Remote Patient Monitoring Systems based on Conversational Agents for

Health Data Collection

Pedro Dias

1,2 a

, Miguel Cardoso

1 b

, Federico Guede-Fernandez

1 c

, Ana Martins

1 d

and Ana Londral

1,2 e

Value for Health CoLAB, Lisboa, Portugal

Comprehensive Health Research Center, Nova Medical School, Nova University of Lisbon, Lisboa, Portugal

Keywords:

Healthcare, Chatbot, Conversational Agents, Cardio-thoracic Surgery, Hypocoagulability, Telemonitorization,

Patient-reported Outcomes.

Abstract:

The pursue of digital health has been increasing in the past years and the COVID-19 pandemic promoted it fur-

ther. Remote monitoring health care allows patients to report health outcomes and receive a proper follow-up

from home and personalized health care by preventing unnecessary trips to hospitals. The design, development

and use of two rule-based chatbots for data collection and guidance providing in two health telemonitoring

contexts, post-cardiothoracic surgery for derived-complications control and patients with hypocoagulation, is

described in this paper. The designed chatbots have the goal of being simple, modular and human guided.

The ﬁrst chatbot was used to collect photos from the surgical wound and the second was used to collect the

INR value (from a coagulometer) and six related questions, following a measurement plan. In both use cases

the clinical team could analyze the collected data and interact with patients using a web application. This

chatbot may contribute to the increase of the safety perception of the patient and their engagement with their

health status. The inclusion of the clinical team in the development was key to identify the requirements and

to improve the user experience.

1 INTRODUCTION

The pursue of digital health increased during the past

years and the COVID-19 pandemic promoted it fur-

ther. Preventing unnecessary trips to hospitals by con-

stantly monitoring patients through Internet of Things

(IoT) devices, mobile apps, or simply through mes-

sages shows promising results. Severe acute respi-

ratory syndrome corona-virus 2 (SARS-CoV-2) is a

highly contagious corona-virus which has put at risk

human health since 2019. Due to its high conta-

gious rate and severity, it spread all over the world

and caused difﬁcult prognosis or even death to elderly

individuals and individuals with non-communicable

diseases (NCDs) (P

ecout et al., 2021). NCDs are

long-term and in some cases life-long pathologies and

include cardiovascular diseases, cancer, chronic respi-

https://orcid.org/0000-0002-8875-2228

https://orcid.org/0000-0002-6290-9098

https://orcid.org/0000-0003-2762-0333

https://orcid.org/0000-0002-4890-3941

https://orcid.org/0000-0002-8002-6790

ratory diseases, diabetes and neurological disorders.

Patients living with NCDs require constant care to

delay disease progression. This leads to both health

and economic consequences due to the growing age

of the worldwide population (Vandenberghe and Al-

brecht, 2020). During the pandemics, patients have

been avoiding hospitals to prevent contagion which

might lead to an increased disease severity and higher

costs to the healthcare system (Kardas et al., 2021).

Remote monitoring healthcare allows patients to

receive a proper follow-up from home and personal-

ized health care, which has been revealed especially

important since the Covid-19 outbreak (Mantena and

Keshavjee, 2021). In the healthcare context, IoT-

enabled devices can be deﬁned as any device, includ-

ing computers, mobile phones, and wearable sensors

that may allow data collection, transfer and storage.

These devices can be used to monitor patients’ symp-

toms in real time (Mamdiwar et al., 2021). Point-

of-care (POC) testing devices are particularly useful

during telemonitorization because relevant parame-

ters can be measured and reported from home. In

the case of people on long-term oral anticoagulation,

812

Dias, P., Cardoso, M., Guede-Fernandez, F., Martins, A. and Londral, A.

Remote Patient Monitoring Systems based on Conversational Agents for Health Data Collection.

DOI: 10.5220/0011011000003123

In Proceedings of the 15th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2022) - Volume 5: HEALTHINF, pages 812-820

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

these tests can be used to measure the international

normalized ratio (INR) and adjust accordingly the

medication dose (Heneghan et al., 2016).

Conversational agents (CAs) are software tools

that mimic a human-like entity during a conversation

using, text, voice, images or a mixture of all (Laranjo

et al., 2018). CAs are usually deployed on messag-

ing apps, websites or mobile-phone applications, as

well as multimodal platforms. They have been used

in a wide range of domains such as customer service,

technical support, marketing, education and health-

care. Several authors have demonstrated that humans

tend to attribute human-like attributes to non-human

agents. Due to this trait about human-machine inter-

action, CAs are perceived as social interaction part-

ners, capable of establishing a meaningful relation-

ship (Bickmore and Cassell, 2001; Epley et al., 2007).

Therefore, a CA can be perceived by the patient as a

health counselor that is available to provide continu-

ous guidance. This continued support can be given by

providing reminders or information about medication

or treatment plan, assisting on self-monitoring tasks,

or by interacting with healthcare professionals report-

ing the results of monitorization.

POC testing alone can be used to do self-

monitoring and self-management. In this case, the

patient is responsible for managing her/his treatment

plan and does not have a proper feedback about

her/his recovery. IoT have changed this paradigm by

allowing healthcare professionals to monitor in real-

time the state of the patient and intervene if there is an

abnormality, which increases patient’s safety. A CA

interacts directly with the patient and can be used to

increase the patient’s literacy, help managing medica-

tion doses and treatment based on the patient’s mea-

surements and prognosis. A system that integrates

all of these technologies might promote patient’s self-

care.

