Leveraging Adaptive Sessions Based on Therapeutic Empathy

Through a Virtual Agent

Adrián Bresó, Juan Martínez-Miranda and Juan Miguel García-Gómez

Grupo de Informática Biomédica, Instituto ITACA, Universitat Politècnica de València, Spain

1 STAGE OF THE RESEARCH

This document describes the work under

development as part of a PhD Thesis carried out in

the context of the European Project “Help4Mood –

A Distributed System to Support the Treatment

of Patients with Major Depression”

(www.help4mood.info) [FP7-ICT-2009-4; 248765].

Help4Mood aims to support the treatment of patients

with Major Depression using Information and

Communications Technology (ICT). The resultant

Personal Health System (PHS) of Help4Mood uses a

set of activity sensors (such as wrist watch, key ring,

or under-mattress sensor) that gather information

about the daily physical and sleep activities from the

patient. This information and other data obtained

through standardized questionnaires are analysed in

order to infer some recommendations (or alerts) and

offer adaptive and tailored tasks as part of the

treatment in the form of daily interactive sessions

provided by a Virtual Agent (VA).

The initial developed work for the project

resulted in the MSc Thesis entitled “Generic Data

Processing & Analysis Architecture of a Personal

Health System to Manage Daily Interactive

Sessions in Patients with Major Depression”

(Bresó, 2013) within the Artificial Intelligence,

Pattern Recognition and Digital Imaging MSc

program at the Technical University of Valencia. In

this MSc Thesis, the author describes the design and

the implementation of the Data Processing &

Analysis layer of the Help4Mood’s PHS. It had been

conceived as the module responsible to analyse

relevant patient’s data, plan the daily interactive

sessions and recommend a set of tailored activities

configured by clinicians that help in the recovery of

the patient. Additionally, this system included an

initial cognitive-emotional module aimed to infer the

specific set of emotions to be displayed by the VA

during the interaction

The research work presented in this document is

an extension of the work done in the MSc's Thesis.

The proposed extension will allow a deeper research

on how to improve current Human-Agent

Interactions particularly addressed to users that are

under treatment of Major Depression. Some theories

in the areas of psychology and cognitive science will

be the basis of a computational model that is

expected to improve the production of the daily

session’s content and the adequate –emotional–

fashion for a better engagement of the users

promoting the long-term use of the system.

2 OUTLINE OF OBJECTIVES

The proposed research work is focused on the

improvement of the treatment adherence and the

user acceptability through two main contributions in

the computer-based psychotherapy:

1. Get a modular and flexible computational

architecture to improve Human-Agent

Interactions during the daily sessions provided

as part of treatment to patients with Major

Depression.

2. Encourage the system’s realism and reliability

through the generation of VA’s empathic

responses based on a modelling of Therapeutic

Empathy.

3 RESEARCH PROBLEM

Depression is expected to be the disorder with the

highest disease burden in high-income countries by

the year 2030 (WHO, 2012). Patients with Major

Depression should follow a specific and continued

treatment based on therapeutic sessions (and/or

drugs) and evaluations to treat the illness and

achieving recovery and prevent future relapse

(Vittengl, 2007). The lengthening of treatment and

relapses increases the healthcare resources and the

Bresó A., Martínez-Miranda J. and Miguel García-Gómez J..

Leveraging Adaptive Sessions Based on Therapeutic Empathy Through a Virtual Agent.

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

healthcare costs, and reduces the quality of life in

the patients.

The magnitude of the lack of therapeutic

adherence is estimated close to 50% in patients with

chronic diseases (Sabatâe, 2003). In particular, non-

adherence in psychiatric disorders is 75% (Rigueira,

2001). The benefits of the therapy are drastically

diminished due to the lack of adherence (Sotoca-

Momblona, 2006). Despite the wide range of studies

on interventions to improve adherence to treatment

(DiMatteo, 1993; Dulmen, 2007; Sabatâe, 2003;

Coombs, 2003; Katon, 1996; Thompson, 2012),

there is no clear evidence on the effect to promote

compliance or clinical parameters (Márquez, 2001;

Rigueira, 2001).

Currently, the use of ICT in mental health is

fairly new and there are open questions and

challenges to effectively apply it. In particular,

patients with Major Depression often have

associated a distorted and negative thinking which

makes them prone to suffer anxiety when interacting

with computer systems (Safford, 1999). Much effort

is needed to develop systems that can be widely

accepted and improve the adherence to computer-

based psychotherapy.

4 STATE OF THE ART

4.1 Human-Agent Interaction in

Mental Health

The development of Human-Agent Interactions

(HAI) applied to Mental Health is a difficult task

because many of the content in therapeutic sessions

are related with personal issues (e.g. thoughts,

emotions and feelings). Therefore, a feasible

computer-based psychotherapy must gain an

acceptable level of trust from the patient.

Particularly important is the content of the message

disclosed to the patient, and the way in which the

VA displays the message.

Systems based on VA have evolved greatly over

the past two decades, improving the interactions

with users. Many studies have demonstrate the

utility of using VAs for training and learning

purposes (Core, 2006; Martínez-Miranda, 2008;

Castellano, 2013), and games (Gebhard, 2008;

Mateas and Stern, 2003; Andrist, 2013).

