JMIR MEDICAL INFORMATICS Fernandes & Lucena

Original Paper

A Software Framework for Remote Patient Monitoring by Using

Multi-Agent Systems Support

Chrystinne Oliveira Fernandes*, MSc; Carlos José Pereira De Lucena*, PhD

Department of Informatics, Pontifical Catholic University of Rio de Janeiro (PUC-Rio), Rio de Janeiro, Brazil
*
all authors contributed equally

Corresponding Author:
Chrystinne Oliveira Fernandes, MSc
Department of Informatics
Pontifical Catholic University of Rio de Janeiro (PUC-Rio)
RDC Bldg, 4th Fl
225 Marquês de São Vicente St
Rio de Janeiro, 22451-900
Brazil
Phone: 55 521 3527 1510
Fax: 55 21 3527 1530
Email: [email protected]

Abstract
Background: Although there have been significant advances in network, hardware, and software technologies, the health care
environment has not taken advantage of these developments to solve many of its inherent problems. Research activities in these
3 areas make it possible to apply advanced technologies to address many of these issues such as real-time monitoring of a large
number of patients, particularly where a timely response is critical.
Objective: The objective of this research was to design and develop innovative technological solutions to offer a more proactive
and reliable medical care environment. The short-term and primary goal was to construct IoT4Health, a flexible software framework
to generate a range of Internet of things (IoT) applications, containing components such as multi-agent systems that are designed
to perform Remote Patient Monitoring (RPM) activities autonomously. An investigation into its full potential to conduct such
patient monitoring activities in a more proactive way is an expected future step.
Methods: A framework methodology was selected to evaluate whether the RPM domain had the potential to generate customized
applications that could achieve the stated goal of being responsive and flexible within the RPM domain. As a proof of concept
of the software framework’s flexibility, 3 applications were developed with different implementations for each framework hot
spot to demonstrate potential. Agents4Health was selected to illustrate the instantiation process and IoT4Health’s operation. To
develop more concrete indicators of the responsiveness of the simulated care environment, an experiment was conducted while
Agents4Health was operating, to measure the number of delays incurred in monitoring the tasks performed by agents.
Results: IoT4Health’s construction can be highlighted as our contribution to the development of eHealth solutions. As a software
framework, IoT4Health offers extensibility points for the generation of applications. Applications can extend the framework in
the following ways: identification, collection, storage, recovery, visualization, monitoring, anomalies detection, resource notification,
and dynamic reconfiguration. Based on other outcomes involving observation of the resulting applications, it was noted that its
design contributed toward more proactive patient monitoring. Through these experimental systems, anomalies were detected in
real time, with agents sending notifications instantly to the health providers.
Conclusions: We conclude that the cost-benefit of the construction of a more generic and complex system instead of a
custom-made software system demonstrated the worth of the approach, making it possible to generate applications in this domain
in a more timely fashion.

(JMIR Med Inform 2017;5(1):e9) doi: 10.2196/medinform.6693

KEYWORDS
eHealth systems; remote patient monitoring; biometric sensors

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 1

(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

and systems. In this context, things could be identified solely

Introduction by using radio-frequency identification (RFID) [4] tags in order
Innovative Technological-Based Solutions to be connected to the Internet and publish their information.
Things are physical objects such as refrigerators, cars, walking
Technological solutions can be applied to deal better with sticks, dog collars, and whatever object comes to mind.
current operational problems involving the delivery of health
care. The application of computational tools to hospital activities Thus, using sensors, actuators, and RFID-like technology,
has the capacity to transform the present operational objects in the environment could be viewed, identified, and
environment through activities such as improvement of work controlled more autonomously. In this case, things themselves
processes. Examples of improvements brought by the use of could specify when they needed to be replaced, fixed, or report
innovative technological solutions are as follows: if they could provide data [3].

1. Change in the way the physician-patient–relationship occurs, IoT Technologies: RFID, Microcontrollers, and Sensors
because of Remote Patient Monitoring (RPM) possibilities [1] To develop the IoT patient-monitoring application described in
2. Ease of information access and sharing among the medical this paper, 3 main IoT technologies have been used: RFID,
team and the patients’ relatives [2] micro-controllers, and sensors.