The aim of the present study is to demonstrate

the applicability of ruled-based chatbots on two dif-

ferent clinical follow-up services: post-cardiothoracic

surgery and patients with hypocoagulation. In both

cases, a telemonitoring system capable of collecting

and reporting data and provide support to the patient

was developed. The principles followed to design

both chatbots as well as the technology used to de-

velop them are described in Section 3. The applica-

tion, data collection procedure, implementation and

results for each chatbot are further explained in Sec-

tion 4.2 and 4.3. In Section 6, conclusions and future

work are presented.

2 STATE OF THE ART

In 1966, the ﬁrst rule-based chatbot was developed

by Joseph Weizenbaum at MIT Intelligence labora-

tory and it was named ELIZA. ELIZA was designed

to mimic a patient-centered Rogerian psychotherapist

and answer based on the identiﬁcation of keywords

in a sentence using pattern matching (Weizenbaum,

1966). In 1972, PARRY was developed by psychia-

trist Kenneth Colby to simulate the behavior of a pa-

tient with schizophrenia (Colby et al., 1971). Psyx-

pert was an expert on disease diagnosis developed in

1987 and used to aid psychiatrists in diagnosing psy-

chotic disorders (Overby, 1987). SESAM-DIABETE

was developed in 1989 to be an interactive educa-

tional expert and provide support to insulin-requiring

diabetic patients by giving personalised guidance and

therapeutic support (Levy et al., 1989). In 1995, AL-

ICE (Artiﬁcial Linguistic Internet Computer Entity)

was the ﬁrst computer program to use Natural Lan-

guage Process (NLP) to interpret user input (Wallace,

2009).

Nowadays, CAs can be delivered using smart-

phone, web or computer-based apps (Chaix et al.,

2019; Denecke et al., 2018; Kamita et al., 2019),

smartphone embedded software (Griol and Callejas,

2016) or any other messaging app, such as Telegram

or Facebook Messenger (van Heerden et al., 2017;

Fitzpatrick et al., 2017; Casas et al., 2018). It was

shown that the use of apps that need to be regularly

updated leads to high dropout rates and non-usage

(Lee et al., 2018). Messaging applications are the

preferable mean to convey a conversation since most

of the population is used to use them in the daily life

(Tudor Car et al., 2020).

In healthcare, CAs have been used as a mean of

conveying information between healthcare providers

and patients. CAs have been applied to many health

care sectors such as mental health (Vaidyam et al.,

2019; Luo et al., 2021), physical activity (Schachner

et al., 2020), HIV prevention (Marcus et al., 2020),

oncology (Abd-Alrazaq et al., 2021), nutritional dis-

orders (Pereira and D

ıaz, 2019), neurological disor-

ders (Pereira and D

ıaz, 2019) and chronic diseases

(Schachner et al., 2020; B

erub

e et al., 2021).

Personal patient data and data collected during a

conversation that might change the patient behavior

can be used to customize the output messages and im-

prove the user experience (Abashev et al., 2017). Ad-

ditionally, in cases where the CA has medical knowl-

edge provided by a secure source, such as medical

databases, helpful context can be automatically gener-

ated through the CA which is particularly relevant for

educational chatbots (Bickmore et al., 2016). An ex-

Remote Patient Monitoring Systems based on Conversational Agents for Health Data Collection

813

ample of educational chatbot is the electronic medica-

tion management assistant (eMMA). eMMA was de-

veloped to empower patients by giving them the max-

imum information about their treatment. Reminders,

interactive medication plans with information about

each medicine and food interaction data are provided

to the user (Tschanz et al., 2018).

Human-like communicative behaviors that can

have positive effects on this therapeutic relationship

have been largely studied (Van Pinxteren et al., 2020).

Potts et al. (2021) studied the preferred human-like

characteristics that would be desired for a chatbot to

help support a mental health service. They found that

the desired chatbot would have a positive outlook, ac-

cessible for any person and be able to support while

recalling previous conversations (Potts et al., 2021).

During the interaction, CAs must endure a relation-

ship with the user. Moore et al. (2017) mentioned that

it is important to ensure that the user is engaged: (1)

the dialogue is tailored to match the user’s level of un-

derstanding, (2) interactions are simple and synthetic

and (3) rephrasing when the user does not understand

the message (Moore et al., 2017).

NCDs are hard to manage for patients recently di-

agnosed, who need to learn how to adapt their lives

and to this new normality. CAs were identiﬁed as im-

portant factors in mitigating knowledge gaps and as

supportive agents since a closer relationship between

the patient and the healthcare provider might be en-

dorsed and, additionally, might be used as a mean to

increase literacy and signalling high-risk situations by

monitoring vital parameters (Heneghan et al., 2016;

Guhl et al., 2020; Echeazarra et al., 2021; Bian et al.,

2020; de Pennington et al., 2021).

Long-term oral anti-coagulation monitorization

has been used to estimate the dosage of anticoagu-

lant given the international normalized ratio (INR)

test. At ﬁrst, the patient would need to provide to

her/his healthcare provider the INR and accordingly

to its value the doctor or nurse would be responsi-

ble for choosing the right anti-coagulation dose. Self-

management would be required to trained patients. In

this case, given the INR test, they would be responsi-

ble for interpreting the results and adjusting the med-

ication (Heneghan et al., 2016). This was particularly

relevant on increasing medication adherence.