Regarding the health context, we can find

systems that train clinicians (Raij, 2007; Kenny,

2007), control patient adherence to medication

(Bickmore and Pfeifer, 2008), and promote physical

exercise (Bickmore, 2006). There are also works

applied to mental health, as the system that support

patients with phobias (Krijn, 2004; Brinkman,

2008), post-traumatic stress disorder (Rizzo and

Josep, 2011), systems aimed at children with autism

(Tartaro and Casell, 2008; Bernardini, 2013),

support interventions to reduce alcohol consumption

(Yasavur, 2013) or systems that implements

standardized questionnaires in order to treat

depression (Pontier and Siddiqui, 2008).

Although initial results are promising, much

effort is still needed to develop systems in which

human-agent interactions are widely accepted for its

use in computer-based psychotherapeutic

interventions. Initiatives such as the developed in the

project Help4Mood would never be able to replace

psychotherapists with the same level of

competences, but may become essential tools to

offer personalized and remote support on daily basis

during the treatment. The key lines of research in

Help4Mood include the mechanism for the planning

of the daily sessions, the generation of the verbal

and nonverbal behaviour in the VA, and the

continuous collection of physical and sleep activities

from the patient through a monitoring system.

Taking the particular characteristics of the addressed

users, the Human-Agent ·Interactions need to be

carefully designed to avoid counter-productive

effects in the target users.

Our proposal, as part of the PhD Thesis is the

construction of a computational model that improve

the current developed system through the generation

of better adaptive sessions regarding to patient’s

condition and the production of an adequate

emotional behaviour in the VA that simulates

therapeutic empathy reactions during the interaction

with the user.

4.2 Treatment Adherence

Treatment adherence was defined by World Health

Organization (WHO) in 2001 as “the extent to which

a person’s behaviour – taking medication, following

a diet, and/or executing lifestyle changes,

corresponds with agreed recommendations from a

health care provider”.

Accurate assessment of adherence behaviour is

necessary for effective and efficient treatment

planning, and for ensuring that changes in health

outcomes can be attributed to the recommended

regimen. Adherence may be measured using either

process-oriented or outcome-oriented definitions.

Outcome-oriented definitions use the end-result of

treatment, e.g. cure rate, as an indicator of success.

LeveragingAdaptiveSessionsBasedonTherapeuticEmpathyThroughaVirtualAgent

Process-oriented indicators make use of intermediate

variables such as appointment-keeping or pill counts

to measure adherence.

There are some variables which researchers are

studying in order to determine if they are influencing

the treatment adherence. Neil et al. (Neil, 2009)

investigated personal predictors of adherence in

people who use a Cognitive Behaviour Therapy

website. Neil et al. (Neil, 2009) found to be

associated with increased adherence were lower

baseline rates of depression, younger age, female

gender, and less knowledge of psychological

treatments. Regarding the factors that an ICT

solutions can influence in the treatment adherence,

we can found reviews on treatment adherence based

on "Interventions on technical solutions" (Yildiz,

2004) which are aimed at simplifying the medication

regimen in order to improve the adherence. Other

are based on "Educational interventions”

(Vergouwen, 2003) which are aimed at informing

the patient about the disease and the treatment in

order to empower him to take a more active

participation in disease management. Some others

are based on "Behavioral interventions” (Yildiz,

2004; Pampallona, 2002) such as alerts and

reminders to improve adherence.

Regarding the interventions on technical

solutions, the treatment complexity has been

identified as a cause of poor adherence. It may be

considered therefore reasonable to think that

interventions aimed at simplification would improve

compliance (Orueta, 2005). Therefore, treatment

must be customized and adapted to the daily needs

of the patient and should be less invasive as

possible.

The educational interventions strategies that

increase knowledge about the disease and treatment

seem necessary, but not sufficient to change

behaviour regarding adherence (Orueta, 2005).

However, other studies focused on schizophrenia

claim that psycho-educational interventions have no

effect on adherence (Dolder, 2003; Zygmunt, 2002;

Lincoln, 2007). In depression studies are less

numerous and with inconclusive results.

Additionally, the systems which provide

information and education about the disease, get

empower patients. The most relevant for obtain the

expected therapeutic results factor is the quality of

the doctor-patient communication (Márquez, 2001;

Donovan, 1997; Thompson, 2012). In the

psychological domain, good communication is

achieved through active listening and therapeutic

empathy.

Health care providers must understand behaviour

change as part of an interpersonal process.

Effective behavioural interventions involve the

actions that help the patients to become responsible

for their own decisions and self-care behaviours.

This behaviour change can be reflected into an

increment of treatment adherence. ICT solutions can

contribute to this change through reminders, alerts,

recommendations, self-reflection, or the planning of

routines (Delamater, 2006). Specifically, in the

treatment of major depressive disorders,

psychologists use several tools such as self-

reflection techniques that may improve

psychological outcomes (Delamater, 2006).