3. More mobility for patients, whose health status can be RFID is an automatic identification method that utilizes radio
monitored from home or work, without being restricted to signals, recovering and storing data remotely through devices
hospital facilities called RFID tags. These devices are used for identification,
sensing, and communication [5].
4. Possibility of collaborative work between the local team and
external professionals; it allows a second opinion about patients’ Arduino [6] microcontrollers, which are open source platforms
diagnoses and treatments, as patient information is already in for electronic prototyping, are also used: Uno R3 [7] and Yún
a distributed database [8] models (Figure 1). Microcontrollers can be programmed to
process inputs and outputs of connected external components
5. Possibility of automatic processes such as vital patient data (Figure 1). One can use embedded computing to allow the
collection by using sensors construction of systems that interact with the environment using
6. Remote and real-time monitoring of patient health conditions hardware and software [9].

7. Alerts to health care professionals in emergency situations A variety of sensors can be used to collect data for IoT
applications such as temperature, humidity, light level, oxygen
8. Decrease in elapsed time for detection of anomalies in the level, and sensor presence, among others.
vital signs of monitored patients, by using software agents; in
this context, software agents consist of computational entities In eHealth, it is common for some devices to contain a number
that perform activities in response to emergency situations of sensors linked together, such as in the HealthPatch MD [10]
Vital Connect health-monitoring sensor (Figure 2). The sensor
Investments in RPM technology can provide better support for is a small adhesive patch with a module that measures heart
patients from their health care team and perhaps make resources rate, breathing frequency, body temperature, posture, detection
available for other health-related activities. of falls, and also has Internet connectivity. Another example is
Theoretical Background the eHealth Sensor Platform Complete Kit [11]. It contains an
eHealth Sensor Shield compatible with Arduino and Raspberry
Internet of Things (IoT) Pi [12] microcontrollers (Figure 2), plus 10 sensors to collect
IoT is a field within Computer Science that has grown quickly biometric data (Figure 2): pulse, oxygen levels in blood, airflow
in recent years. Kevin Ashton introduced the term “Internet of (breathing), body temperature, electrocardiogram (ECG),
Things” in 1999 [3]. One can define IoT as a global network of glucometer, galvanic skin response, blood pressure, patient
smart devices that can sense and interact with their environment position (accelerometer), and muscle or electromyography
for communication with users and other things (smart devices) sensor (EMG).

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 2

(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

Figure 1. Arduino micro-controllers, Uno R3 (on the left) and Yún (on the right) models.

Figure 2. HealthPatch MD (on the left), e-Health Sensors Shield and e-Health Sensor Platform Complete Kit (on the right).

Frameworks contain fixed and flexible points known as frozen

Software Agents spots and hot spots, respectively. Hot spots are extension points
A software agent [13] is an element of a computational system that allow developers to create a new application from the
that is situated in an environment where it can perform framework instantiation process. In this case, developers should
autonomous actions in order to reach its assigned goals. An create specific application code for each hot spot, through the
agent is both autonomous and capable of learning from its implementation of abstract classes and methods defined in the
experience. Autonomy has been acknowledged as a key framework. Frozen spots consist of the framework’s kernel,
characteristic of an agent in satisfying its goals [14]. In this corresponding to its fixed parts, previously implemented and
context, autonomy means operating without the intervention of hard to change. A frozen spot calls one or more of the
humans or other systems, although the set of possible actions application’s hot spots and is present in each framework’s
should be previously defined. instance [16].
Although agents control their behavior and internal states, they Creating a new instance of a framework consists of 3 main steps:
do not have full control of the environment in which they (1) Domain analysis, (2) Design, and (3) Instantiation. The
operate. Agents contain a set of actions that can carry out tasks, domain analysis step includes requirements elicitation including
the execution of which can result in changes in the definitions of hot and frozen spots. The design step is
environments. For this reason, one can consider that an agent responsible for specifying the hot and frozen spots through a
can have partial control and influence over its environment modeling language such as UML [17] diagrams. Design patterns
depending on the action performed [15]. [18] are also used in this phase. The instantiation phase
In general, the use of software agents is justified by the fact that corresponds to the application generation phase through hot
asynchronous software systems require autonomous operation, spot implementation [16].
a general argument that can be applied to our solution. Related Work
General Concepts About Software Frameworks Our proposal takes a similar approach to that in [19]. This paper
Frameworks are tools used to generate applications related to shows the implementation of a distributed information
a specific domain; that is, to cope with a family of related infrastructure that uses the intelligent agent paradigm for: (1)
problems [16]. The choice of using existing frameworks or automatically notifying the patient’s medical team regarding
developing new application generators is based on whether the the abnormalities in his or her health status; (2) offering medical
framework can offer design and code reuse. Thus, frameworks advice from a distance; and (3) enabling continuous monitoring
can usually increase software development productivity and of a patient’s health status. In addition, the authors have
shorter time-to-market, compared with traditional approaches. promoted the adoption of ubiquitous computing systems [20]
and apps that allow immediate analysis of a patient’s
physiological data such as a personalized feedback of their
http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 3
(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

