour
characteristics registered with the previous
instruments. Each sensor is electronically
identified by: date, time, sensor identifier and
sensor message (Fig. 17).
Input data acquired (saved in a SQL
database) will be a sequence of sensor events
E, sub divided in e1, en… This signal identifies a
task being performed by the participant that
can be decomposed in subtasks: A1…An. Activity
subtasks can be initiated in a random order,
what matters are the different interwoven
activities being done. Mistakes like forgetting to
turn off the burner or taking a long time to
complete a simple task, may indicate some
health condition. Machine learning techniques
were then used to automatically quantify the
quality of a performance related to an activity,
comparing to other individuals.
In order to gather data to learn and test
algorithms, patients were asked to perform a
complex task. The objective was to imagine that
they wanted to meet a friend in a museum at a
certain hour and, after that, they would dine
with him at his home. So, they needed to
perform a DOT (Day Out Task) to ensure
everything go as planned:
1. Magazine: Choose a magazine from
the coffee table to read on the bus ride.
2. Heating pad: Microwave for 3 minutes
a heating pad located in the kitchen
cupboard to take on the bus.
3. Medication: Right before leaving, mime
taking motion sickness medicine found
in the kitchen cabinet.
4. Bus map: Plan a bus route using a
provided map, determine the time that
will be needed for the trip and calculate
when to leave the house to make the
bus.
5. Change: Gather correct change for the
bus.
6. Recipe: Find a recipe for spaghetti
sauce in a book and collect ingredients
to make the sauce with a friend.
7. Picnic basket: Pack all the items in a
picnic basket located in the closet.
Exit: When all the preparations are made,
take the picnic basket to the front door.
Fig. 17 - Date, time, sensor ID and message emitted
by each sensor. [29]
Fig. 18 - Sensors’ setup used to detect mental
impairment. [29]
20. Dawadi et al. [29] observed the manner
179 patients act during the experiment
described above. 14 had to be excluded from
the study, as they didn’t conclude at least 2
subtasks (baseline criteria established at the
beginning). Moreover, the mean time needed
to perform the 8 steps was 10.33 ± 3.85
minutes.
Task scoring was given in a scale from 1
to 6, such that the last value means the activity
wasn’t performed in a regular way. This
evaluation was attributed by a
neuropsychologist, based on the time needed
to complete an activity (which is influenced by
their ability to multitask) and the quality of the
same. Joining this classification to the data
obtained from the sensors, supervised learning
algorithms were used.
Just as Skubic et al. [28], they also made
use of unsupervised learning to split all the
results in 3 different groups (healthy, MCI and
dementia). With this technique, it was possible
to assign to each group, what type of sensors
were usually active or not and correlate it with
their mental health. One important source of
information was for example the one present in
Fig. 19.
The correlation was finally assessed
using a coefficient of determination, which
value was registered has being 0,62. This means
that 62% of the variations in the dependent
variable can be explained by the variation in the
independent one.
Sources of uncertainty in the results,
were identified by the small number of
participants and neurophysiologists, which can
lead to some biased results. Future studies that
includes the use of wearables are also thought
to provide more accurate results in the future.
Fig. 19 - Sequence score attributed to each patient
vs health condition. [29]
21. Privacy and Security issues
Despite the broad consensus that IoT will
deliver a great value to the Healthcare sector
[13, 14, 15], this emergent technology still faces
one big challenge before its wide
implementation.
If it is true that generating large
amounts of data will enable large-scale
statistical studies and the finding of
unsuspected correlations through machine
learning algorithms, it is also true that large
amounts of data will raise privacy issues.
Although it is overwhelming that an in-body
medical device can be remotely accessible (e.g.,
to perform a software update on an aged
pacemaker, without the need for a surgical
procedure), this also means that in-body
medical devices can be remotely hacked.
In fact, these topics didn´t miss the eye
scope of policy makers. In 2013, the European
Commission issued a document [16] about
Privacy and Security in IoT. In this document are
listed the following objectives to be attained in
the design of any IoT system: the right of
deletion; the right to be forgotten; data
portability (the possibility for the user to
change IoT provider and “carry” his data with
him, this requires some compatibility between
different providers); privacy and data
protection principles.