It is easier to make assessments of adherence

when drug treatment is performed (e.g. by applying,

Directly-Observed Therapy (DOT)). But when

treatment involves psychotherapeutic sessions, the

assessment is often made through comparisons about

Treatment Adherence Therapy (TAT) (Staring,

2006) versus Treatment As Usual (TAU) as control

group (Staring, 2010; Gilden, 2011) or using

questionnaires (Titov, 2013; Lin, 2012).

5 METHODOLOGY

The main hypothesis under research in this thesis is

that “The use of a computational model that plans

personalized daily sessions and generates

appropriate empathic responses will improve the

adherence in depressed patients avoiding an early

discontinuation of the treatment, which can result in

new episodes or relapses into major depression”.

The following sections outline the proposed

experiments and methodologies designed to address

the proposed hypothesis.

5.1 Adaptive and Flexible Daily Session

Planning

We are extending the generic and flexible session

planner implemented in the previous MSc. Thesis

(Bresó, 2013). The session planner is able to process

and analyse data from different sources (activity

sensors, demographical data, questionnaire scores,

etc.) in order to infer the patient’s condition and

generate an appropriate set of more than 10 tasks

(such as the collection of speech patterns, the

negative thoughts reflexion activity, sleep

questionnaire, or daily mood check questionnaire) as

part of the treatment. For instance, when the patient

answers the Patient Health Questionnaire (PHQ-9),

the system infers and updates the patient’s

ICAART2014-DoctoralConsortium

depression level. If the system detects a health

deterioration compared to historical values, the

system will provide a message for the patient to

contact clinical staff and switch the system to the

“emergency mode”. In this mode the system will

plan only a basic set of essential tasks specified by

the clinicians. A complementary example is when

the system detects a critical situation such as suicidal

risks. This situation can be inferred from different

questionnaires. When it is inferred, the planning of

the system is stopped and only one task will be

planned: “crisis plan” task, which will try to calm

and inform the patient about local emergency

telephones to contact for assistance.

All defined tasks are managed in order to

generate an adaptive daily session which is used by

the VA as the core content to communicate with the

users. In this sense, we are working to improve the

treatment adherence as follows:

 Interventions based on improved technical

solutions: Future sessions may be more

personalized and suitable to patient’s condition in

order to reduce the complexity of the treatment. This

personalization will be carried out by improving the

analysis of patient activity data for the patient to

estimate objectively the patient’s condition and

expanding the set of activities available to make

more suitable recommendations. The more tasks we

add to the system, more variability and more

personalization can be done. For instance, the

system should be able to automatically adjust the

length of the session to patient stamina level in order

to the patient does not get tired. Currently we are

obtaining the patient stamina level subjectively

through a direct question: “What length of session

you prefer (short/medium/long)?”. We are working

in order to infer the stamina level from the trend of

the past sessions and from gathering activity data.

 Educational interventions: The system

provides a full weekly report which contains the

patient progress and the highlights values. We would

add some psycho-educational tasks such as those

used to inform the patient about recommendations

on good habits. For instance, if the system analyses

the data acquired from mattress sensor device and

from sleep questionnaire and it detects a disturbance

in sleep, the system should provide information

related to healthy rest and sleep tips.

 Behavioural interventions: Currently, the

system alerts the patient when he/she is not using the

system or the sensors devices. We plan to generate

more sessions with content-related alerts or

recommendations such as “You reject this suggested

activity twice, are you sure that you want to reject it

again?” Additionally, we are defining a behaviour

activation task in order to suggest (and check its

execution) the patient with a set recommended

activities such as walking or socializing. Other

interesting task that we are implementing is the

relaxation task, in which the patient listen a set of

relaxing exercises in order to facilitate relaxation

and self-reflection.

Currently, some of the tasks defined in the

system are based on Cognitive Behaviour Therapy

(CBT) (Gray et al., 2002), which have demonstrated

high adherence because it uses effective cognitive

and behavioural strategies to motivate patients to

complete homework, modify their maladaptive

thinking and behaviour, and solve problems on their

own (Safren, 2012).

Clinicians can configure the available set of

tasks, their frequency, constrains and priorities in

order to customize the system to different clinical

centres, patients or different stages of the treatment.

This Planner is implemented using a (1) Rule Based

System (RBS) running with the JESS engine

(http://herzberg.ca.sandia.gov/) and a (2) planning

algorithm that we have developed.

The rule based system allows to generate the

inference of a clinically adequate set of suggested

tasks. The easy replacement of the set of rules gives

the flexibility to adapt the planner to different

clinical domains such as other mental disorders

(anxiety, phobias, panic, or bipolar). The current

rules are codding the clinical preferences of the

clinician staff involved in the project but it is also

able to code rules from clinical guidelines. Finally,

our designed architecture infers clinical concepts

(using SNOMED-CT terminology) in order to store

valuable information for clinical research purposes.

Most of these concepts have been inferred from the

analysis of the activity sensor data (e.g. “Restless”

and “Activity Sleep Time”).

Regarding the planning algorithm, it is in charge

of planning a personalised daily patient session

based on clinical specifications (frequencies,

constrains, etc.), historical tasks executions,

currently user responses, and the level of stamina.

5.2 Therapeutic Empathy Responses

The modelling of emotional competence in virtual

agents is an active research area where theoretical,

technical, and ethical considerations need to be

addressed. There are currently some computational

models of emotion developed to cover specific

components of the emotional phenomenon.