In 2020, a randomized controlled trial (RCT)

aimed on increasing the adherence to medication by

sending reminders and the quality of life of patients

with atrial ﬁbrillation. An embodied CA was used

to improve communication, prevent hospital readmis-

sion, and educate patients. This study revealed im-

provements in patient’s quality of life, daily activity

and a self-reported adherence to anticoagulation ther-

apy (Guhl et al., 2020).

A RCT using a Telegram based chatbot assistant,

TensionBot, was performed to monitor patients with

high blood pressure (BP). Its main functionalities

are sending reminders to patients to ensure that BP

is measured, storing measurements and letting data

available to healthcare professional and, giving addi-

tional support about how to measure BP (Echeazarra

et al., 2021). This study revealed that patients using

this CA were able to improve their ability to properly

measure BP and diminish the paperwork for health-

care providers (Echeazarra et al., 2021).

A postoperative follow-up system of orthopedic

patients was performed in an exploratory quantita-

tive and qualitative study in 2020. This system was

based on a CA that would perform a series of ques-

tions through a cellphone call based on a template

given by the medical staff and patient’s personal data.

Patient’s responses would be processed using speech

recognition and spoken language understanding tech-

niques. This dialogue was about patient’s satisfaction

regarding the hospital service, health education and

wound recovery. Doctors and nurses would have ac-

cess to the feedback report and be able to intervene

when needed. They found that AI-assisted follow-up

could replace traditional follow-up (Bian et al., 2020).

In 2021, a proposal project that aimed to evalu-

ate the effectiveness, usability, and acceptability of

a CA named Dora. This conversation model would

integrate speech recognition and generation to per-

form telephone follow-up by asking questions regard-

ing patient’s recovery (de Pennington et al., 2021).

3 DESIGN OF CHATBOT

The designed chatbot can be categorized as a rule-

based task-oriented chatbot which has the goal of be-

ing clear, modular and human guided. Saﬁ et al.

(2020) identiﬁed four stages that should be considered

during CA design to facilitate user engagement: text

understanding, dialog management, data management

and text generation (Saﬁ et al., 2020). The design

of the proposed chatbot has considered the aforemen-

tioned four stages:

• Text understanding: it refers to the process of ﬁg-

uring out what is the meaning or intention of the

user’s input. This process can be done by using

pattern matching (keyword or string matching) or

intelligent models using NLP or Machine Learn-

ing (ML) models (Saﬁ et al., 2020). Some stud-

ies revealed the embryonic stage of NLP mod-

els in healthcare due to their lack of consistent

methodology and evaluation methods (Schachner

Scale-IT-up 2022 - Workshop on Scaling-Up Health-IT

814

et al., 2020). The proposed chatbot is used for

telemonitoring purposes and vital signs gathering,

thus it exists in a structured and organized context

which opposes to conversational agents that work

directly with free-text.

• Dialog management: it is the process of establish-

ing a link between user’s input and the CAs re-

sponse. This can be handled using ﬁnite, frame

or agent-based dialogue system. A ﬁnite-based

dialogue is a state transition network that can be

deﬁned by a sequence of pre-determined interac-

tions between the user and the CA. In this case,

user’s dialogue is limited to a set of options and

the dialog ﬂow is ﬁxed. A frame-based dialogue

is characterized by a dialog ﬂow determined by

the content of the user’s input and the CAs data.

The answer given by the user will be used to ﬁll

the missing data from the task. Agent-based di-

alogue is deﬁned as a dynamic dialogue between

two agents that have enough intelligence to pur-

sue a conversation based on reasoning and the

sequence of arguments discussed during the di-

alogue (McTear et al., 2016; Saﬁ et al., 2020).

In this study, a ﬁnite-state dialogue machine was

developed to increase the simplicity of the inter-

action with the user while offering the necessary

modularity to give support during the follow-up.

• Data management: the proposed CA was de-

signed to collect patient’s data to generate a daily

report to the healthcare professional for further

analysis. Additionally, health-related information

is led available to users, in order to improve their

health literacy.

• Text generation: it is responsible for answering

the user. This answer can be ﬁxed or generated

in case the output is built on top of intelligent

models that use NLP or DL (Saﬁ et al., 2020).

Systems that give ﬁxed responses can be tedious

and frustrating to the user due to their inﬂexi-

bility (Williams et al., 2018; Irfan et al., 2020).

Intelligent agent might increase user engagement

(Schumaker et al., 2007), the lack of robust eval-

uation methods together with the black box effect

related to machine learning models (Tudor Car

et al., 2020) might result in adverse consequences

to the patient (Laranjo et al., 2018). Therefore, the

text generation of the proposed chatbot is based on

pre-deﬁned and ﬁxed text.

That being said, following the proposed design we

implemented two distinct chatbots for two use cases

using Django, a Python-based framework as their

backbone and Twilio as the cloud communication

platform to send and receive text messages through

Short Message Service (SMS) and WhatsApp mes-

sages. What is more, to handle data management we

leverage the open-source database management sys-

tem PostgreSQL. Finally, to host both of our applica-

tions we use the platform as a service (PaaS) Heroku.