condition in real time, by using an alarm-and-remember 1. Five measurement points were identified in the
mechanism. In this solution, patients can be evaluated, Agents4Health’s workflow related to the tasks performed by
diagnosed, and cared for through a mode that is both remote agents and were labeled as Timestamps (T1 through T5) as
and ubiquitous. In the case of rapid deterioration of a patient’s follows:
condition, the system automatically notifies the medical team
T1. The Agents4Health application retrieves the patient data
through voice calls or SMS messages, providing a first-level
from the cloud and the monitoring agent analyses them,
medical response. This proposal differs from ours, in that the
searching for anomalies. If no anomaly is detected, the system
resulting application is closed, as opposed to our broader eHealth
remains in a loop collecting more data until an anomaly is found.
application generator.
Once an anomaly is detected the application continues to T2
The approach in [21] focuses on design and development of a
T2. This second step is reached when the monitoring agent
distributed information system based on mobile agents to allow
detects an anomaly and then calls the notification agent.
automatic and real-time fetal monitoring. Devices such as a
PDA, mobile phone, laptop, and personal computer are used to T3. The notification agent initiates the routine to notify the
capture and display the monitored data. health care providers;
In [22], mobile health apps are proposed as solutions for (1) T4. The notification agent sends information about the detected
overcoming personalized health service barriers; (2) providing anomaly to the patient’s health care providers;
opportune access to critical information on a patient’s health
T5. The health care providers receive the notification message
status; (3) avoiding duplication of exams, delays and errors in
on their mobile phones.
patient treatment.
2. Agents4Health is executed and the timestamps are measured
Methods and registered.

Main Research Goals 3. Four delays defined as follows are captured for the different
agent’s execution tasks:
Our main research goal is to demonstrate that the formulation
of a software framework to generate IoT applications in the Detection anomaly interval (DAI)=T2−T1. The anomaly's
eHealth domain does effectively support RPM. The aim is to detection delay in the monitoring routine.
analyze the tradeoffs involved in the challenge of building a Notification start interval (NSI)=T3−T2. The delay between the
flexible and powerful tool to help deal with the constraints found anomaly detection and the initiation of the notification routine.
in a medical care environment. This initial version is totally
experimental; it has not been tested in real medical care Notification period (NP)=T4−T3. Duration of the notification
environments. Regarding the long-term goals of the research, routine by agents.
the aim is to apply this software framework in a real medical Notification routine interval (NRI)=T5−T4. Time elapsed
care environment to assess its effective use as well as adequacy between the sending of the notification and its receipt by the
in terms of regulatory approval. health provider.
Methodology These delays were calculated to serve as a concrete measure of
We decided to build an IoT framework to allow the how quickly and proactively the solution can respond to the
characterization of the RPM domain by using framework design environment, as well as to support the assertion that this system
techniques that encompass software agents. Framework performs anomaly detection in real time.
methodology was chosen to assess its suitability for the RPM
To confirm the fulfillment of the main research goal, the
domain and its potential to generate customized applications
experiment described above was conducted and the relevant
that achieve the stated goals of more closely connecting patients
results have been tabulated in the Discussion section.
to their health care team. As a proof of the concept, 3
applications were developed with different implementations for
each hot spot of the framework.
Results
An application named, Agents4Health was selected to illustrate IoT4Health Framework
the instantiation process and the IoT4Health framework
Domain Analysis
operation. Furthermore, IoT devices were built from scratch to
collect patient data for the Agents4Health application by using In this step, problems that health professionals currently deal
hardware prototypes comprised of biometric sensors and with in their patient monitoring routines are considered. As
Internet-enabled microcontrollers to send the sensed data to the mentioned earlier in this paper, the decision was made to build
cloud automatically. a software framework instead of one or more apps. The choice
to use framework design techniques was motivated by the fact
To measure the ability of the tool to respond proactively to that the construction of a more generic and complex system
adverse conditions such as anomalies in patients’ vital signs, would provide a cost-benefit, in that frameworks can usually
and its capacity to notify health providers in real time, the increase software development productivity and shorter
following step-by-step experiment was conducted: time-to-market.

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 4

(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

IoT4Health Design IoT4Health Architecture

Regarding the IoT4Health’s design, besides the hot and frozen The IoT4Health’s architecture is structured in 3 layers, each
spots that were modeled as UML diagrams, the following with well-defined functionality (Figure 3).
architecture was defined.
Figure 3. The IoT4Health’s architecture with its three layers (L1-L3). Data Management Layer (L3) and Distribution Data Layer (L1) interact through
the Data Communication Layer (L2). The IoT4Health’s frozen spots can be extended by hot spots.