The selection of a specific emotion theory as the

LeveragingAdaptiveSessionsBasedonTherapeuticEmpathyThroughaVirtualAgent

basis of a computational model depends on the

aspects of emotion that the computational model

tries to represent. From the current existent theories

of emotion, the one that predominates above the

others in the efforts dedicated to implement

computational models of emotions is the cognitive

appraisal theory of emotions (Lazarus, 1991;

Scherer, 2001). The core concept of appraisal

theories refer that the events in a person‘s

environment are constantly identified and evaluated

by the individual. This cognitive evaluation (or

appraisal) process leads to an emotional response

(according to the event‘s relevance for the person)

which in turn generates a specific behaviour to cope

with the appraised events.

In the context of the Help4Mood project and in

the subsequent MSc's Thesis (Bresó, 2013), we

developed a Cognitive-Emotional module which is

responsible to produce emotional responses in the

VA during the interaction with the user. This module

has been developed as a stand-alone Java application

which makes use of the FAtiMA (FearNot Affective

Mind Architecture) software libraries

(http://sourceforge.net/projects/fatima-modular/)

(Dias, 2011). FAtiMA is one of the existent

(appraisal theory based) computational architecture

of emotions which its initial objective was the

creation of empathic agents interacting in a virtual

environment to tackle and eventually help to reduce

bullying problems in schools.

With the use of FAtiMA, the emotional

behaviour of a virtual agent has been generated

through the authoring (in a XML-based format) of

their emotional reactions, action tendencies, goals

and actions based on the detection and appraisal of

the events occurred in the agent‘s environment. The

environment of the VA in Help4Mood is basically

composed from all the data received (directly and

indirectly) from the user. These data include the

objective user‘s information collected through the

sensor devices (i.e. actigraphs to identify sleep and

physical activity patterns) and the subjective self-

reported information through standardised

questionnaires. All this information is interpreted

and transformed into the specific events used as the

input to the cognitive emotional module. The

received events will produce the specific VA‘s

emotional behaviour during the interaction with the

user according to the specific goals, emotional

reactions and action tendencies defined as the

internal state of the VA.

The detection and appraisal of the events

produced in the VA’s environment lead to the

selection of specific dialogues which are jointly used

with some facial expressions to display the

emotional behaviour in the VA. When an event is

appraised as desirable for the VA (e.g. the daily

logging of the user into the system facilitating the

VA’s goal related to maintain a good level of

adherence in the patient) or desirable for the

patient’s well-being (e.g. good self-reports about

moods or thoughts) positive emotions are generated

in the VA. These emotions will disclose specific

utterances that complement the VA’s feedback

dialogues such as “Thanks for logging in today, I am

happy to see you again” or “Great! I‘m glad to hear

that”. Negative emotions are produced when the

detected events are appraised as not desirable for the

VA or the patient, which in turn produce feedback

such as “I’m sorry to know that” or “That’s not

really good”.

The displaying of negative emotions when

interacting with people recovering from depression

should be addressed carefully. In this sense, one of

the research lines proposed in this PhD is an

extension of the current implemented cognitive-

emotional module to better cope with the negative

situations reported or detected in the patient. A key

difference in the empathic behaviour of the

Help4Mood‘s VA regarding the original application

of FAtiMA is that the empathic responses (aimed to

be displayed through the dialogue content and some

facial expressions) needs to be modulated according

to the special characteristics of the target users. Our

VA must not display a “pure emotional” empathic

behaviours by adopting the same typically negative

mood of the patient. The disadvantage is that these

behaviours can be interpreted as sympathetic

expressions of condolence that may imply a sense of

unintended agreement with the patients (negative)

views (Clark, 2007). What is most beneficial from a

clinical perspective is not to produce “only” natural

empathic reactions as response to the patient’s input,

but to generate therapeutic-empathy responses in the

agent.

It is important to distinguish natural empathy

(experienced by people in everyday situations) from

therapeutic empathy in order to provide the patients

with useful feedback for their particular condition.

One of the differences between natural and

therapeutic empathy is the “addition of the cognitive

perspective-taking component to the emotional one;

the cognitive component helps the therapist to

conceptualize the client’s distress in cognitive

terms” (Thwaites, 2007). That means that a therapist

should “assume both the role of an emotional

involvement in an interview with a patient and an

ICAART2014-DoctoralConsortium

emotional detachment that allows for a more

objective appraisal” (Clark, 2007).

The extension proposed for the VA’s cognitive-

emotional module is focused in the modelling of a

perspective-taking component aimed to produce in

the VA the required emotional detachment or

emotional distance at specific stages of the

interaction with patients with major depression. The

main idea is the implementation of some strategies

for emotion regulation as defined in (Gross, 2007).

One of these strategies is the cognitive change

(through the re-appraisal process) used to assess

with a different (more positive) perspective a

particular event. This reappraisal mechanism will be

integrated as an extension of the FAtiMA

architecture.

