A Framework for AI-enabled Proactive mHealth with Automated

Decision-making for a User’s Context

Muhammad Sulaiman, Anne Håkansson and Randi Karlsen

Department of Computer Science, UiT The Arctic University of Norway, Tromsø, Norway

Keywords: mHealth, Proactive Health, Artificial Intelligence, Machine Learning, Wearables, Just-in-Time Adaptive

Interventions, Automated Decision-making, Digital Health Intervention, Self-empowerment.

Abstract: Health promotion is to enable people to take control over their health. Digital health with mHealth empowers

users to establish proactive health, ubiquitously. The users shall have increased control over their health to

improve their life by being proactive. To develop proactive health with the principles of prediction, prevention,

and ubiquitous health, artificial intelligence with mHealth can play a pivotal role. There are various challenges

for establishing proactive mHealth. For example, the system must be adaptive and provide timely

interventions by considering the uniqueness of the user. The context of the user is also highly relevant for

proactive mHealth. The context provides parameters as input along with information to formulate the current

state of the user. Automated decision-making is significant with user-level decision-making as it enables

decisions to promote well-being by technological means without human involvement. This paper presents a

design framework of AI-enabled proactive mHealth that includes automated decision-making with predictive

analytics, Just-in-time adaptive interventions and a P5 approach to mHealth. The significance of user-level

decision-making for automated decision-making is presented. Furthermore, the paper provides a holistic view

of the user's context with profile and characteristics. The paper also discusses the need for multiple parameters

as inputs, and the identification of sources e.g., wearables, sensors, and other resources, with the challenges

in the implementation of the framework. Finally, a proof-of-concept based on the framework provides design

and implementation steps, architecture, goals, and feedback process. The framework shall provide the basis

for the further development of AI-enabled proactive mHealth.

1 INTRODUCTION

Center for disease control and prevention (CDC)

defined public health as "the science and art of

preventing disease, prolonging life, and promoting

health through the organized efforts and informed

choices of society, organizations, public and private

communities, and individuals" (Winslow, 1920).

This definition likewise emphasizes the need of

promoting health and preventing disease providing a

holistic solution for the individual. To promote health,

people must be enabled to increase control over their

health to be able to improve it, World health

organization (WHO) (WHOa, 2021).

Health promotion and wellbeing are vital as

healthcare globally is dealing with many challenges

(Haseltine, 2021). One major challenge is an aging

population, life expectancy has increased in the last

century or so for instance in Norway it is 84.2 years

for women and 80.6 years for men (NIPH, 2018).

This increase intensifies the development of multiple

chronic diseases like cardiovascular disease, stroke,

cancer, osteoarthritis, and dementia (Atella et al.,

2018). WHO estimates that about half of the disease

burden is from chronic illness (WHOb, 2021). In the

US, about 6 out of 10 adults suffer from chronic

diseases (CDC, 2021).

The healthcare sector must endure the pressure of

dealing with a public health crisis. A recent example

is the pandemic (WHOc, 2020) coronavirus (COVID-

19) that provides insights on how healthcare has to

cope with one of the most contagious diseases that hit

mankind in the past decades. COVID-19 is not the

first and certainly not the last of these viruses. During

a public health crisis, it is important to provide regular

care to people at a distance. COVID-19 certainly

fueled the need for new tools and practices for

healthcare digitalization to provide care to people

away from the hospital settings.

Digitalization in healthcare to support self-

management is not a new concept. Digital health is

playing a pivotal role to support healthcare by

Sulaiman, M., Håkansson, A. and Karlsen, R.

A Framework for AI-enabled Proactive mHealth with Automated Decision-making for a User’s Context.

DOI: 10.5220/0010843200003123

In Proceedings of the 15th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2022) - Volume 5: HEALTHINF, pages 111-124

ISBN: 978-989-758-552-4; ISSN: 2184-4305

111

transforming existing practices (Hermes, Riasanow,

Clemons, Böhm & Krcmar, 2020). It contributes to

promoting health by providing tools to empower the

user (FDA, 2021). Digital health also incorporates

mobile health (mHealth) to provide healthcare

services using mobile and wireless technology.

mHealth combines wearables to render health

services to anyone, anywhere at any time (WHOd,

2018). This ubiquitous notion is the key in mHealth

to deal with health delivery constraints like location,

time, and cost. mHealth and wearables market is

growing and is estimated to reach 149.3 billion USD

by 2028 (GRAND View Research, 2021).

mHealth facilitates the paradigm shift in self-

management by providing tools to the users, so they

can become more aware and conscious about their

health. The user-centric approach can empower the

user by providing new insights into the health

information gathered from wearables and mobile

devices. These devices that users carry at all times,

and the collected health information they generate

provide the need for user-level decision-making.

The reactive approach in health is to act when a

crisis occurs with damage control (Amir, 2019). It is

practical in many circumstances, but it diminishes

health promotion and self-management. Proactive

health, in contrast, is to act before a crisis to predict and

prevent a situation promptly (Sharma, Singh Aujla &

Bajaj, 2019). Proactive health promotes wellbeing by

empowering the user and making them aware of an

anomaly beforehand. The active participation of the

user in their health enables more health information.

Proactive health can promote wellbeing, so the

ultimate goal of proactive health is to be predictive and

preventive with personalization. The system should

account for the uniqueness of the user. This can enable

care for the user ubiquitously but can also save lives

and support the healthcare system.

To develop proactive health with the principles of

prediction, prevention, and ubiquitous health

Artificial Intelligence (AI) with mHealth can play a

pivotal role. Many different definitions of AI are

available over the decades that serve well for many

use cases. IBM defined AI as "artificial intelligence

is a field, which combines computer science and

robust datasets, to enable problem-solving. It also

encompasses sub-fields of machine learning and deep

learning, which are frequently mentioned in

conjunction with artificial intelligence. These

disciplines are comprised of "AI algorithms which

seek to create […] systems which make predictions

or classifications based on input data (IBMa, 2021).

This definition serves well for enabling AI in

proactive health.

Wearable and mobile devices which the user

carries, have many sensors to collect health data,

which can be the key for finding patterns and making

accurate predictions for early intervention. AI can

apply reasoning and negotiation to the gathered data

to automate processes and facilitate decision-making.