4 APPLICATIONS

In this section we will describe two applications de-

veloped under a multidisciplinar team of healthcare

professionals, engineers and linguists: CardioFollow

and HemoControlBot. These applications aimed at

improving the follow-up of post-surgery (CardioFol-

low) and patients with hypocoagulation (HemoCon-

trolBot) by giving them support and additional in-

formation. Both use cases had the objective of col-

lecting patient reported outcomes measures (PROM)

and to display this data on a web application to the

pertinent clinical staff (doctors and nurses), follow-

ing the human computer interaction principles. The

chatbot and the web application were developed in a

co-creation environment, meaning that the members

of the clinical staff had a pivotal role in their design

and improvement. The target population for both use

cases were patients that were recovering from a car-

diothoracic surgery. What is more, the chatbots im-

plemented for the introduced use cases leverage either

SMS or Whatsapp. In this section, we deﬁne the com-

mon web application that sustained both use cases,

specifying the details of each use case. Afterwards

we explain the functionality of each developed chat-

bot.

4.1 Clinical Team’s Web Application

The web application is the bridge that connects pa-

tients with the clinical team. Whilst patients inter-

act through the chatbot, the clinical staff monitors the

patients’ health closely through the developed appli-

cation. The application goal is to show patients data

through different formats and layouts which were de-

signed side by side with the clinical team - through

icons, tables and interactive graphics. The use of

icons allow a quick understanding of a patient’s health

status, namely changing colors according to speciﬁc

events - a sudden alteration in any collected measure

or value outside of a speciﬁed range. The graphics

were used to display the data overtime, giving the

clinical team the ability to compare the last reported

value with the previous ones.

A screenshot of the web application is depicted in

Fig. 1. This ﬁgure provides information regarding all

patients and offers an overview of their health status

Remote Patient Monitoring Systems based on Conversational Agents for Health Data Collection

815

with the goal of being easily readable to the clinical

team: all measurements are identiﬁed with a speciﬁc

icon and their values are displayed in the same rectan-

gle. When the values outbound the predeﬁned healthy

range, the color will change to properly identify the

high-risk situations. Thus, allowing the clinical team

to prioritize the patients according to the number of

alerts.

Figure 1: List of telemonitored patients with the respective

outcomes and alerts. Measure that are outside of the normal

range are represented with the color yellow. The names of

the participants were hidden with a grey box.

Furthermore, the web application also has direct

access to the dialogues within the chatbot which can

be used for the clinical team to communicate directly

to the patient, through the developed web application.

4.2 CardioFollow

This chatbot was used in a pilot that consisted of the

telemonitoring of patients that underwent cardiotho-

racic surgery. Complications during surgery or hos-

pitalization are common (Crawford et al., 2017), but

risk also extends to the postoperative period, lead-

ing to hospital readmission of around 15-20% of pa-

tients during the ﬁrst month and 30% in the ﬁrst year

(Efthymiou and O’Regan, 2011; McElroy et al., 2016;

Khoury et al., 2020). This pilot had the participation

of 35 patients so far, as depicted in Table 1. These

patients, after the hospital discharge, were telemoni-

tored for 1 month.

Table 1: Overview of both use cases. TD - Telemonitoring

Duration; PTP - Previously Telemonitored Patients; PTF -

Patients to be Telemonitored in the Future.

Use Case TD PTP PCT PTF

CardioFollow 1 month 35 0 150

HemoControlBot 6 months 0 1 29

During this period, the patients had a daily rou-

tine where they would report their blood pressure,

weight, number of steps, heart rate and answer six

questions, using IoT devices connected to a mobile

application (Lopes et al., 2019). Whereas, the de-

veloped chatbot was used to collect daily photos of

the patients’ surgical wounds. That being said, we

needed only a subset of the design deﬁned above.

Namely, dialog management, text generation and data

management. More speciﬁcally, when the patient

sends the photo of his/her surgical wound, the chat-

bot needs to be able to understand it, respond with

clinically approved messages, and store the picture

for further analysis by the clinical team. The chat-

bot text generation feature can be divided into two

parts. The ﬁrst part consists of an appreciation mes-

sage that was used to engage the patient and to cre-

ate some kind of empathy with him/her. The second

part consists of a set of day-dependent literacy mes-

sages used to provide guidance during the patient’s

recovery/telemonitoring. This message calendar was

developed with the clinical team based on leaﬂets that

were given to patients after hospital discharge. The

message was sent right after the photo was sent be-

cause it is at that moment that the patient has his/her

attention focused on the smartphone.

This speciﬁc use case is going to be expanded to

more 150 patients, as depicted in Table 1 and evalu-

ated throughout a larger span of time where more im-

provements resulting from the clinical staff require-

ments will be developed.

4.3 HemoControlBot

This use case albeit developed in a similar context

that Cardiofollow, was targeted speciﬁcally to as-

sess hypocoagulation, through the International Nor-

malized Ratio (INR) and additional health outcomes,

through multiple-choice questions with the goal of

evaluating if the use of a coagulometer allowed the

patient a longer and better stay in the therapeutic

range, when compared to standard clinical follow-up.

That being said, the entirety of the interaction of the

patient occurs through the chatbot, which is respon-

sible for delivering the multi-step questionnaires, val-

idating and storing the data. The patients are given

a coagulometer (used to measure the INR), the chat-

bot’s phone number and a code representing a mea-

surement plan deﬁned by the clinical staff. A mea-

surement plan is a set of measures with a certain pe-

riodicity. In this speciﬁc use case the periodicity is

set to 7 days and consists of 7 items: INR value and

6 questions, related to medication, bruise, bleeding,

feces, nausea and trips to the hospital/health center.