When a new event is received from the

environment a prospective appraisal will be

executed to assess if the event derives from a

desirable or undesirable situation in terms of the

VA’s goals. The result of this prospective appraisal

is the projection of the potential emotional state

produced by this event. In other words, our model

“simulates” the appraisal and affect derivation

processes to analyse the emotional consequences of

the current situation, but without producing the full-

blown emotional responses. If the projected

emotional state includes the activation of a negative

emotion with intensity greater than a pre-defined

maximum threshold, the corresponding pre-defined

alternative event(s) is selected for reappraisal which

would construct a more positive meaning of the

original situation.

An example of this process is when the patient is

reporting a low mood during the today session. The

VA can appraise this event as highly undesirable for

the patient’s condition generating a strong negative

emotion. The new reappraisal component can

change the meaning of this situation using an

alternative view. In the example, the VA can analyse

the results obtained in the mood questionnaire

during previous sessions (stored in the model of the

patient) and check whether these results show a

positive tendency in the patient’s condition during

the past days. If a positive tendency is found, the

original event would be reappraised as “not much

undesirable” to the patient (thought the current level

of patient’s mood is not the optimal). This

reappraisal can change the emotional state or the

emotion’s intensity in the VA which is reflected in

the feedback provided to the patient, something like

“Ok, it seems that your current mood is not very

good, but in general terms you are making good

progresses in the last days”.

This proposed extension of the emotional responses

is expected to improve the VA-patient interaction

and increase the patient engagement, and therefore

contribute with the adherence to the treatment.

5.3 Clinical Pilots

There are planned 3 incremental pilots (see Table 1)

to evaluate the Help4Mood system in three

European clinical centres: (1) The Clinical Centre of

Psychology and Psychotherapy at the University of

Babes-Bolyai (Cluj-Napoca), (2) the Health

Foundation of Sant Joan de Deu (Llobregat,

Bacelona), and (3) the Centre for Population Health

Sciences, University of Edinburgh (Edinburgh, UK).

Table 1: Help4Mood pilots.

Pilot Country Patients Tasks Days

1 Edinburgh 5 1 7

2 Cluj-Napoca 4 5 15

3 Catalonia >10 >5 >15

3 Edinburgh >10 >5 >15

3 Cluj-Napoca >10 >5 >15

The first and second one has already been

performed, the third one is under development and it

is planned the first semester of 2014. The last pilot

will include the new developments that are described

in this paper. So the final testing results will evaluate

the work carry out in this PhD Thesis.

At the end of each pilot the clinicians administer

a personal interview to the patients in order to obtain

data about acceptability for each of the system’s

components.

5.4 Statistical Analyses

The gathering data from all these pilots will be

analysed to test our main hypothesis and study 3

factors: (1) system usage, (2) system engagement,

and (3) adherence to the treatment. .

A clinical requirement is that the system must be

used daily, so we can check in system logs the

patient logins in order to obtain the system usage.

Additionally, clinicians set the minimal and maximal

executions of each task to be offered to the user. In

the first pilot, only one task was configured, which

should be performed once per day during the seven

days of the system’s use. In the second pilot,

clinicians establish 5 tasks (see Table 2). In order to

obtain the adherence to the treatment, we calculated

the ratio between the maximal executions

established by clinicians with the number of tasks

completed by the user using the system.

LeveragingAdaptiveSessionsBasedonTherapeuticEmpathyThroughaVirtualAgent

Table 2: Minimum and Maximum number of weekly

executions in each task during the second pilot.

Task

Minimum

number of

executions

per week

Maximum

number of

executions

per week

Daily Mood Check

Questionnaire

7 7

Speech Activity 3 5

Patient Health

Questionnaire (PHQ-9)

1 1

Sleep Questionnaire 0 7

Negative Thought

Questionnaire

3 7

Total 14 27

System engagement is measured based in the

length (Long, Medium, Short) of the daily session

selected by the patient. The choice of long sessions

suggests greater involvement and motivation. On the

other hand, the selection of short sessions suggested

little engagement and commitment.

5.5 Initial Outcomes

The system was tested in two incremental pilots. The

first one was carried out in Edinburg during 7 days

with 5 patients in 2013. In this version, the system

only offered a Daily Mood Check questionnaire.

The results of this first pilot (see Table 3)

showed a 100% of system usage and treatment

adherence in 4 patients, and 71% in one patient. The

results related with the treatment adherence are the

same than the results of system usage due that only

one task was possible to be used. If the patient

accessed the system, he performed the activity. If

not, he could not do the task and he could not get it

back another day. In this first pilot it was not

possible to clearly measure the patient engagement

due to the same reason that only one task was

proposed every day, so the patient was not offered

with the selection of the session’s length.

Table 3: Assessment of the patient system usage (%) and

adherence to the treatment (%) in pilot 1.

P1 P2 P3 P4 P5 AVG

System

usage

1 0.71 1 1 1 0.942

Treatment

Adherence

1 0.71 1 1 1 0.942

The second pilot was carried out in Cluj-Napoca

during 15 days with 4 patients in 2013. This version

was more complete. The system could plan 5

different tasks and allow the user to select the length

of the sessions. The results for the second pilot are

showed in the following tables:

Table 4: Assessment of the patient system usage (%) and

adherence to the treatment (%) in pilot 2.