This paper explores the fundamentals of AI-

enabled proactive mHealth by a comprehensive study

for the framework, challenges in design and

implementation of a system with automated decision-

making. The paper also presents the implementation

goals, and a model with an architecture for developing

AI-enabled proactive mHealth based on the proposed

framework. The result is a proof-of-concept that

renders an implementation view of the system with

technical aspects of design steps, goals, and a proposed

input/output mapping with the architecture. The paper

will provide the basis for further development of AI-

enabled proactive mHealth.

2 RELATED WORK

Proactive mHealth is to predict and prevent a

situation beforehand. A system that can provide

proactiveness must have a clear goal to predict and

prevent. The level of proactiveness when it comes to

proactive mHealth is broad. A basic level of

proactiveness provides benefits of being proactive to

manage a disease, a medium level corresponds to

being predictive, and a high level of proactiveness

allows a system to be predictive and preventive. Some

studies (McConnell et al., 2018; Aguilera et al., 2020;

Korpershoek et al., 2020; Baig, 2017) presented a

basic level of proactiveness, and the benefits of being

proactive. These are targeted for a specific need to

self-manage the disease. Hence, they are not adaptive

to new situations which require more level of

proactiveness. A few studies (Aguilera et al., 2020;

Baig, 2017) focused on managing a chronic disease

proactively.

A study (Nag, Pandey & Jain, 2017) presented the

importance of proactive health and significance by

giving a health map example. The health-map has

states of the user drawn in form of a map.

A few research studies (Dijkhuis et al., 2018;

Baig, 2017; Rojas & Dey, 2019) used wearables as

data sources but with activity data only. Activity data

can be useful for promoting health but when it comes

to providing prediction activity. A study (Baig, 2017)

concludes wearables as a key in providing health for

anyone at any time. Another study (Menictas, Rabbi,

Klasnja, & Murphy, 2019) highlights the importance

of decision-making with mHealth.

HEALTHINF 2022 - 15th International Conference on Health Informatics

112

2.1 Applications of AI in mHealth

AI is one of the factors driving healthcare towards

digitalization. It represents several technologies that

enable machines to sense, comprehend, act, and learn

(Matthew & Richard, 2021). The booming increase in

generated data today (Statista, 2021) has fuelled the

need for AI to support and automate healthcare

dilemmas. Some examples of domains where AI

contributes are administrative workflows, fraud

detection, dosage error detection, diagnosis

assistance, virtual assistance, decision-support,

automated processes, drug discovery, personalized

treatment, disease screening and early detection

(Matthew & Richard | Accenture, 2021). These

application areas provide the basis for the framework

of AI-enabled proactive mHealth.

mHealth application and available wearables data

have proven another domain where AI is

contributing. Many studies have used mHealth with

AI, a review (Naseer Qureshi, Din, Jeon & Piccialli,

2020) presented mHealth applications that use

machine learning. People are becoming more aware

of their health. Statistics show that 62% of

smartphone owners search the internet for health-

related information (Smith | Pew research, 2015).

Estimation indicates that about 80% of mHealth

applications will use AI by 2025 (Ghazaryan, 2021).

However, not all websites contain the correct

information, and an AI system shall be able to filter

out the non-correct parts and provide information

about the valid and trustworthy websites.

Some use cases (Ghazaryan, 2021) of mHealth

with AI are prediction models, personalized

treatments, early detection, recommender systems,

screening, and triage, and chatbots. For the

framework some of these use-cases are included.

mHealth with AI can support users in decision-

making to promote health. mHealth is beneficial

because of the features it provides e.g.,

personalization and ubiquitousness. It is utterly

necessary to account for the uniqueness of the user,

handled by user preferences. Table 1 presents some

mHealth applications and their features.

Table 1: mHealth applications.

Purpose Features

Diabetes control (Curran,

Nichols, Xie & Harper,

2010)

The paper focuses on

adjusting insulin levels

using mHealth

Activity (Yom-Tov et al.,

2021

)

The aim of the study is to

romote walkin

Blood pressure level

(Toro-Ramos et al., 2017)

The aim of the study is BP

monitoring

The related work provides the effectiveness of

being proactive but in the context of self-managing

the disease. There is no prior research that defines

proactive mHealth with AI with the capabilities of

prediction and prevention. A system to predict and

prevent a situation promptly, targeting a user for

promoting wellbeing before becoming sick. Such a

system can consider multiple parameters of wellbeing

i.e., environment or surroundings and current state of

the user. The critical analysis determines that AI in

mHealth is rapidly growing, but with a focus on

monitoring and self-management which depreciates

proactiveness.

3 USER-LEVEL

DECISION-MAKING FOR

SELF-EMPOWERMENT

Decision-making is the process that can have an

impact on our lives (Steph | Medium, 2020).

Decision-making requires time and effort to

comprehend details, make decisions on knowledge at

hand, plus decision alternatives to choose from based

on the circumstances. Effective decision-making

requires a step-by-step approach. One strategy

proposes a seven-step model for effective decision-

making (Dartmouth, 2021). The steps are as follows:

to identify the decision, collect information, identify

alternatives, weigh the evidence, choose among

alternatives, take action and finally, review the

decision.

Decision-making in healthcare often involves

several stakeholders in this process, such as doctors,

and nurses. Most health-related decisions are said to

be in the grey area of decision-making

(Abbasgholizadeh Rahimi, Menear, Robitaille &

Légaré, 2017). The grey area represents the scenario

where there is no right or wrong approach. Health-

related decisions can be critical with life-threatening

impacts for the users. Thus, there is a need to have

more informed choices and insights on the patient to

help with effective decision-making. mHealth is

contributing to this process by engaging patients in

decision-making to support health professionals with

new insights.

Most of these mHealth solutions are decision-

support systems for clinical decision-making. This is

an approach where the system supports evidence-

based decision-making. These solutions are crucial,

and many healthcare professionals rely on them to

provide shared decisions by having a combined

A Framework for AI-enabled Proactive mHealth with Automated Decision-making for a User’s Context

113

opinion between the healthcare professional and the

mHealth solution (Abbasgholizadeh et al., 2017).

These shared decisions are indeed beneficial

although, this diminishes the fact that many decisions

are taken at the user level (Krist, Tong et al., 2017).

The health decisions or health choices can have a

direct impact on their health, i.e., prevent disease,

prolong life, and promote wellbeing. Table 2 presents

the types of decisions that can have an impact on the

health of a user and the contributors involved in this

decision-making. mHealth and wearables are being

used by users all the time. There is new health

information about the states of the user that is

dynamic and can provide health data required for

decision-making. The table has two columns, first

column provides a case of decision-making. The

following column renders detail about contributors in

making that decision.