Figure 4. The Agents4Health’s architecture with its three layers (L1-L3). The Agents4Health application (L3) interacts with Parse (L1) through REST
API (L2).

examples of how our framework can be extended by specific

L1: Data Distribution Layer application code implemented in IoT4Health’s instances.
This layer works as a remote database of a patient’s vital signs.
Identification Module (M1)
L2: Data Communication Layer It should be implemented by IoT4Health’s instances to support
Through this layer, the L3 can communicate with the L1 layer’s the patient identification process. The IoT4Health’s architecture
remote database. offers the possibility of customizing this process, allowing
developers to use different strategies, including: (1) The use of
L3: Data Management Layer
a unique identification code, such as the patient’s ID Card; (2)
The L3 Layer is responsible for the entire information utilization of RFID tags that can be inserted into objects like
management of the instantiated applications. It is composed of bracelets, cards, or other elements with radio-frequency
10 modules with well-defined responsibilities. These modules identification capability; and (3) Biometry.
are shown next, along with a description of their purpose and

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 5

(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

Collection Module (M2) email message, voice call, or by Bluetooth. This module also
It provides the collection of both patient and his or her requires the configuration of some parameters, as follows:
environment data. The collection process can be realized 1. Health provider responsible for an anomaly: a health
manually or automatically. professional should be selected to deal with each anomaly
Storage Module (M3) described.
Its implementation lets the application store the vital collected 2. Notification details: The type of message for each health care
data. Examples of storage strategies that can be developed as provider indicated previously should be specified (ie, SMS,
an extension of IoT4Health’s architecture are (1) Local storage, email, voice call, or Bluetooth), along with the details such as
(2) Cached storage, and (3) Remote storage. email address or phone number.
Visualization Module (M4) This module’s result is the communication process between
It was designed to provide users with ways to visualize storage agents. Agents that monitor patient data send a message to
data. Developers can implement some visualization strategies, agents that send notifications when they detect an abnormality
utilizing the Web or a mobile application. in the patient’s condition, based on the predefined anomaly
settings already mentioned.
Recovery Module (M5)
Resource Negotiation Module (M9)
It is responsible for recovering patient data stored on the
cloud-based platform. It utilizes the concept of cognitive agents that, in this context,
would be responsible for the use of argumentation techniques
Monitoring Module (M6) [23] to achieve resource sharing in a collaborative way, by
It was designed to continuously monitor the sensed data through making its management more effective. An application could
software agents. Agents evaluate if the sensed data are within implement cognitive agents, responsible for adopting negotiation
normal ranges, thus monitoring them to find anomalous values strategies, to obtain hospital resources for a particular patient.
(AV). These normal ranges are defined for each patient,
Dynamic Reconfiguration Module (M10)
accounting for age, gender, other individual patient conditions,
and each sensor in use. The system has a mandatory Its goal is to provide applications with context-sensitive
configuration step for each patient that can easily be completed capability so that these systems could be capable of responding
by an administrator filling out a form through a system interface. to changes in the environment. A change of a patient’s room
In this step, the following parameters are defined: could affect the defined parameters for monitoring, anomaly
detection, and notification modules, becoming inappropriate in
1. Desired value range (DVR): They are the normal values the new context. In this case, the applications’ values must be
collected from sensors; that is, values within an acceptable limit. reconfigured. This reconfiguration can be carried out manually
They can correspond to an interval such as 36.0-36.6 for body by an administrator user or autonomously by cognitive agents.
temperature, for example. They should be defined for each
sensor in use. Frozen Spots and Hot Spots
2. AVs: They are values outside the DVR, which are associated The IoT4Health contains 11 hot spots, offering developers the
with anomalies. Regarding the DVRs from the previous opportunity to create customized applications. Each one of these
example, one could have 37.8 as an example for a temperature modules has extension points that broaden our framework’s
AV. architecture, as shown above in Figure 3.

3. Label of Anomaly: They are associated with the AVs and The Application Agents4Health as an
must also be defined for each sensor. Regarding data like
temperature, it can be associated with the following anomalies:
Illustrative Instantiation of IoT4Health
(1) Hyperthermia, for example, can be the label specified for The Agents4Health application [1] is an example of the
anomalies associated with AVs higher than 36.6; (2) IoT4Health’s instantiation process, which was developed to
Hypothermia, to AVs lower than 36.0. illustrate the generative power of our framework Figure 4. It
As one can observe, each such anomaly will receive a consists of a multi-agent system that autonomously conducts
meaningful label regarding the health care context, so that it monitoring and notifying tasks. To access the patient data sensed
makes sense to a domain specialist. The goal is to enable a by real biometric sensors and remotely stored through Arduino,
health care provider to identify quickly what problem is the Agents4Health communicates with the cloud via REST
occurring when the system has detected an anomaly. application programming interface (API).