P1 P2 P3 P4 AVG

System usage 1 0.6 1 0.87 0.87

Treatment

Adherence

0.77 0.51 0.88 0.72 0.72

In the second pilot, the treatment adherence is

lower than in pilot 1. Patients were requested to

perform a minimum of 28 activities in 15 days and a

maximum of 54 (optimal result that we take as a

reference to calculate the treatment adherence).

Additionally, patients could select the length of the

session that allows us to analyse the patient

engagement to the system (see Table 5). Patient P2

had a low system usage (60%), so adherence was

also low (35%). The patients P1 and P3 got a 100%

use of the system. Regarding adherence, they

presented different results. This is due to the

selection of each patient in the length of the

conducted sessions. P1 selected mainly short

sessions, while P3 selected long sessions. Similarly,

P4 has less system usage than P1 but has almost the

same treatment adherence. That is because P4 has

made many more long sessions than P1, so P4 was

able to complete more tasks.

Table 5: Assessment of the patient engagement (%) in

pilot 2, based on type session selection (long, medium, or

short).

P1 P2 P3 P4 AVG

Long

selection

0.13 0.625 0.7 0.45 0.48

Medium

selection

0.2 0.125 0.2 0.0 0.13

Short

selection

0.66 0.25 0.1 0.54 0.39

The results obtained from these two pilots inform

us about the trends of the use of the system, the

treatment adherence, and patient engagement. The

low number of patients evaluated has been a

limitation of the study. In any case we started to

analyse the correlation between system usage,

treatment adherence and patient engagement. We

expect that the patient engagement can be improved

in order to obtain better results than the results

presented in Table 5 through the new contributions

described in this document. This improvement could

ICAART2014-DoctoralConsortium

affect in the treatment adherence even when the

system usage remains low.

Regarding the exit interview of the patient with

the clinician, the results obtained showed different

results. Some patients thought that the system plans

too long sessions, other described that sessions are

appropriate and others that there are too short

sessions.

Additionally, the patients of the pilot 2 were

asked about how much empathic was the virtual

agent. Two patients thought that the VA was

empathic and correct. In contrast, the third patient

answered that sometimes the VA was not really

empathic.

6 EXPECTED OUTCOME

The new contributions described in this document

(extension of the Session Planner and Cognitive-

Emotional module) will be implemented and

evaluated in the final pilot. We expect to obtain

more data that allow us to asset better conclusions

about the patient engagement and adherence

regarding our developed system.

In the long run, future pilots should include

patients control groups in order to compare the

obtained results between those patients that used the

system and those patients with the treatment that

does not include the use of the system. This would

help to conclude whether the use of our system

influenced and improved the adherence to the

treatment.

ACKNOWLEDGEMENTS

This paper reflects only the author’s views. The

European Community is not liable for any use that

may be made of the information contained herein.

This research is carried out within the EU FP7

Project “Help4Mood – A Computational

Distributed System to Support the Treatment of

Patients with Major Depression” [FP7-ICT-2009-

4; 248765].

REFERENCES

Andrist S., Leite I., Lehman J. 2013. Fun and fair:

influencing turn-taking in a multi-party game with a

virtual agent. In Proceedings of the 12th International

Conference on Interaction Design and Children

(IDC'13) pp., 352-355, New York.

Bernardini S., Porayska-Pomsta K., Sampath H. 2013.

Designing an Intelligent Virtual Agent for Social

Communication in Autism. Proceedings of the Ninth

AAAI Conference on Artificial Intelligence and

Interactive Digital Entertainment. pp. 9-15.

Bickmore T., Gruber A., Intille S., Mauer D. 2006. A

Handheld Animated Advisor for Physical Activity

Promotion. In American Medical Informatics

Association Annual Symposium, Washington, DC.

Bickmore T., Pfeifer L. 2008. Relational Agents for

Antipsychotic Medication Adherence. CHI'08

workshop on Technology in Mental Health Florence,

Italy.

Bresó, A. 2013. MSc Thesis: Generic Data Processing &

Analysis Architecture of a Personal Health System to

Manage Daily Interactive Sessions in Patients with

Major Depression. http://personales.upv.es/

adbregua/docs/A_Breso.pdf.

Brinkman W. P., van der Mast C. A. P. G., de Vliegher D.

2008. Virtual Reality Exposure Therapyfor Social

Phobia: A Pilot Study in Evoking Fear in a Virtual

World. In Proceedings of HCI2008 Workshop – HCI

for Technology Enhanced Learning. Liverpool, UK.

Castellano G., Paiva A., Kappas A., Aylett R., Hastie H.,

Barendregt W., Nabais F., Bull S. 2013. Towards

Empathic Virtual and Robotic Tutors. In Artificial

Intelligence in Education Lecture Notes in Computer

Science Volume 7926, pp 733-736.

Clark, A. J. 2007: Empathy in Counseling and

Psychotherapy. Perspectives and Practices. Lawrence

Erlbaum Associates.

Coombs, T., Deane, F. P., Lambert, G., & Griffiths, R.

2003. What influences patients’ medication

adherence? Mental health nurse perspectives and a

need for education and training. International journal

of mental health nursing, 12(2), 148-152.