Table 2: User-level decisions with contributors’ example.

Cases for decision-

making

Contributors involved in

decision-making

From Air quality index

(AQI = 157), the user

profile (asthma), The air

quality tomorrow is

unhealthy. You are in a

risk group, stay home or

wear a mask when you go

out.

Sensors, mobile and the

user

Low activity, the user

profile (goals), You are

not very active lately. The

weather is pleasant today,

have a walk for ten

minutes.

Wearables, sensors,

mobile devices, and the

user

The pollen count, User

profile (have pollen), The

pollen count will be

higher tomorrow. You are

in a risk group, stay home

or wear a mask when you

go out.

Meteorological data,

user profile, mobile

device

These decisions cases are significant for a person

since they are not part of the clinical decision-making

approach that requires a doctor to examine but must

be handled by the user. Let us consider the case of a

user called “A” that is wearing a sleep tracker to sleep

every day. The information from that tracker can lead

to choices that user can take to improve his/her health.

These decisions are taken by User-A at the user level

and can promote health.

4 THE CONTEXT OF THE USER

The user’s context is "the interrelated conditions in

which something exists or occurs" (Merriam-

Webster, 2021). Context is important when providing

proactive mHealth because it consists of parameters

that can have a direct impact on the health of the user.

Many different circumstances e.g., environment,

surroundings, and user-profiling contribute to the

context. To emphasize, consider an example of a

system that can predict a health issue – atrial

fibrillation. Which is a heart condition that causes an

irregular heart rate. The system must include every

attribute about the user, family history, disease

history, current and previous states from the

wearables and sensors and user-profiling with user

characteristics. A context is anything relevant, can

have an adverse impact on health.

A context-aware system must be able to collect

information about surroundings and adapt to the

environment. Forming a context is vital because it

considers the uniqueness of the users and more

information on the states of the user. To define the

context, it is required to determine the target person

by choosing a user or group.

A user’s context accounts for user characteristics

and profiling. For someone who is suffering from

asthma caused by an inflammatory reaction

(Djukanović, 2021) the context of air pollution shall

be included.

Moreover, for a user in a risk group of infectious

disease it is essential to know about an outbreak of flu

or the rise of a number of flu cases in the area. In

contrast, for a specific group of people in risk groups

like the elderly or multi-disease, the context of flu

spread can be the same. In addition, a storm warning

can have a similar impact on everyone in the area.

The context can be elaborated and defined in many

ways, but the limitation comes to what is measurable

from the sensors and other sources and what can be

predicted and prevented. The context can also include

another person in the surroundings, i.e., one infected

person can infect another person. In the same way, a

driver on the road who suffers from a heart problem or

epilepsy can be a threat to the nearby surroundings. But

this is not measurable or predictable, in the current

systems. With the usage of the Internet of Things, IoT,

and autonomous vehicles, it can be a possibility to

provide solutions for measuring and predicting failures

and epileptic seizures.

Context is dynamic with ever-changing

surroundings. A system must be adaptive to cope with

the dynamic changes in context. A person suffering

from a pollen allergy shall have as context the amount

HEALTHINF 2022 - 15th International Conference on Health Informatics

114

of pollen in the environment, which is fluctuating over

time. Everything that can have an impact on one's

health, shall be part of the person’s /user’s context.

With the above-mentioned definition of context, it

is assumed to have different attributes e.g.,

environment, surroundings, user-profile, and

characteristics. The table shows several parameters,

used for developing the context and how they are

related to a user.

Figure 1: The context of a user.

An example of a user’s context is presented with

details on what it holds. The figure, Figure 1, shows

how a user’s context can be and what parameters it

includes. The center of the figure displays a user with

a user profile. Different parameters that form the

context are presented: user’s profile, characteristics,

environment. Table 3 demonstrates each parameter in

detail.

Table 3: Parameters for a user’s context

The context

User characteristics User profiling

Environment

(anything

that is

around and

can have an

impact)

Daily

patterns

Behaviour Physical

attributes

History

State

Weather, air

pollution,

threats, and

outbreak

Physical

activity

Screen

time

To adapt,

actions

and

behaviour

BMI

Allergies

Weight

Family,

Disease

Location,

Health

information

Activity

data, SpO2,

and feve

These parameters require multiple data sources as

input to the system. In section 5, a comparison of

available resources with the parameters of the context

is established.

5 THE FRAMEWORK OF

PROACTIVE mHealth

A framework provides the supporting structure

(Cambridge, 2021) to support building software. The

specified framework provides abstraction, which

supports the development of systems over it. The

framework also defines a set of rules to follow when

developing applications.

A conceptual framework defines concepts

collected after extensive research into a topic. A

definition of the conceptual framework is a

"conceptual framework as a network, or a plane, of

interlinked concepts that together provide a

comprehensive understanding of a phenomenon or

phenomena" (Jabareen, 2021). It is essential to

understand the link between these concepts.

This paper presents, a framework of AI-enabled

proactive mHealth. The framework is derived from a

systematic literature review of the topic and existing

systems. To show proof-of-concept, each part of the

framework is defined with examples and use-cases.

Following are the components of the framework:

 Automated decision-making with predictive

analytics

 P5 approach to mHealth

 Just-in-time adaptive interventions

Figure 2: Proactive mHealth framework.

Figure 2 shows the components of the framework:

AI with automated decision-making, Just-in-time

adaptive intervention, and the P5 approach to

mHealth. Each component has variables with relation

to other components.

5.1 Automated Decision-making with

Predictive Analytics

Automated decision-making (ADM) with predictive

analytics is defined as “decisions by technological

means without human involvement" (EDPB, 2021)

recognizing patterns from extensive information to

provide decisions (data-driven). Booming increase in

the amount of digital data and ever-growing AI,

decision-making is empowered to automated

processes (Saha, 2021).

The first step is to process information by

applying algorithms and making informed decisions.

ADM that is powered with predictive analytics can

gather, process, and model health information to

render an automated decision (Araujo, Helberger,

Kruikemeier & H. de Vreese, 2020).

A Framework for AI-enabled Proactive mHealth with Automated Decision-making for a User’s Context

115

As presented in section-3, user-level decision-

making with multiple use-cases and examples. The

emphasis on user-level decision-making proves to be

the basis of providing ADM for AI-enabled proactive

mHealth. A system must consider multiple factors as

input and adapt to sudden changes in the states but

eventually, the benefit for the user is to get an

automated decision from the system, which is valid

for the user. These automated decisions are a part of

the prediction and prevention mechanism to promote

wellbeing.