Anomalies Detection Module (M7) The Arduino integrated development environment (IDE) was
used to implement the M2 and M3 modules of the
It is supported by the use of reactive agents. This entity triggers
Agents4Health in the C++ language. The other modules were
alerts to health providers when case anomalies are detected.
created with the Java language. The software agents were
Notification Module (M8) programmed with the version 4.3.0 of the JADE tool [24]. JADE
It offers the possibility of using different strategies to send alerts is a free software distributed by Telecom Italia (the copyright
to the medical team, such as by short message service (SMS), holder), in open source under the terms and conditions of the

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 6

(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

second version of the Lesser General Public License (LGPL) However, because the Parse hosted service will be retired in
license. It is a framework to develop agent systems in Java. It early 2017, we are moving our database to another platform
simplifies multi-agent systems’ implementation through called MongoDB [27].
Foundation for Intelligent, Physical Agents (FIPA)-compliant
middleware [25]. The JADE API offers 2 types of behavior
L2: Data Communication Layer
classes that can be extended by agents: Primitive and Composite. The Agents4Health communicates with Parse (L1) through the
The Agents4Health agents’ behavior was implemented using REST API [28]. The application sends and retrieves data to and
the Primitive behavior class. Each application agent is an from the cloud through HTTP requests.
extension of the Agent Class and has a corresponding behavior
L3: Data Management Layer
to Behavior’s extension class. The behavior of each of the
system’s agents is defined by its setup method, where behavior The data management layer comprises the IoT application, with
was configured through the addBehavior method. its 8 modules (M1-M8) as follows:

The Agents4Health application’s reactive agents present 2 types M1

of behavior: TickerBehavior and OneShotBehavior. To identify a patient in the application, an RFID strategy has
TickerBehavior type behavior is executed cyclically. That is, been chosen. We have used an RFID system that includes a tag
agents in our scenario that must carry out continuous monitoring and a reader. This process is performed through an RFID
activities implement this behavior, as is the case of interface, where each patient receives a bracelet containing an
MonitoringSensorTemperatureDataAgent class. Other agents RFID tag that will be used as a unique ID code in the system.
are responsible for executing tasks that are not realized in a
M2
predefined interval of time, occurring only on demand in
response to a specific event. This is the case of In the Agents4Health application, both pulse and body
NotificationBySMSAgent, which sends messages to the medical temperatures are collected. This module may be extended to
team. collect other patient data such as electroencephalography (EEG),
EMG, as well as environment data such as light, noise levels,
The Agents4Health Instance and data about the device such as battery status. This process,
which is also called sensing, is performed automatically in
L1: Data Distribution Layer
Agents4Health. Arduino is used, together with sensors for
To provide the data distribution service to the application, a heartbeat and temperature that form an IoT device capable of
remote data storage service called Parse [26] was utilized. collecting patient data without human intervention (Figure 5).
Figure 5. Our IoT device for patient monitoring that contains an Arduino microcontroller and biometric sensors (on the left), Pulse and Oxygen in
Blood Sensor (SPO2) sensors and Body Temperature sensor (on the right).

M5
M3
A Web application is provided to support the visualization of
Once collected, the application transfers the patient’s data over
the patient data. In the current implementation, there is a line
the Internet to the Parse.
chart for each one of the sensors used and they are updated in
M4 real time (Figure 6).
The remote storage allows any authorized user to access the
data by means of a user-friendly interface [1], through any
device (computer, mobile phone, or tablet).

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 7

(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

Figure 6. Example of the visualization module (on the left). Example of an action taken by the NotificationBySMSAgent agent (on the right).

M7
M6
During this phase, a specific label for each anomaly is defined
Following the IoT4Health’s protocol, in this step, an
for each of the sensors: heartrate sensor (Table 1) and
administrative user defines the DVR and the AV for each sensor.
temperature sensor (Table 2).

Table 1. Configuring an example for the anomaly detection module, considering cardiac heartbeat.

AVa for cardiac heartbeat Kinds of associated types of anomalies

Heartbeat values <60 Bradyarrhythmias such as sinus bradycardia or atrioventricular block

Heartbeat values >110 Tachyarrhythmias such as atrial fibrillation, supraventricular tachycardia, and ventricular tachycardia

a
AV: anomalous value.