Core M., Traum D., Lane H. C., Swartout W., Gratch J.,

van Lent M., Marsella S. 2006. Teaching Negotiation

Skills through Practice and Reflection with Virtual

Humans. Simulation Vol. 82(11), pp. 685–701.

Delamater A, 2006. Improving Patient Adherence.

Clinical Diabetes, vol. 24 no. 2 71-77.

Dias, J, Mascarenhas, S, & Paiva, A. 2011. Fatima

modular: Towards an agent architecture with a generic

appraisal framework. In Proceedings of the

International Workshop on Standards for Emotion

Modeling.

DiMatteo, M. R., Sherbourne, C. D., Hays, R. D., Ordway,

L., Kravitz, R. L., McGlynn, E. A., ... & Rogers, W.

H. 1993. Physicians' characteristics influence patients'

adherence to medical treatment: results from the

Medical Outcomes Study. Health psychology, 12(2),

93.

Dolder C. R., Lacro J. P., Leckband S., Jeste D. V., 2003.

Interventions to improve antipsychotic medication

adherence: review of recent literature. J Clin

Psychopharmacol; 23:389-99.

LeveragingAdaptiveSessionsBasedonTherapeuticEmpathyThroughaVirtualAgent

Donovan J. L., 1997. El paciente como decisor.

Utilización racional de medicamentos. International

Journal of Technology Assessment in Health Care (2ª

parte). Series divulgativas, documento nº 2, Sociedad

Española de Farmacéuticos de Atención Primaria

(SEFAP).

Gebhard P., Schröder M., Charfuelan M., Endres C., Kipp

M., Pammi S., Rumpler M., Türk O. 2008.

IDEAS4Games: Building Expressive Virtual

Characters for Computer Games. Proceedings of th 8th

International Conference on Intelligent Virutal

Agents, H. Prendinger, J. Lester and M. Ishizuka

(Eds.): LNAI 5208, pp. 426-440.

Gilden J, Staring A.B.P, Van der Gaag M, Mulder C.L,

2011. Does Treatment Adherence Therapy reduce

expense of healthcare use in patients with psychotic

disorders? Cost-minimization analysis in a randomized

controlled trial. Schizophrenia Research, vol. 133, no.

1–3, pp. 47–53.

Gross, J. J., & Thompson, R. A. 2007. Emotion regulation:

Conceptual foundations. Handbook of emotion

regulation, 3, 24.

Katon, W., Robinson, P., Von Korff, M., Lin, E., Bush, T.,

Ludman, E., & Walker, E. 1996. A multifaceted

intervention to improve treatment of depression in

primary care. Archives of General Psychiatry, 53(10),

924.

Kenny P., Parsons T., Gratch J., Leuski A., Rizzo A. 2007.

Virtual Patients for Clinical Therapist Skills Training.

7th International Conference on Intelligent Virtual

Agents, pp. 197-210, Paris France.

Krijn M., Emmelkamp P. M. G., Olafsson R. P., Biemond

R. 2004. Virtual reality exposure therapy of anxiety

disorders: A review. Clinical Psychology Review, Vol.

24(3) pp.259-281.

Lazarus, R. S. 1991. Emotion and adaptation (p. 557).

New York: Oxford University Press.

Lin, E. H., Von Korff, M., Ciechanowski, P., Peterson, D.,

Ludman, E. J., Rutter, C. M., & Katon, W. J. 2012.

Treatment adjustment and medication adherence for

complex patients with diabetes, heart disease, and

depression: a randomized controlled trial. The Annals

of Family Medicine, 10(1), 6-14.

Lincoln TM, Wilhelm K., Nestoriuc Y. 2007.

Effectiveness of psychoeducation for relapse,

symptoms, knowledge, adherencie and functioning in

psychotic disorders: a meta-analysis. Schizophr Res;

96:2232-45.

Márquez E, Casado JJ, Márquez JJ. 2001. Estrategias para

mejorar el cumplimiento terapéutico. FMC; 8:558-73.

Martínez-Miranda J., Jung B., Payr S., Petta P. 2008. The

Intermediary Agent’s Brain: Supporting Learning to

Collaborate at the Inter-Personal Level. In

Proceedings of the 7th. Conference on Autonomous

Agents and Multiagent Systems AAMAS'08, Volume 3,

pp. 1277 – 1280 IFAAMAS.

Mateas M., Stern A. 2003. Façade: An experiment in

building a fully-realized interactive drama. In Gama

Developers Conference, Game Design Track.

Neil, A. L., Batterham, P., Christensen, H., Bennett, K., &

Griffiths, K. M. 2009. Predictors of adherence by

adolescents to a cognitive behavior therapy website in

school and community-based settings. Journal of

Medical Internet Research, 11(1).

Orueta R. 2005. Estrategias para mejorar la adherencia

terapéutica en patologías crónicas. Inf Ter Sist Nac

Salud; 29(2):40-8.

Osterberg L, Blaschke T. Adherence to medication. N

Engl J Med 2005; 353(5):487-97.

Pampallona, S., Bollini, P., Tibaldi, G., Kupelnick, B., &

Munizza, C. 2002. Patient adherence in the treatment

of depression. The British Journal of Psychiatry,

180(2), 104-109.