Machine learning algorithms such as random

forest (RF), decision trees (DT), regression models,

artificial neural networks (ANN) can contribute to

developing the engine for ADM. Choosing the

technique depends on the type of data and decisions

to provide to the user.

It is imperative to understand the types of

decisions. These automated decisions are for

predicting and preventing health issues to promote

wellbeing. The system does not provide clinical

decision-making and hence, cannot be used as one.

Some examples of automated decisions are:

 The system predicts the air quality to be

unhealthy tomorrow, wear a mask.

 High pollen in the area tomorrow you are in

the risk group, please use medication.

 There is an outbreak in the area you are in the

risk group. Stay home stay safe.

These decisions are based on the parameters (user

profiling, context, and characteristics) of the user.

The decisions are provided as information to the user

to prevent a possible illness or health issue.

5.2 P5 Approach to mHealth

The P5 approach to mHealth is based on the P4

(Sagner et al., 2018) spectrum of medicine presented

as an extension to personalized medicine. It proffered

new insights into designing health systems to be more

vigilant in considering users’ uniqueness and

timeliness. The spectrum is described as predictive,

preventive, personalized and participatory.

There have been many additions to this P4

spectrum. A P5 approach was presented (Gorini et al.,

2018) that added a P as Psycho-cognitive to provide

more details on the user’s decision-making regarding

their health. P5 approach to mHealth is relevant

because of the capabilities of mHealth to provide

more information about the user and a platform to

observe timeliness.

Table 4 illustrates the P5 principles for designing

health systems in detail. The 5 P’s are predictive,

preventive, personalized, participatory, and psycho-

cognitive.

Table 4: P5 Approach to mHealth (Gorini et al., 2018).

5 Ps Definition

Predictive

Allowing a precise prediction

about the future state

Preventive Timely preventive interventions

Personalized The uniqueness of the user

Participatory Not as passive recipients rather

active decision-makers

Psycho-cognitive Improving the ability to make

decisions

AI-enabled proactive mHealth system must

adhere to this design principle to be more effective for

the users. These design principles are part of our

framework of AI-enabled proactive mHealth.

The order of these principles or implementation

depends on the use case at hand or the targeted

outcome of the application. A motivating factor is the

availability of resources and tools to implement that.

With AI and the availability of wearable devices,

most of the health can be gathered for a more precise

targeted outcome. The design implementation can be

at different levels as well. Personalization can be

achieved for a user by targeting user preferences, but

at a community level, it depends on the characteristics

of the society.

5.3 Just-in-Time Adaptive

Interventions

Just-in-time adaptive interventions (JITAI) are based

on health interventions which are a way of providing

health services using mobile applications. Digital

health intervention aims to deliver information that is

useful for the user using digital platforms (Soobiah,

Cooper & Kishimoto, 2021).

Interventions are categorized as nudges, boosts

and recommendations (Hertwig & D Ryall, 2020). A

nudge is defined as "A nudge is an aspect of the

choice architecture that alters people's behavior in a

predictable way without forbidding any options or

significantly changing their economic incentives"

(Osman, 2016). A boost differs as its objective is to

improve someone's ability to make their own

decisions (Hertwig & D Ryall, 2020). In this study,

the focus is not on the behavioral science aspect of

these interventions. The property relevant to the

framework is the importance of interventions and

their impact.

HEALTHINF 2022 - 15th International Conference on Health Informatics

116

The focus of the framework is on the architecture

of digital interventions that can provide timely

interventions to the user but with adaptive behavior.

The Just-in-time adaptive intervention (JITAI) is an

intervention design that provides the right type of

support at the right time, by adapting to one's varying

internal and contextual state (Nahum-Shani et al.,

2017). JITAI intervention design focuses on 3

principles that are more of a challenge when it comes

to implementing JITAI.

 When: When to intervene

 What: What information to provide

 Whom: To whom, the target user receiver of

the intervention

Figure 3: JITAI intervention points (Nahum-Shani et al.,

2017).

Figure 3 shows JITAI intervention points: when

in time, what intervention options, to whom who is

the target user and decision rules. An AI-enabled

proactive mHealth system must adhere to these

principles to provide timely interventions. That also

means it is necessary to understand the current state

of the target user before providing interventions. The

term internal and contextual state refers to the current

state of the user as well as its context.

Figure 4: Health map JITAI intervention points.

Figure 4 shows the health map, to highlight the

use of JITAI benefits for health promotion. The start

state represents the state of the user. The goal state is

the state representing goal. The intervention point is

a user-level decision required to change the state of

the user. The impact after this intervention is

displayed with an ascent in green.

6 IDENTIFICATION OF

SOURCES: WEARABLES,

SENSORS, AND AVAILABLE

RESOURCES

Identification of data collecting sources is the first

step in developing an AI-enabled proactive mHealth

system using the framework. The parameters that

provide health information about the user and its

context must be gathered as input. Most importantly,

information about the current state of the user is

crucial when providing JITAI.

Several different resources shall be included for

information collection to provide a holistic view of

the user’s preferences and current health. Table 5

provides some details about the most commonly used

sources and features they offer. The first column

identifies different sources, the second column

features list the detail of each. Finally, the listed

factors match what to collect from each source.

Table 5: Sources with features and factors.

Sources Features Factors

Wearables activity data SpO2

(oxygen

saturation), heart

rate, body temp,

Physical activity

and Screen time

(inactivity time)

User

profile

User

characteristics

Current

state

Daily patterns

Sensors Weather, Air

pollution

Context

Environment/Surroundings

Available

sources

Outbreaks,

Threats, risks in

the nearby area.

Pollen, Storms,

cyclone, and

avalanche

Context

Environment/Surroundings

In this paper, sources are categorized as

wearables, other sensors, and available resources.

Each source is examined with an example to collect

data necessary to build a holistic view of the user, but

also the challenges it possesses

6.1 Wearable Devices

Wearable devices have sensors that the user wear, are

portable, comfortable, and can collect data, which are

combined to produce information (Wu & Luo, 2021).

A Framework for AI-enabled Proactive mHealth with Automated Decision-making for a User’s Context

117

These devices also provide data even when they are

not being used (running in the background).