Table 2. Configuring an example for the anomaly detection module considering temperature.

AVa for temperature Kinds of associated types of anomalies

Temperature values <36 Hypothermia

Temperature values >36.6 Hyperthermia

a
AV: anomalous value.

In Agents4Health, the criteria used by the reactive agents to some interference such as a strand of hair between the sensor
detect anomalies are defined by the domain specialists and coded and the skin, the signal quality will indicate this situation. In
in the XML language. They will form the agents’ knowledge that case, the application can be configured to ignore the sensed
bases (Figure 7). data until the signal quality provides a reliable value. The latter
is useful to make the AVs flexible, taking into consideration
To prevent the system from detecting false abnormalities and
the context of the patient being monitored. To avoid mistakenly
triggering false alarms caused by simple patient movements or
detecting a heartbeat anomaly, for example, when a patient is
exercising, 2 strategies are being developed: (1) filtering the
engaged in physical activity, we can use sensors such as an
sensed data by using information provided by its own sensors
accelerometer to collect context information.
related to the signal quality; (2) adding the environment’s
sensors to collect information about the context of the M8: For Agents4Health, the choice was to send SMSs as a
measurement. The former is performed when the sensors in use notification strategy, using the Twilio [30] library. Twilio is a
provide information about signal quality. Sensors such as the platform using API communication that offers Web-service
Mindwave Mobile Headset (NeuroSky) [29] are used to collect APIs, allowing users to construct their own SMS communication
EEG data to provide this type of information. In this particular applications.
case, if the sensor is not in contact with the skin or if there is

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 8

(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

Figure 7. XML document that constitutes the agents’ knowledge base.

In a main application experiment, generating a complete instance

Discussion of the framework validated IoT4Health. This experiment, called
Conclusions and Future Work Agents4Health, made it possible to observe that its design
contributed toward making the patient’s environment more
The main objective of this paper was to report on the proactive. Through this experimental system, it has also been
construction of the IoT4Health framework based on a possible to detect anomalies in real time and to send alerts
requirements analysis of the RPM domain. This was instantly and autonomously to health providers. Thereby,
accomplished by using framework design techniques and the professionals responsible for taking action in the case of
introduction of software agents in the framework design to allow abnormalities in a patient’s condition can immediately react to
autonomic behavior [31]. To deal with the aforementioned goal, these events.
experiments for the IoT4Health framework were created and
tried. As mentioned in the Methods section, Table 3 offers the results
of the experiment conducted to offer a more concrete way to
indicate the performance of the simulated environment:

Table 3. Examples of timestamps for agents’ behavior and task delays.

Timestamp T1 Timestamp T2 Timestamp T3 Timestamp T4 Timestamp T5 DAIa NSIb NPc
(s) (s) (s)
2016-11-04-173424 2016-11-04-173427 2016-11-04-173427 2016-11-04-173428 2016-11-04-1734 3 0 1
2016-11-04-173430 2016-11-04-173434 2016-11-04-173434 2016-11-04-173436 2016-11-04-1735 4 0 2
2016-11-04-173442 2016-11-04-173445 2016-11-04-173445 2016-11-04-173448 2016-11-04-1735 3 0 3
2016-11-04-173450 2016-11-04-173454 2016-11-04-173454 2016-11-04-173455 2016-11-04-1735 4 0 1
2016-11-04-173458 2016-11-04-173501 2016-11-04-173501 2016-11-04-173502 2016-11-04-1735 3 0 1
2016-11-04-173504 2016-11-04-173507 2016-11-04-173507 2016-11-04-173509 2016-11-04-1735 3 0 2

a
DAI: detection anomaly interval.
b
NSI: notification start interval.
c
NP: notification period.

On average, the DAI for the Agents4Health experiment results precision, there is only an approximate measurement for NRI,
is 3.5 s. The NSI presented zero delays for all results in this which was less than 1 min on average.
experiment. The NP averaged 1.75 s. And, finally, as mobile
We have been involved in a number of practical developments
phones do not provide the SMS reception time with millisecond
based on our framework. One consists of the use of Bluetooth

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 9

(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

for communication with the medical team in the absence of As future work, we are planning a rigorous formal
Internet access. The use of machine learning is also examined characterization of the patient monitoring domain as well as the
in the patient monitoring domain. This approach is performed formal characterization of the family of applications reachable
by creating melanoma and mammography classifications as a through the framework flexible points. The application of
black box accessible to the system agents. IoT4Health is also cognitive agents as elements of the software framework is also
being used as the basis of a complex patient monitoring system being considered.
under development in our laboratory [32], with our participation
and which has been named portable care.