This chatbot is currently being tested with one patient

and 29 more will enter the program, as depicted in Ta-

ble 1. Each patient will report his/her outcomes to the

chatbot for 6 months.

The chatbot implemented for this use case used

all the four features deﬁned in Section 3. Namely, in

order for the patient to register into the chatbot we

Scale-IT-up 2022 - Workshop on Scaling-Up Health-IT

816

make use of the four features since we offer a self-

enroll approach. That being said, the interaction is

triggered by a patient’s message saying ”hello”. The

chatbot initiates the enrollment process requesting the

measurement plan code, name, birth date and gender,

step by step. Thus, the chatbot needs to understand,

store, manage and validate the entirety of the interac-

tion. Furthermore, the chatbot validates the user input

accepting only valid answers, that are measurement

dependent and in turn are deﬁned by regular expres-

sions. Moreover, by leveraging the dialog manage-

ment and text generation capabilities of the chatbot,

we display the expected input format and give proper

feedback if the input does not match with the expected

format. Furthermore, if the measurement is a ques-

tion, we showcase the possible answers to the user.

All the answers are preceded with a number that al-

lows the user to answer with the respective number.

This way, the answering process is faster and easier to

the patient. In addition, due to the data management

capability, we can store all the measures and user-

deﬁned thresholds for each measurement, which then

permits the chatbot to send an alert (for the patient and

the clinical staff) whenever any measure goes outside

of the expected ranges, deﬁned by the clinical team.

Finally, with the goal of allowing the patient to vali-

date his/her answers, the chatbot always asks for con-

ﬁrmation before storing the reported outcomes. If the

patient does not validate the answers, the chatbot re-

sets to the beginning of the measurement plan. The

ﬂow behind this chatbot is depicted in Fig.2.

An example of a conversation between a patient

and the chatbot is depicted in Fig. 3. The green

speech bubbles represent the patient and the black

speech bubbles represent the chatbot.

In this use case, the requirements were trans-

lated into new features to the web application.

Namely, manage measurement plans speciﬁc for a

given service, in this case there is only the cardio-

thoracic service. Nevertheless, multiple services

from multiple hospitals could use the developed soft-

ware/solution/application. A measurement plan is

used within two contexts: periodic reporting as noted

previously, and on-demand reporting, which is trig-

gered by the clinical staff. Thus, our web application

was extended to be able to create and manage mea-

surement plans, as well as their periodicity and allow-

ing the clinical staff to request speciﬁc measures if de-

sired. Intuitively, we offer a easy to use interface for

the clinical staff to manage all the features described

above.

Figure 2: Measure collection ﬂow. The circle represents the

patient, the diamond represents the ﬁnite state machine and

the rectangle represents the chatbot.

Figure 3: Example of a conversation between a patient and

the chabot.

Remote Patient Monitoring Systems based on Conversational Agents for Health Data Collection

817

5 DISCUSSION

As it is mentioned in subsections 4.2 and 4.3, two CAs

were developed to provide support to patients after

discharge from the hospital. The information deliv-

ered to the patient was deﬁned after several meetings

with the clinical team and it contains guidelines to im-

prove recovery. These chatbots are specially impor-

tant due to the huge amount of information provided

to the patient after diagnosis, which usually lets the

patient overwhelmed and might be difﬁcult to remem-

ber when away from the hospital. Thus, with these

solutions we aim at providing health literacy, solve

the issue regarding the lack of understanding of med-

ications regimens, monitor vital signs while targeting

high-risk situations, and equip the clinical team with

an easy to use web application to communicate and

understand the patient’s recovery.

The CAs design aimed at the development of a

system adapted to most patients. Not all patients

could be included due to the exclusion criteria of the

two use cases. These criteria were illiteracy and the

inability to manage a cellphone. However, if the pa-

tient spends his/her recovery period with a capable in-

formal caregiver, the telemonitoring can still occur,

which happened with some patients from CardioFol-

low.

The user experience of the selected patients might

be improved by attending to the degree of patient’s

literacy and adjusting the dialogue to improve the en-

gagement with the CA. This could be done by having

different dialogue-frames given the patient’s literacy

level.

Patients mentioned the importance of having a

non-invasive tool that at speciﬁc timings can provide

support when needed. After diagnosis, patients tend

to look for adapted normality. A supportive tool that

empowers patients by capacitating them to take their

own decisions and feel safe from home, might be the

key to giving them this adapted normality.

6 CONCLUSIONS

In this work, we successfully deﬁned a generalized

design of a rule-based task-oriented chatbot, which

we then applied to two different use cases. Our initial

experiments showed that offering a conversational-

based interface to collect PROMs is an efﬁcient way

to keep the end-users - the patients - engaged. Mo-

bile phones are ubiquitous, so most of the popula-

tion can beneﬁt from this way of outcome collection.

Using SMS or Whatsapp as an outcomes collection

channel is a good way to interact with patients be-

cause nowadays anyone carries their cell phone with

them, thus being able to respond quickly to what is re-

quested. Although the interface is simple and easy-to-

use, managing a mobile phone requires some level of

digital literacy, thus not being useful for every type of

patient. Both use cases described in this paper, were

developed with the objective of being extended to col-

lect outcomes from patients with other health condi-

tions.