Another definition of wearable technology is

“seamlessly embedded portable computers … worn

on the body” (Janet, 2016). In health, wearables are

adopted as e.g., fitness trackers, biosensors, and smart

health watches (Janet, 2016).

Some key features (Bove, 2019) that wearable

provides are: Accelerometer, Altimeter,

Electrocardiogram, Location GPS, Microphone,

Oximeter, Thermometer, Pressure, and Stress.

Table 6 shows common fitness trackers (Fitbit

charge 4, Garmin Vivosmart 4, OURA Ring, Polar

M430, and Mi band 6) with their features.

Table 6: Wearables with features: a comparison.

Fitbit Charge

Garmin

Vivosmart

OURA Ring Polar

M430

Mi band 6

Activity

tracking,

GPS,

Continuous

heart rate,

breathing

rate, Stress

management

score, SpO2,

Skin

temperature,

sleep and

inactivity

Activity,

Sleep, heart

rate

measureme

nt and

inactivity

Sensors:

accelerome

ter,

pedometer

Activity,

Steps,

Inactivity

time, heart

rate, body

temperature

and sleep

Sensors:

acceleromete

r, optical

heart rate

monitor

Activity,

Sleep,

heart rate

measurem

ent and

inactivity

Sensors:

accelerom

eter,

optical

heart rate

monitor

Activity

tracking,

GPS,

Continuous

heart rate,

Stress

management

score, SpO2,

sleep and

inactivity

time

Data produced by these wearable devices can be

collected and used for AI-enabled proactive mHealth.

Unfortunately, the interoperability challenges

between vendors complicate this data collection

process. Most modern wearable sensors provide their

software development kit (SDK). An SDK is a set of

tools and programs made available by the vendor for

developers to work on. It provides several APIs for

the developers to use. Although it is a challenge when

it comes to data collection and having different

sensors connected to a single platform.

Vendors are working on solving the problem with

the non-compatibility of wearables, and there have

been many recent developments in this. For a clear

view of the compatibility problem, a comparison is

drawn between two of the vendors and the SDKs they

provide. Tables 7 present a comparison between

Google Fit and Apple HealthKit in terms of systems

and options available and data transfer options.

Table 7: System and options available (De Arriba-Pérez et

al., 2016).

Platforms SDK-

Sensors

SDK-

Warehouse

REST API

Apple

HealthKit

  

Google Fit

  

For storage, Google Fit provides warehouse and

cloud services with an SDK for developers. It also

provides REST API for third-party systems, a very

beneficial feature for to getting endpoint access.

Another essential feature is access to raw sensor data.

It is vital when real-time data from the user is needed.

An example of this is to use Sensor API for access to

raw sensor data in Google Fit.

The Apple HealthKit does not provide direct data

access to the wearable or the warehouse. The only

way is to access data by a query. A REST API service

is not provided which must be built first to store or

retrieve desired data.

Choosing the wearables for AI-enabled proactive

mHealth depends on the following factors: what we

want to measure, what is essential and what features

we can use.

6.2 Other Sensors

To collect real-time information about the context

more information from sensors is needed. A proposed

project can use the following setup to render

information.

 A microcontroller: Arduino Uno Rev3

(Arduino, 2021): It is based on ATmega328P,

has 14 digital input/output pins, a 16 MHz

ceramic resonator.

 Sensor: MQ-135 (Winsen, 2021): It is a

semiconductor sensor to measure air quality, it

is widely used as an alarm.

 Sensor: BME 280 (Bosch, 2021): It is an

environmental sensor that measures humidity,

pressure and temperature.

 Sensor: PM 2.5 (Adafruit, 2021): It is an air

quality sensor that measures air quality in real-

time.

For AI-enabled proactive mHealth to adapt to

real-time context, it is imperative to use these sensors

together to aim an overall view of surroundings.

6.3 Other Available Resources

To make accurate predictions, historical data of

context is required, i.e., air quality data from previous

HEALTHINF 2022 - 15th International Conference on Health Informatics

118

years can contribute to early detection. These

historical data can contribute to an early detection of

an outbreak. Cities or communities already provide

datasets for their environments. For example,

metrological data provide information measured with

different parameters regarding the environment,

including weather data, warnings data, temperature,

outbreaks, and cyclones. The predictions using these

resources are possible by training and modelling

systems.

This metrological data, combined with other

sensors and wearables cover the need for AI-enabled

proactive mHealth. A system must consider using

multiple parameters as input to provide automated

decision-making to the user.

7 CHALLENGES

To establish AI-enabled proactive mHealth many

challenges must be tackled.

One challenge is the complexity of being

proactive. In terms of implementation or design, it is

difficult for defining proactive. As discussed, the

system must consider multiple parameters, i.e.,

environment, surroundings, user profile, and

characteristics, but it makes the system more complex

when not every parameter is relevant for automated

decision-making. Another challenge is the

availability of these various parameters. To address

this issue a clear target must be considered. For

example, if a system is to provide timely intervention

to support a user at the right time, the system must

alert the user and show if it is safe to go outside or not

before the user leaves the house. The goal reduces

complexity to exclude data as not relevant for the

moment. For example, sleep data is not applicable for

air quality. Hence, it is to use the right, proper data at

the right time and for the right purpose. In the case of

air quality, it is data from the city's sensors that have

collected information about the current air quality, the

user's existing health information like asthma and

user preferences like personal acceptance of the air

quality in the surroundings outside the house.

Another challenge is to understand the target of

AI-enabled proactive mHealth, that is if the target is

for a user or a specific population. The

implementation for a user must adhere to the

uniqueness of the user, choices, and patterns essential

to have an efficient system implementation.

One system challenge is the importance of human

behavior as a part when implementing a system. The

system must cope with user characteristics and adapt

to a healthy lifestyle. The system must adapt to

change in lifestyle and the actions that can impact

health. This requires the system to constantly get

feedback, as a feedback loop from the user to model

the behavior and capture user preferences. The

feedback shall not require the user to provide

information manually. Rather, the systems shall learn

from how the user act, in a situation. For example, did

the user go out, although the system has alerted the

user about the bad air quality.

The timeliness of JITAI is an implementation

challenge, as well. A well-directed timely

intervention can save lives. Thus, a system must

handle precisions, i.e., when is the time for an

intervention with what type of information to the user

and finally, considering the user preferences when

providing this intervention.