Acknowledgments
This work was supported by grants from National Counsel of Technological and Scientific Development (CNPq).

Conflicts of Interest
None declared.

References
1. Fernandes CO, Lucena CJP. An Internet of Things Application with an Accessible Interface for Remote Monitoring Patients.
In: Design, User Experience, and Usability: Interactive Experience Design. Springer International Publishing Switzerland:
Springer International Publishing; 2015 Jul 06 Presented at: HCI International 2015; August 2-7, 2015; Los Angeles, CA,
USA p. 651-661 URL:http://doi.org/10.1007/978-3-319-20889-3_60 [doi: 10.1007/978-3-319-20889-3_60]
2. Fernandes CO. Enabling a Smart and Distributed Communication Infrastructure in Healthcare. In: Lucena CJP, Lucena
CAP, Azevedo BA, editors. Innovation in Medicine and Healthcare 2015. Springer International Publishing Switzerland:
Springer International Publishing; Aug 12, 2015:435-446.
3. RFIDjournal. 2009. That 'Internet of Things' Thing - In the real world, things matter more than ideas URL:http://www.
rfidjournal.com/articles/view?4986 [accessed 2016-08-22] [WebCite Cache ID 6jxDkcPuS]
4. Finkenzeller D, Cox MK. RFID Handbook: Fundamentals and Applications in Contactless Smart Cards, Radio Frequency
Identification and Near-Field Communication. Third edition. Hoboken, US: Wiley; Jun 14, 2010.
5. Atzori L, Iera A, Morabito G. The Internet of Things: a survey. Computer Networks 2010 Oct;54(15):2787-2805. [doi:
10.1016/j.comnet.2010.05.010]
6. Arduino. URL:https://www.arduino.cc/ [accessed 2016-08-21] [WebCite Cache ID 6jwDCGjRi]
7. Arduino. Arduino UNO & Genuino UNO URL:https://www.arduino.cc/en/Main/ArduinoBoardUno [accessed 2016-10-22]
[WebCite Cache ID 6lSEHmYSS]
8. Arduino. Arduino Yún Shield (USA Only) & Genuino Yún Shield (Outside USA) URL:https://www.arduino.cc/en/Main/
ArduinoYunShield [accessed 2016-10-22] [WebCite Cache ID 6lSEaH2wB]
9. Doukas C. Building Internet of Things with the Arduino (Volume 1). USA: CreateSpace Independent Publishing Platform;
2012.
10. Engadget. 2015. HealthPatch MD alerts your doctor about heart problems in real time URL:https://www.engadget.com/
2015/01/06/healthpatch-md-vital-connect/ [accessed 2016-08-22] [WebCite Cache ID 6jxH7bcek]
11. Cooking-hacks. e-Health Sensor Platform V2.0 for Arduino and Raspberry Pi [Biometric / Medical Applications] URL:https:/
/www.cooking-hacks.com/documentation/tutorials/ehealth-biometric-sensor-platform-arduino-raspberry-pi-medical [accessed
2016-08-22] [WebCite Cache ID 6jxCIvFSO]
12. Raspberrypi. URL:https://www.raspberrypi.org/ [accessed 2016-08-22] [WebCite Cache ID 6jxI7NEyb]
13. Wooldridge M. Intelligent Agents. In: Multi-agent Systems. Cambridge, MA: The MIT Press; 1999.
14. Wooldridge MJ. An introduction to multiagent systems. Chichester, UK: John Wiley & Sons; 2009.
15. Russell SJ, Norvig P. Artificial intelligence: a modern approach. Upper Saddle River, NJ: Prentice Hall; 2010.
16. Markiewicz M, de Lucena CJ. Object oriented framework development. Crossroads 2001 Jul 01;7(4):3-9. [doi:
10.1145/372765.372771]
17. UML. What is UML URL:http://www.uml.org/ [accessed 2016-08-22] [WebCite Cache ID 6jxJ32ugU]
18. Gamma E, Helm R, Johnson R, Vlissides J. Design patterns: elements of reusable object-oriented software. Boston, MA:
Addison-Wesley; 1995.
19. Su CJ, Wu C. JADE implemented mobile multi-agent based, distributed information platform for pervasive health care
monitoring. Appl Soft Comput 2011 Jan;11(1):315-325. [doi: 10.1016/j.asoc.2009.11.022]
20. Weiser M. Some computer science issues in ubiquitous computing. Commun ACM 1993;36(7):75-84. [doi:
10.1145/159544.159617]
21. Su CJ, Chu TW. A mobile multi-agent information system for ubiquitous fetal monitoring. Int J Environ Res Public Health
2014 Jan 02;11(1):600-625 [FREE Full text] [doi: 10.3390/ijerph110100600]
22. Mohammadzadeh N, Safdari R. Patient monitoring in mobile health: opportunities and challenges. Med Arch 2014;68(1):57-60
[FREE Full text] [Medline: 24783916]
http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 10
(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