From our experience, telemonitored patients feel

safer and more engaged regarding their health status,

so this work is a starting point to pose remote patient

monitoring as a way of preventing clinical complica-

tions, due to its close follow-up, and of leading to an

increase of the patients health literacy level. More-

over, from the clinical team point of view, having

the possibility to analyze more outcomes will help to

make better clinical decisions and adopt a more per-

sonalized treatment for each patient.

Finally, regarding the chatbot and web applica-

tion development, including the clinical team in the

process is of the utmost importance for two reasons.

First, they are the domain experts with the necessary

knowledge to identify the requirements for such sys-

tems to be of any use. Second, they are the end-users,

and user experience is extremely important within

this context. Ultimately, the developed systems must

serve the clinical team, not the opposite.

7 FUTURE WORK

Both of the use cases described above took place

within the context of cardiothoracic surgery in Por-

tugal. Thus, intuitively, future work lies in expanding

these conversational agents to other contexts, which

will require further development to adapt to speciﬁc

requirements that might arise. However, these two

use cases are still undergoing as depicted in Table 1.

There is currently no way to show to the patient a

summary of his/her latest measurements, so develop-

ing a way of showcasing this data is a priority in the

near future. Regarding user interaction, increasing the

intelligence of the chatbot might be a way of decreas-

ing input errors and giving advice. This can be done

by suggesting, predicting or correcting the user’s in-

put.

ACKNOWLEDGEMENTS

The authors would like to acknowledge the Roche Di-

agnostics and the cardiothoracic service of the Hos-

pital de Santa Marta for their contributions to car-

Scale-IT-up 2022 - Workshop on Scaling-Up Health-IT

818

rying out the study. This research has been sup-

ported by the project DSAIPA/AI/0094/2020 from the

Fundac¸

ao para a Ci

encia e Tecnologia AI 4 COVID-

19 Program.

REFERENCES

Abashev, A., Grigoryev, R., Grigorian, K., and Boyko,

V. (2017). Programming Tools for Messenger-Based

Chatbot System Organization: Implication for Outpa-

tient and Translational Medicines. BioNanoScience,

7(2):403–407.

Abd-Alrazaq, A. A., Alajlani, M., Ali, N., Denecke, K.,

Bewick, B. M., and Househ, M. (2021). Perceptions

and Opinions of Patients About Mental Health Chat-

bots: Scoping Review. Journal of Medical Internet

Research, 23(1):e17828.

Bian, Y., Xiang, Y., Tong, B., Feng, B., and Weng, X.

(2020). Artiﬁcial Intelligence–Assisted System in

Postoperative Follow-up of Orthopedic Patients: Ex-

ploratory Quantitative and Qualitative Study. Journal

of Medical Internet Research, 22(5):e16896.

Bickmore, T. and Cassell, J. (2001). Relational Agents:

A Model and Implementation of Building User Trust.

In Proceedings of the SIGCHI conference on Human

factors in computing systems, pages 396–403, Seattle,

WA, USA.

Bickmore, T. W., Utami, D., Matsuyama, R., and Paasche-

Orlow, M. K. (2016). Improving Access to Online

Health Information With Conversational Agents: A

Randomized Controlled Experiment. Journal of Med-

ical Internet Research, 18(1):e1.

erub

e, C., Schachner, T., Keller, R., Fleisch, E., V Wan-

genheim, F., Barata, F., and Kowatsch, T. (2021).

Voice-Based Conversational Agents for the Preven-

tion and Management of Chronic and Mental Health

Conditions: Systematic Literature Review. Journal of

Medical Internet Research, 23(3):e25933.

Casas, J., Mugellini, E., and Khaled, O. A. (2018). Food

Diary Coaching Chatbot. In Proceedings of the 2018

ACM International Joint Conference and 2018 In-

ternational Symposium on Pervasive and Ubiquitous

Computing and Wearable Computers, pages 1676–

1680, Singapore Singapore. ACM.

Chaix, B., Bibault, J.-E., Pienkowski, A., Delamon, G.,

Guillemass

e, A., Nectoux, P., and Brouard, B. (2019).

When Chatbots Meet Patients: One-Year Prospective

Study of Conversations Between Patients With Breast

Cancer and a Chatbot. JMIR Cancer, 5(1):e12856.

Colby, K. M., Weber, S., and Hilf, F. D. (1971). Artiﬁcial

Paranoia. Artiﬁcial Intelligence, 2(1):1–25.

Crawford, T. C., Magruder, J. T., Grimm, J. C., Suarez-

Pierre, A., Sciortino, C. M., Mandal, K., Zehr, K. J.,

Conte, J. V., Higgins, R. S., Cameron, D. E., and

Whitman, G. J. (2017). Complications After Cardiac

Operations: All Are Not Created Equal. The Annals

of Thoracic Surgery, 103(1):32–40.

de Pennington, N., Mole, G., Lim, E., Milne-Ives, M., Nor-

mando, E., Xue, K., and Meinert, E. (2021). Safety

and Acceptability of a Natural Language Artiﬁcial In-

telligence Assistant to Deliver Clinical Follow-up to

Cataract Surgery Patients: Proposal. JMIR Research

Protocols, 10(7):e27227.

Denecke, K., Hochreutener, S., P

opel, A., and May, R.