Wearables and sensors provide great details into

real-time information, such as about the current state

of the user. Wearables and sensors indeed support the

timeliness principle of these interventions, since real-

time data is necessary to provide just-in-time

information to the user.

Using this real-time information is beneficial but

also a challenge when it comes to implementation.

Decision rules that include this information

simultaneously with the historical data are tricky to

implement because any sudden change must be

accepted by the model. Different viewpoints i.e.,

conditions that can suddenly change are a must for

designing decision rules with real-time data. Raw

data processing is another complicated process

because gathered data from sensors are to be filtered

before inclusion.

Sudden change in the context and user

characteristics is difficult to include in the system. It

produces a challenge of using real-time raw data with

historical data. So, the more information sources and

resources, the better the adaption mechanisms

because raw data must be filtered and processed

before inclusion. A system must update itself with the

incorporation of new information.

An added challenge is the evaluation of the system

and if it works, the accuracy of interventions and their

impact on health. It is beneficial to get feedback from

the user or a design principle where the system gets a

notification from the user, i.e., if they endure a

decision or not is beneficial.

Choosing the precise machine learning algorithms

depends on the available datasets. Many machine

learning techniques i.e., Support-vector machines,

SVM, decision trees, DT, and artificial neural

networks, ANN, are proven to be accurate. A system

must adapt to variations and the accuracy of the

predictions and preventions must be evaluated by a

A Framework for AI-enabled Proactive mHealth with Automated Decision-making for a User’s Context

119

defined evaluation mechanism to determine the best

model for the dataset.

In decision-making, it is very crucial to interpret

and provide transparency. But there is a trade-off

between automation and transparency. This

opaqueness and black-box system directly impact

trustworthiness. eXplainable AI (XAI) (IBMb, 2021)

can be applied to making AI techniques more

explainable. That is to have a system whose decisions

are understandable by humans.

Implementation of the system requires starting

from prediction, then moving towards prevention. So,

to promote the health of a user the system is built like

a stack.

 a prediction that someone can become sick.

 an action, i.e., activity or prevention for the

user (a personalized activity)

 Feedback from the user, which becomes a new

input for the system.

User participation is a design principle called

participatory from the P5 principles of system

implementation, which is very complicated when

designing the system. The system must handle

usability, i.e., to make the system easy to use and

acceptable for everyone. The usability of the system

is about engaging the user to contribute to the system.

8 PROPOSED DESIGN AND

IMPLEMENTATION

The implementation goal is to adopt the framework

of AI-enabled proactive mHealth explained in Figure

2. The AI-enabled proactive mHealth system must

incorporate automated decision-making with

predictive analytics, the P5 approach for design, as

well as JITAI for providing interventions to the user.

Implementation of a system using the AI-

enabled proactive mHealth framework includes steps,

architecture, and processes.

8.1 Building Systems using the

Framework

The framework presented above, supports building

AI-enabled proactive mHealth systems.

8.1.1 Implementation Steps

The implementation phase starts with identifying

sources, then collecting relevant data, preparing data

for feature extraction, choosing the algorithms,

defining decision rules, training the model, and

making a prediction. Also matching the outcome with

an automated decision, and finally allowing feedback

for evaluation must be taken into account.

 Identification of parameters: This step is

necessary when considering what to predict

and what attributes to include. For AI-enabled

proactive mHealth, the system must recognize

parameters of the context, user characteristics,

and user profiling.

 Classifying sources: Another step is to look for

sources and resources for the parameters. For

the AI-enabled proactive mHealth system,

these are wearables, other sensors and

available resources.

 Collecting data: Data collection is a challenge,

considering different heterogeneous sources.

In addition, information is required from the

user to provide intervention, such as, JITAI.

So, this step is significant to collect every

possible data from the available sources and

then provide a just-in-time module.

 Preparing data: Data preparation is the step

where data is processed, to find duplications,

noise, distortion, and skewed data. Well-

processed and prepared data improves the

quality of the system.

 Choosing the algorithm: Choosing the machine

learning algorithm is dependent on the type of

data that is available at this stage. Several

techniques can be adopted, i.e., Deep learning

(DL) for image and speech processing, and

reinforcement learning (RL) which enables an

agent to learn through actions in a specific

environment. RL can improve precision by

learning optimal decision rules, adapts and

adjusts to user preferences to enhance the

accuracy of the system.

 Defining decision rules: Decision rules are

what make this system operational and reach

decisions. A better design is to adapt, though,

with more inputs, and the current state of the

user.

 Training the model: In this step, algorithms are

applied to train the model by loading data and

get an outcome from the system.

 Making a prediction: The outcome of the

model is the prediction that is interpreted and

processed to match a decision based on the

user. For example, a prediction that the air

quality will be unhealthy tomorrow.

 Matching the prediction with an automated

decision: A processed outcome is matched

with a decision to provide the user with an

HEALTHINF 2022 - 15th International Conference on Health Informatics

120

automated decision on time. The prediction

that the air quality will be unhealthy tomorrow

will follow up with a decision to use an

antihistamine.

 Feedback for evaluation: The system must

adapt to the user’s actions, i.e., behavior, by

automatically getting feedback from the user

on an intervention. The feedback is the user’s

action after an intervention. This feedback

works as an input to the system.

8.1.2 APIs Endpoints

The system shall have an API for sending requests

and receiving a response. Using an application

programming interface (APIs) enhances the system

effectiveness. The endpoint allows a layer or a system

to use this to build the applications. The endpoint

provides the following operations.

 Receiving a request

 Loading the model

 Making the prediction

 Sending back the response

It allows an application to use the operations

where information is added as input and an output is

a response from the model.

8.1.3 The Proposed Process with Steps

Figure 5: Proposed process with steps.

Figure 5 presents the proposed model with an

iterative approach from the parameter’s identification

on the left to automated decision on the right.

8.1.4 The Feedback Mechanism for Inputs

The outcome of a JITAI is feedback from the user. It

is used as an input to the system. This feedback is

either explicit or implicit:

 Implicit feedback: This is the feedback is

automatically collected from the user based on

the decision provided to the user and the action

user takes. The health map provides a way to

gather implicit feedback.

 Explicit feedback: It is the feedback provided

directly by the user of the system. The JITAI

must have a mechanism to collect the feedback

this. It can be a simple question to the user

regarding the intervention.

8.1.5 Features Important for Systems Built

on the Framework

Table 8 below presents other features that contribute

to the system strength. The features of a system are

robust, secure, and private. These features are

important for the system that is built on the

framework.