We note, however, that fulfilling
simultaneously all these objectives might be
difficult. For example, if the patient data are to
be integrated in a database, or cloud, in order
to perform statistical studies or run machine
learning algorithms, this data integration
process should be done in an anonymous way,
so that it is not possible to track the data back
to the patient. The anonymity ensures privacy
but somehow prevents the possibility of data
deletion by the patient (once the individual
data are uploaded to the database/cloud, since
there is no way to link the data back to a specific
patient, the user may erase the data in a local
level: in his device; but not in a higher level).
In the above cited document are
studied the following 4 possible lines of
regulation: Do nothing. Soft law / Self-
regulation. Co-regulation. Binding law.
The last option, an agreement
consciously made between the IoT service
provider and the user/patient, where certain
actions are either required or prohibited, was
considered the most preferable. Such an
Agreement would be similar to the Terms and
Conditions every Smartphone user has to agree
with before installing any new application. A
compromise between user and provider parts
may help solving the privacy issues. In the
document, the writing commission recognizes
that “this [binding law] should be accompanied
by effective and efficient means of data
protection enforcement” [16]. The privacy issue
has been tackled, is left for others to present
solutions for the security one.
Nevertheless, in general, the IoT was
found to be such a hot issue that, in early 2015,
the European Commission founded AIOTI
(Alliance for Internet of Things Innovation), an
organism whose mission is “to contribute to a
dynamic European IoT ecosystem” [17].
Security hasn’t gone unnoticed either.
In fact, the great companies of software and
communication protection are allocating
resources to find solutions to the IoT security
issue (McAfee [18], wolfSSL [19], Symantec
[20], Arxan [21]).
The McAfee report of March 2015 [22]
emphasizes that: security must be kept in mind
from the conception of the IoT system to the
development of the networks and devices,
rather than being postponed as an
afterthought; the regulation paradigm for the
approval of medical devices needs to evolve, to
incentive innovation and protect the public
interest; security must be taken into account in
the whole healthcare system, from the device,
to the network, to the data center.
22. In the concrete case of the previously
analyzed Multi-tier residential system, “the
problem of security arises at all three tiers of a
WWBAN-based telemedical system” [1],
however, for the small number of nodes in a
typical WWBAN-based telemedical system and
the short communication ranges this shouldn’t
be a hard problem.
23. Conclusion
We have been watching an exponential
development on the types of internet
connected devices that can serve as biological
sensors, in fact, the technological evolution is
such that several classes of these sensors have
been implemented in wearable devices. We
have also seen that a great number of solutions
for the devices’ wireless communication is
available, at affordable prices, and covering
different engineering requisitions (like range or
data transmission). Allowing the designing of
solutions specialized to each different problem.
The wide offer of sensors and sensors’ wireless
communication protocols, at low costs, makes
an Internet-of-Things residential based system
economically viable. We have studied one
possible architecture for these systems which
could be a great solution for the integration of
the data collected from many different home or
personal acquisitions systems. Data integration
and transmission is essential for the
composition of an Electronic Health Record
which will provide integrated access to
patient’s data from any Internet accessing
device, among several other advantages
previously mentioned. Empowering people to
manage the entire network from something
that has already entered the daily life of most
of the Portuguese population which is the
Smartphone. The interaction of the user with
the Personal Server operating in the
Smartphone must be mediated through an app
specifically designed for that effect. The
available diversity of home or wearable sensors
accurate at the medical level and at affordable
prices, the existence of healthcare specific apps
that integrate and treat health parameters
gathered in sensors and the case studies
monitoring elder people using a network of
connected home sensors and machine learning
algorithms with positive feedback from both
patients and clinicians are good indicators that
Internet of Things in Healthcare has approved
in the first steps. Regulators, like the European
Commission, are already attempting the next
big step: the integration of all these small
already existing systems at a higher
(governmental) level. Problems like privacy and
security concerns arise, although a first solution
to the privacy issue has already been found, the
security problem remains an open problem
which many software security companies are
trying to tackle. We think that IoT in Healthcare
has a great potential and will bring a great value
to this sector in a near future.