(2018). Self-Anamnesis with a Conversational User

Interface: Concept and Usability Study. Methods of

Information in Medicine, 57(05/06):243–252.

Echeazarra, L., Pereira, J., and Saracho, R. (2021). Ten-

sioBot: a Chatbot Assistant for Self-Managed in-

House Blood Pressure Checking. Journal of Medical

Systems, 45(4):54.

Efthymiou, C. A. and O’Regan, D. J. (2011). Postdischarge

complications: what exactly happens when the patient

goes home? Interactive CardioVascular and Thoracic

Surgery, 12(2):130–134.

Epley, N., Waytz, A., and Cacioppo, J. T. (2007). On seeing

human: a three-factor theory of anthropomorphism.

Psychological review, 114(4):864–886.

Fitzpatrick, K. K., Darcy, A., and Vierhile, M. (2017). De-

livering Cognitive Behavior Therapy to Young Adults

With Symptoms of Depression and Anxiety Using a

Fully Automated Conversational Agent (Woebot): A

Randomized Controlled Trial. JMIR Mental Health,

4(2):e19.

Griol, D. and Callejas, Z. (2016). Mobile Conversational

Agents for Context-Aware Care Applications. Cogni-

tive Computation, 8(2):336–356.

Guhl, E., Althouse, A. D., Pusateri, A. M., Kimani, E.,

Paasche-Orlow, M. K., Bickmore, T. W., and Mag-

nani, J. W. (2020). The Atrial Fibrillation Health Lit-

eracy Information Technology Trial: Pilot Trial of a

Mobile Health App for Atrial Fibrillation. JMIR Car-

dio, 4(1):e17162.

Heneghan, C. J., Garcia-Alamino, J. M., Spencer,

E. A., Ward, A. M., Perera, R., Bankhead, C.,

Alonso Coello, P., Fitzmaurice, D., Mahtani, K. R.,

and Onakpoya, I. J. (2016). Self-monitoring and

self-management of oral anticoagulation. Cochrane

Database of Systematic Reviews, 2021(4).

Irfan, B., Hellou, M., Mazel, A., and Belpaeme, T. (2020).

Challenges of a Real-World HRI Study with Non-

Native English Speakers: Can Personalisation Save

the Day? In Companion of the 2020 ACM/IEEE In-

ternational Conference on Human-Robot Interaction,

pages 272–274, Cambridge United Kingdom. ACM.

Kamita, T., Ito, T., Matsumoto, A., Munakata, T., and In-

oue, T. (2019). A Chatbot System for Mental Health-

care Based on SAT Counseling Method. Mobile In-

formation Systems, 2019:9517321.

Kardas, P., Boven, J. F. M. v., Pinnock, H., Menditto,

E., Wettermark, B., Tsiligianni, I., and

Agh, T.

(2021). Disparities in European healthcare system

approaches to maintaining continuity of medication

for non-communicable diseases during the COVID-

19 outbreak. The Lancet Regional Health - Europe,

4:100099.

Khoury, H., Ragalie, W., Sanaiha, Y., Boutros, H., Rudasill,

S., Shemin, R. J., and Benharash, P. (2020). Read-

mission After Surgical Aortic Valve Replacement in

the United States. The Annals of Thoracic Surgery,

110(3):849–855.

Laranjo, L., Dunn, A. G., Tong, H. L., Kocaballi, A. B.,

Chen, J., Bashir, R., Surian, D., Gallego, B., Magrabi,

Remote Patient Monitoring Systems based on Conversational Agents for Health Data Collection

819

F., Lau, A. Y. S., and Coiera, E. (2018). Conversa-

tional agents in healthcare: a systematic review. Jour-

nal of the American Medical Informatics Association,

25(9):1248–1258.

Lee, K., Kwon, H., Lee, B., Lee, G., Lee, J. H., Park, Y. R.,

and Shin, S.-Y. (2018). Effect of self-monitoring on

long-term patient engagement with mobile health ap-

plications. PLOS ONE, 13(7):e0201166.

Levy, M., Ferrand, P., and Chirat, V. (1989). SESAM-

DIABETE, an expert system for insulin-requiring di-

abetic patient education. Computers and Biomedical

Research, 22(5):442–453.

Lopes, I., Sousa, F., Moreira, E., and Cardoso, J. (2019).

Smartphone-Based Remote Monitoring Solution for

Heart Failure Patients. Studies in Health Technology

and Informatics, 261:109–114.

Luo, T. C., Aguilera, A., Lyles, C. R., and Figueroa, C. A.

(2021). Promoting Physical Activity Through Conver-

sational Agents: Mixed Methods Systematic Review.

Journal of Medical Internet Research, 23(9):e25486.

Mamdiwar, S. D., R, A., Shakruwala, Z., Chadha, U., Srini-

vasan, K., and Chang, C.-Y. (2021). Recent Advances

on IoT-Assisted Wearable Sensor Systems for Health-

care Monitoring. Biosensors, 11(10).

Mantena, S. and Keshavjee, S. (2021). Strengthening

healthcare delivery with remote patient monitoring in

the time of COVID-19. BMJ Health Care Informatics,

28(1):e100302.

Marcus, J. L., Sewell, W. C., Balzer, L. B., and Krakower,

D. S. (2020). Artiﬁcial Intelligence and Machine

Learning for HIV Prevention: Emerging Approaches

to Ending the Epidemic. Current HIV/AIDS reports,

17(3):171–179.