Table 8: Features of the system.

Features Definition

Robust The system must be robust

Secure It should account for the

CIA principle of security

Private User data must be private

8.1.6 The Architecture of the System

Figure 6: Architecture of the system.

Figure 6 presents the architecture of the system. The

figure is drawn in three segments, the user, AI-model

and sources with parameters (attributes). Cloud

presents storage of data from sources into two

sections, raw data, and historical data. This data is an

input for the AI model and is provided via an API

endpoint. The model then makes a prediction and

provides an automated decision in red to the user. The

outcome of this is feedback from the system to learn

from, presented in green.

A Framework for AI-enabled Proactive mHealth with Automated Decision-making for a User’s Context

121

8.1.7 Categorization of Interventions

In a proposed model, JITAI or automated decisions

can be categorized as Red, yellow, and green. Red-

colored intervention is critical and needs urgent

attention i.e., a vibration along with the intervention

to get the user’s attention. Yellow intervention does

not have an immediate effect, A green intervention is

an acknowledgement.

Figure 7: Categorization of interventions with examples.

Figure 7 presents some examples of this approach

using a mobile application screen. The colours

represent the type of intervention with notifications.

9 CONCLUSIONS AND FUTURE

WORK

In this paper, the impact of user-level decision-

making for AI-enabled proactive mHealth is

presented. The paper reasons the importance of

context for establishing proactiveness and provides a

framework of AI-enabled proactive mHealth. The

framework aims at providing insights into the design

and implementation goals of the system. This paper

also considers various implementation challenges.

The framework identifies the need for multiple

attributes as input and sources e.g., wearables,

sensors, and other resources. Finally, a system is

proposed based on the framework, which is adaptive

and provide timely interventions, JITAI architecture

with automated decision-making is fundamental for

the implementation. To this extent, the proposed

system considers multiple sources as input to provide

timely intervention to the user. This intervention is an

automated decision, which is built on the user’s

preferences. The outcome of the system is collected

implicitly or explicitly as feedback for ensuring

adaptiveness.

For future work, the system will be developed

using the proposed framework. The system will

adhere to the prediction and prevention mechanism to

provide timely intervention with personalization.

Multiple parameters will provide a holistic view of

the user. The implementation will be an AI engine,

which depending on the datasets and availability of

the features will handle different machine learning

algorithms for clustering and classifications. The

core part of the system depends on machine learning

techniques for providing an automated decision that

is beneficial for the user. The proof-of-concept model

can then be used for further developing an AI-enabled

proactive mHealth system.

REFERENCES

Winslow, C. (2021). Introduction to Public Health|Public

Health 101 Series|CDC. Cdc.gov. Retrieved 5 October

2021, from https://www.cdc.gov/training/public

health101

WHOa (2021). Health promotion. Who.int., Retrieved 5

October 2021, from https://www.who.int/health-topics

Haseltine, W. (2021). Aging Populations Will Challenge

Healthcare Systems All Over The World. Forbes.

Retrieved from https://www.forbes.com/sites/

williamhaseltine

NIPH (2018). Norwegian Institute of Public Health. Life

expectancy in Norway. from https://www.fhi.no/en/op/

hin/population

Atella, V., Piano Mortari, A., Kopinska, J., et al., (2018).

Trends in age-related disease burden and healthcare

utilization. Aging Cell, 18(1), e12861.

WHOb (2021). GHO. Global Health Estimates: Life

expectancy and leading causes of death and disability.

Who.int., from https://www.who.int/data/gho/data/

themes/mortality-and-global-health-estimates.

CDC (2021). Chronic Diseases in America, from

https://www.cdc.gov/chronicdisease/resources/infogra

phic/chronic-diseases.htm.

WHOc (2020). Director-General's opening remarks at the

media briefing on COVID-19 - 11 March 2020.

Retrieved from https://www.who.int/director-general/

speeches/detail/

Hermes, S., Riasanow, T., Clemons, E., Böhm, M., &

Krcmar, H. (2020). The digital transformation of the

healthcare industry: exploring the rise of emerging

platform ecosystems and their influence on the role of

patients. Business Research, 13(3), 1033-1069.

FDA (2021). What is Digital Health?. U.S. FDA. Retrieved

5 October 2021, from https://www.fda.gov/medical-

devices/digital-health-center-excellence/what-digital-

health.

HEALTHINF 2022 - 15th International Conference on Health Informatics

122

WHOd (2018). GoE. Who.int. Retrieved 5 October 2021,

from https://www.who.int/observatories/global-

observatory-for-

GRAND View Research. (2021). mHealth Apps Market

Growth & Trends. February 2021. Retrieved from

https://www.grandviewresearch.com/press-release/

Amir, M. (2019). The Shift from Reactive to Proactive

Healthcare | Blackcreek. Blackcreek.io. Retrieved 5

October 2021, from https://blackcreek.io/insights/the-

shift-from-reactive-to-proactive-healthcare.

Sharma, D., Singh Aujla, G., & Bajaj, R. (2019). Evolution

from ancient medication to human-centered Healthcare

4.0: A review on health care recommender systems.

International Journal Of Communication Systems,

e4058.

IBMa 2021. Education, I. (2021). What is Artificial

Intelligence (AI)?. Ibm.com. Retrieved 5 October 2021,

from https://www.ibm.com/cloud/learn/

McConnell, M., Turakhia, M., Harrington, R., King, A., &

Ashley, E. (2018). Mobile Health Advances in Physical

Activity, Fitness, and Atrial Fibrillation. Journal Of

The American College Of Cardiology, 71(23), 2691-

2701.

Aguilera, A., Figueroa, C., Hernandez-Ramos, R., Sarkar,

U., Cemballi, A., & Gomez-Pathak, L. et al. (2020).

mHealth app using machine learning to increase

physical activity in diabetes and depression: clinical

trial protocol for the DIAMANTE Study. BMJ Open,

10(8),

Korpershoek, Y., Hermsen, S., Schoonhoven, L.,

Schuurmans, M., & Trappenburg, J. (2020). User-

Centered Design of a Mobile Health Intervention to

Enhance Exacerbation-Related Self-Mgt in Patients

With COPD (Copilot): Mixed Methods Study. JMIR,

22(6)

Baig, M. (2017). Early Detection and Self-management of

Long-term Conditions Using Wearable Technologies.

Auckland University of Technology.