23. Laera L, Blacoe I, Tamma V, Payne T, Euzenat J, Bench-Capon T. Argumentation over ontology correspondences in MAS.
2007 May 14 Presented at: Aamas Conference 2007; May 14 - 18, 2007; Honolulu, Hawaii p. 228. [doi:
10.1145/1329125.1329400]
24. Jade.tilab. JAVA Agent DEvelopment Framework: an open source platform for peer-to-peer agent based applications
URL:http://jade.tilab.com/ [accessed 2016-08-21] [WebCite Cache ID 6jwCwVQvX]
25. FIPA. FOUNDATION FOR INTELLIGENT PHYSICAL AGENTS URL:http://www.fipa.org/docs/wps/f-wp-00018/
f-wp-00018.html [accessed 2016-08-22] [WebCite Cache ID 6jxCpiSP5]
26. Parse. Open Source Hub URL:http://parse.com/ [accessed 2016-08-21] [WebCite Cache ID 6jwBcbFYW]
27. Mongodb. MongoDB Atlas: Hosted MongoDB-as-a-Service URL:https://www.mongodb.com/ [accessed 2016-11-04]
[WebCite Cache ID 6ll1mFYUV]
28. Parseplatform.github. Parse URL:http://parseplatform.github.io/docs/rest/guide/ [accessed 2016-08-22] [WebCite Cache
ID 6jxLgHHwW]
29. Neurosky. URL:http://neurosky.com/biosensors/eeg-sensor/biosensors/ [accessed 2016-11-04] [WebCite Cache ID
6lm37mphP]
30. Twilio. URL:https://www.twilio.com/ [accessed 2016-08-21] [WebCite Cache ID 6jwCh91Vd]
31. Nunes IO, Kulesza U, Nunes C, Cirilo E, Lucena C. Extending web-based applications to incorporate autonomous behavior.
New York, NY, USA: ACM; 2008 Presented at: 14th Brazilian Symposium on Multimedia and the Web (WebMedia '08);
October 26 - 29, 2008; Vila Velha, Brazil p. 115-122 URL:http://dl.acm.org/citation.cfm?doid=1666091.1666112 [doi:
10.1145/1666091.1666112]
32. LES. 2016. URL:http://www.les.inf.puc-rio.br/wiki/index.php/P%C3%A1gina_principal [accessed 2016-08-22] [WebCite
Cache ID 6jxZarAkW]

Abbreviations
API: Application Programming Interface
AV: anomalous values
DAI: detection anomaly interval
DVR: desired value range
ECG: electrocardiography
EEG: electroencephalography
EMG: electromyography
FIPA: Foundation for Intelligent Physical Agents
HTTP: Hypertext Transfer Protocol
IDE: integrated development environment
IoT: Internet of Things
JADE: Java Agent Development Framework
NP: notification period
NRI: notification routine interval
NSI: notification start interval
RFID: radio-frequency identification
SMS: short message service
UML: Unified Modeling Language
XML: eXtensible Markup Language

Edited by G Eysenbach; submitted 24.09.16; peer-reviewed by D Cowan, M Drobics, J Rawstorn, C Matava, V Gay; comments to
author 17.10.16; revised version received 16.11.16; accepted 29.01.17; published 27.03.17
Please cite as:
Fernandes CO, Lucena CJPD
A Software Framework for Remote Patient Monitoring by Using Multi-Agent Systems Support
JMIR Med Inform 2017;5(1):e9
URL: http://medinform.jmir.org/2017/1/e9/
doi: 10.2196/medinform.6693
PMID: 28347973

©Chrystinne Oliveira Fernandes, Carlos José Pereira De Lucena. Originally published in JMIR Medical Informatics
(http://medinform.jmir.org), 27.03.2017. This is an open-access article distributed under the terms of the Creative Commons

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 11

(page number not for citation purposes)
XSL• FO
RenderX
 JMIR MEDICAL INFORMATICS Fernandes & Lucena

Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction
in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete
bibliographic information, a link to the original publication on http://medinform.jmir.org/, as well as this copyright and license
information must be included.

http://medinform.jmir.org/2017/1/e9/ JMIR Med Inform 2017 | vol. 5 | iss. 1 | e9 | p. 12

(page number not for citation purposes)
XSL• FO
RenderX