Pontier M., Siddiqui G. A. 2008. Virtual Therapist That

Responds Empathically to Your Answers. Intelligent

Virtual Agents, pp. 417-425.

Raij A., Johnsen K., Dickerson R., Lok B., Cohen M.,

Duerson M., Pauly R., Stevens A., Wagner P., Lind D.

2007. Comparing Interpersonal Interactions with a

Virtual Human to Those with a Real Human.

Visualization and Computer Graphics, IEEE

Transactions 13(3), pp. 443-457.

Rigueira AI. 2001. Cumplimiento terapéutico: ¿qué

conocemos de España? Aten Primaria; 27(8):559-68.

Rizzo, Albert; Joseph, Brennan J. 2011. Psychology

Today. Virtual Reality, A New Therapy Part Two: The

SimCoach.

Sabatâe, E. 2003. Adherence to long-term therapies:

evidence for action. World Health Organization.

Safford, S. M., & Worthington, J. E. 1999. Computer

anxiety in individuals with serious mental illness.

Computers in human behavior, 15(6), 735-745.

Safren, S. A., O’Cleirigh, C. M., Bullis, J. R., Otto, M. W.,

Stein, M. D., & Pollack, M. H. 2012. Cognitive

Behavioral Therapy for Adherence and Depression

(CBT-AD) in HIV-Infected Injection Drug Users: A

Randomized Controlled Trial. Journal of Consulting

and Clinical Psychology, 80(3), 404-415.

K. R. Scherer, A. Schorr, and T. Johnstone (Eds.).

Appraisal Processes in Emotion: Theory, Methods,

Research. Oxford University Press, 2001.

Sotoca-Momblona J. M., Codina-Jané C. 2006. Cómo

mejorar la adherencia al tratamiento. JANO; 1605:39-

40.

Staring A. B. P., Mulder C. L., Van der Gaag M., Selten J.

P., Loonen A. J. M., Hengeveld M. W. 2006.

Understanding and improving treatment adherence in

patients with psychotic disorders; a review and a

proposed intervention. Curr Psychiatry Rev; 2: 487–

94.

Staring ABP, Van der Gaag M, Koopmans G. T, Selten J.

P, Van Beveren J. M, M. W. Hengeveld M. W,

Loonen A. J. M, Mulder C. L. 2010. Treatment

adherence therapy in people with psychotic disorders:

randomised controlled trial.BJP, 197:448-455.

Tartaro A., Cassell J. 2008. Playing with Virtual Peers:

Bootstrapping Contingent Discourse in Children with

Autism. Proceedings of International Conference of

the Learning Sciences (ICLS), Utrecht, Netherlands.

ICAART2014-DoctoralConsortium

Thwaites, R., Bennett-Levy, J.: Conceptualizing Empathy

in Cognitive Behaviour Therapy: Making the Implicit

Explicit. Behavioural and Cognitive Psychotherapy,

35, pp. 291-612 (2007).

Thompson, L., & McCabe, R. 2012. The effect of

clinician-patient alliance and communication on

treatment adherence in mental health care: a

systematic review. BMC psychiatry, 12(1), 87.

Titov, N., Dear, B. F., Johnston, L., Lorian, C., Zou, J.,

Wootton, B. & Rapee, R. M. 2013. Improving

adherence and clinical outcomes in self-guided

internet treatment for anxiety and depression:

randomised controlled trial. PloS one, 8(7), e62873.

Van Dulmen, S., Sluijs, E., van Dijk, L., de Ridder, D.,

Heerdink, R., & Bensing, J. 2007. Patient adherence to

medical treatment: a review of reviews. BMC health

services research, 7(1), 55.

Vergouwen, A. C., Bakker, A., Katon, W. J., Verheij, T.

J., & Koerselman, F. 2003. A systematic review of

interventions. Adequate follow-up can’t be optional,

64, 21.

Vittengl J. R., Clark L. A., Dunn T. W. Jarrett R. B. 2007.

Reducing relapse and recurrence in unipolar

depression: A comparative meta-analysis of cognitive

behavioral therapy's effect. J Consult Clin Psychol 75:

475-488.

WHO, 2012. Global burden of mental disorders and the

need for a comprehensive, coordinated response from

health and social sectors at the country level.

http://apps.who.int/gb/ebwha/pdf_files/EB130/B130_9

-en.pdf.

Yasavur U., Lisetti, C., Rishe N. 2013. Modeling Brief

Alcohol Intervention Dialogue with MDPs for

Delivery by ECAs. Intelligent Virtual Agents. Lecture

Notes in Computer Science Volume 8108, pp 92-105.

Yildiz, M., Veznedaroglu, B., Eryavuz, A., & Kayahan, B.

2004. Psychosocial skills training on social

functioning and quality of life in the treatment of

schizophrenia: A controlled study in Turkey.

International journal of psychiatry in clinical practice,

8(4), 219-225.

Zygmunt A., Olfson M., Boyer C. A., Mechanic D., 2002.

Interventions to improve medication adherence in

schizophrenia. Am J Psychiatry; 159:1653-64.

LeveragingAdaptiveSessionsBasedonTherapeuticEmpathyThroughaVirtualAgent