Nag, N., Pandey, V., & Jain, R. (2017). Health Multimedia.

Proceedings Of The 2017 ACM On International

Conference On Multimedia Retrieval. https://doi.org/

10.1145/3078971.3080545

Dijkhuis, T., Blaauw, F., van Ittersum, et al., (2018).

Personalized Physical Activity Coaching: A Machine

Learning Approach. Sensors, 18(2), 623.

Rojas, J., & Dey, A. (2019). The Personalization of Mobile

Health Interventions. Corpus ID: 208153767

Steph U. Medium. (2020). How do the choices we make

affect our lives?. In Medium.

https://stephup.medium.com/how-do-the-choices-we-

make-affect-our-lives-20f287487966

Dartmouth, U. (2021). Decision-making process. Retrieved

5 October 2021, from https://www.umassd.edu/fycm

Abbasgholizadeh Rahimi, S., Menear, et al., (2017). Are

mobile health applications useful for supporting shared

decision-making in diagnostic and treatment

decisions?. Global Health Action, 10(sup3), 1332259.

Matthew, C, Richard, F. Artificial Intelligence in

Healthcare | Accenture. (2021). from https://www.

accenture.com/us-en/

Statista (2021). Total data volume worldwide 2010-2025

|.from https://www.statista.com/statistics/871513/

Naseer Qureshi, K., Din, S., (2020). An accurate and

dynamic predictive model for a smart M-Health system

using machine learning. Information Sciences, 538,

486-502.

Smith, A. (2021). U.S. Smartphone Use in 2015. Retrieved

5 October 2021, from https://www.pewresearch.org/

internet

Araujo, T., Helberger, N., et al., (2020). In AI we trust?

Perceptions about automated decision-making by

artificial intelligence. AI & Society, 35(611-623).

Curran, K., Nichols, E., (2010). An intensive insulin

therapy mobile phone application built on artificial

intelligence techniques. Journal Of Diabetes Science,

4(1), 209-220.

Merriam-Webster (2021). Definition of context. (2021).

Retrieved 6 October 2021, from https://www.merriam-

webster.com/dictionary/context

Djukanović, R. (2021). Asthma: A disease of inflammation

and repair. The Journal Of Allergy And Clinical

Immunology, 105(2 Pt 2), S522–S526.

European Data Protection Board (EDPB). guidelines on

automated individual decision-making and profiling –

(2021).

Cambridge (2021). Framework. Retrieved 6 October 2021,

from https://dictionary.cambridge.org/dictionary/

english

Ghazaryan, A. (2021). Applications of AI and ML in 2021

mHealth. DZone.

Gorini, A., Mazzocco, K., Triberti, S., Sebri, V., Savioni,

L., & Pravettoni, G. (2018). A P5 Approach to m-

Health: Design Suggestions for Advanced Mobile

Health Technology. Frontiers In Psychology, 9(2066).

Hertwig, R., & D Ryall, M. (2020). Nudge Versus Boost:

Agency Dynamics Under Libertarian Paternalism. The

Economic Journal, 130

(629), 1384–1415.

Jabareen, Y. (2021). Building a Conceptual Framework:

Philosophy, Definitions, and Procedure. International

Journal Of Qualitative Methods, 8(4), 49-62.

Sagner, M., McNeil, et al. (2018). The P4 Health Spectrum

- A Predictive, Preventive, Personalized and

Participatory Healthspan. Progress In Cardiovascular

Diseases, 59(5).

Saha, D. (2021). How The World Became Data-Driven,

And What’s Next. Retrieved 6 October 2021, from

https://www.forbes.com/sites

Soobiah, C., Cooper, M., (2021). Identifying optimal

frameworks to implement or evaluate digital health

interventions: a scoping review protocol. BMJ Open

2020, 10.

Toro-Ramos, T., Kim, Y., et al. (2017). Efficacy of a mobile

hypertension prevention delivery platform with human

coaching. Journal Of Human Hypertension, 31(12),

795.

Yom-Tov, E., Feraru, G., Kozdoba, et al., (2021).

Encouraging Physical Activity in Patients With

Diabetes: Intervention Using a Reinforcement Learning

System. Journal Of Medical Internet Research, 19(10).

A Framework for AI-enabled Proactive mHealth with Automated Decision-making for a User’s Context

123

Bove, L. (2019). Increasing Patient Engagement Through

the Use of Wearable Technology. The Journal For

Nurse Practitioners, 15(8), 535-539.

IBMb (2021). Explainable AI IBM. Retrieved 6 October

2021, from https://www.ibm.com/watson/explainable-

Winsen (2021). MQ 135 manual. Retrieved 6 October

2021, from https://www.winsensensor.com/d/files/

PDF/

Nahum-Shani, I., Smith, S, et al., (2017). Just-in-Time

Adaptive Interventions (JITAIs) in Mobile Health: Key

Components and Design Principles for Ongoing Health

Behavior Support. Annals Of Behavioral Medicine,

52(6), 446-462.

Osman, M. (2016). Nudge: How Far Have We Come?.

Oeconomia, (6-4), 557-570. doi: 10.4000/oeconomia.2

490

Adafruit 2021. PM2.5 Air quality sensor. (2021). Retrieved

6 October 2021, from https://cdn-learn.adafruit.com/

downloads/pdf/

Bosch (2021). BME280. Retrieved 6 October 2021, from

https://www.bosch-sensortec.com/products/

environmental-

Arduino (2021). UNO R3. Retrieved 6 October 2021, from

https://docs.arduino.cc/hardware/uno-rev3

Janet, H., (2016). Wearable Technology and Mobile

Innovations for Next-Generation Education. Advances

In Educational Technologies And Instructional Design.

Wu, M., & Luo, J. (2021). Wearable Technology

Applications in Healthcare: A Literature Review.

Retrieved 6 October 2021, from https://www.

himss.org/resources/wearable

Menictas, M., Rabbi, et al., (2019). Artificial intelligence

decision-making in Mobile health. The Biochemist,

41(5), 20-24.

De Arriba-Pérez, F., Caeiro, et al., (2016). Collection and

processing of data from wrist wearable devices in

multiple-user scenarios. Sensors, 16(9), 1538. MDPI

AG,

Krist, Alex H et al. “Engaging Patients in Decision-Making

and Behavior Change to Promote Prevention.” Studies

in health technology and informatics vol. 240 (2017):

284-302.

HEALTHINF 2022 - 15th International Conference on Health Informatics

124