McElroy, I., Sareh, S., Zhu, A., Miranda, G., Wu, H.,

Nguyen, M., Shemin, R., and Benharash, P. (2016).

Use of digital health kits to reduce readmission af-

ter cardiac surgery. Journal of Surgical Research,

204(1):1–7.

McTear, M. F., Callejas, Z., and Griol, D. (2016). The

Conversational Interface: Talking to Smart Devices.

Springer.

Moore, R. J., Arar, R., Ren, G.-J., and Szymanski, M. H.

(2017). Conversational UX Design. In Proceedings

of the 2017 CHI Conference Extended Abstracts on

Human Factors in Computing Systems, CHI EA ’17,

pages 492–497, New York, NY, USA. Association

for Computing Machinery. event-place: Denver, Col-

orado, USA.

Overby, M. A. (1987). Psyxpert: an expert system proto-

type for aiding psychiatrists in the diagnosis of psy-

chotic disorders. Computers in biology and medicine,

17(6):383–393.

ecout, C., Pain, E., Chekroun, M., Champeix, C., Kulak,

C., Prieto, R., van Vugt, J., Gilchrist, K., and Lain

Pellet, A.-F. (2021). Impact of the COVID-19 Pan-

demic on Patients Affected by Non-Communicable

Diseases in Europe and in the USA. Interna-

tional Journal of Environmental Research and Public

Health, 18(13).

Pereira, J. and D

ıaz, O. (2019). Using Health Chatbots for

Behavior Change: A Mapping Study. Journal of Med-

ical Systems, 43(5):135.

Potts, C., Ennis, E., Bond, R. B., Mulvenna, M. D., McTear,

M. F., Boyd, K., Broderick, T., Malcolm, M., Kuos-

manen, L., Nieminen, H., Vartiainen, A. K., Koste-

nius, C., Cahill, B., Vakaloudis, A., McConvey, G.,

and O’Neill, S. (2021). Chatbots to Support Mental

Wellbeing of People Living in Rural Areas: Can User

Groups Contribute to Co-design? Journal of Technol-

ogy in Behavioral Science, 6(4):652–665.

Saﬁ, Z., Abd-Alrazaq, A., Khalifa, M., and Househ, M.

(2020). Technical Aspects of Developing Chatbots

for Medical Applications: Scoping Review. Journal

of Medical Internet Research, 22(12):e19127.

Schachner, T., Keller, R., and v Wangenheim, F. (2020). Ar-

tiﬁcial Intelligence-Based Conversational Agents for

Chronic Conditions: Systematic Literature Review.

Journal of Medical Internet Research, 22(9):e20701.

Schumaker, R. P., Ginsburg, M., Chen, H., and Liu, Y.

(2007). An evaluation of the chat and knowledge de-

livery components of a low-level dialog system:The

AZ-ALICE experiment. Decision Support Systems,

42(4):2236–2246.

Tschanz, M., Dorner, T. L., Holm, J., and Denecke, K.

(2018). Using eMMA to Manage Medication. Com-

puter, 51(8):18–25.

Tudor Car, L., Dhinagaran, D. A., Kyaw, B. M., Kowatsch,

T., Joty, S., Theng, Y.-L., and Atun, R. (2020). Con-

versational Agents in Health Care: Scoping Review

and Conceptual Analysis. Journal of Medical Internet

Research, 22(8):e17158.

Vaidyam, A. N., Wisniewski, H., Halamka, J. D., Kasha-

van, M. S., and Torous, J. B. (2019). Chatbots and

Conversational Agents in Mental Health: A Review

of the Psychiatric Landscape. The Canadian Journal

of Psychiatry, 64(7):456–464.

van Heerden, A., Ntinga, X., and Vilakazi, K. (2017). The

potential of conversational agents to provide a rapid

HIV counseling and testing services. In 2017 Inter-

national Conference on the Frontiers and Advances

in Data Science (FADS), pages 80–85, Xi’an, China.

IEEE.

Van Pinxteren, M. M., Pluymaekers, M., and Lemmink,

J. G. (2020). Human-like communication in conversa-

tional agents: a literature review and research agenda.

Journal of Service Management, 31(2):203–225.

Vandenberghe, D. and Albrecht, J. (2020). The ﬁnancial

burden of non-communicable diseases in the Euro-

pean Union: a systematic review. European Journal

of Public Health, 30(4):833–839.

Wallace, R. S. (2009). The Anatomy of A.L.I.C.E. In Ep-

stein, R., Roberts, G., and Beber, G., editors, Parsing

the Turing Test, pages 181–210. Springer Netherlands,

Dordrecht.

Weizenbaum, J. (1966). ELIZA—a Computer Program for

the Study of Natural Language Communication be-

tween Man and Machine. Communications of the

ACM, 9(1):36–45.

Williams, T., Thames, D., Novakoff, J., and Scheutz, M.

(2018). ”Thank You for Sharing that Interesting

Fact!”: Effects of Capability and Context on Indi-

rect Speech Act Use in Task-Based Human-Robot Di-

alogue. In Proceedings of the 2018 ACM/IEEE In-

ternational Conference on Human-Robot Interaction,

pages 298–306, Chicago IL USA. ACM.

Scale-IT-up 2022 - Workshop on Scaling-Up Health-IT

820