Managing Data in Help4Mood

Maria K. Wolters

, Juan Mart´ınez-Miranda

, Helen F. Hastie

and Colin Matheson

School of Informatics, University of Edinburgh, Edinburgh, U.K.

IBIME, Universitat Polit´ecnica de Valencia, Valencia, Spain

School of Mathematical and Computer Sciences, Heriot-Watt University, Edinburgh, U.K.

Abstract. Help4Mood is a system that supports the treatment of people with ma-

jor depression through collecting a wealth of cognitive, psychomotor, and motor

data, which can then be summarised and analysed further. Data is stored in func-

tional units that correspond to treatment relevant entities using a custom XML

DTD. As far as possible, observations and ﬁndings are coded using SNOMED

CT to ensure interoperability with other applications such as Electronic Health

Records.

1 Help4Mood—Avatar-Based Support for People with Major

Depression

Depression is the main cause of disability worldwide [1]. It is characterised by a persis-

tent and intense change of mood which affects behaviour, cognition, and physiology.

There are various forms of depression [2]. Here, we focus on major unipolar de-

pression, which is mostly treated in the community. The core symptoms are persistent

low mood and loss of interest. Figure 1 summarises the deﬁnitive diagnostic criteria for

an episode of major depression, which includes activity and sleep symptoms.

As the DSM-IV deﬁnition suggests, depression also greatly affects psychomotor

function [3,4]. Two types of major depression can be distinguished, a melancholic form

where patients’ movementsare signiﬁcantly slowed down, and a non-melancholicform,

where movements are not affected or agitated. Slowed movements are reﬂected in both

gross motor function, such as gait, and ﬁne motor function, such as movement initiation

and reaction times. They also contribute to a reduced speech rate and a ﬂat intonation

[5,6].

At the moment, recovery is monitored infrequently through self-reported patient

questionnaires that require the person with depression to remember over a period of

time that can be as long as two weeks (e.g., PHQ-9 [7]). Those self-reports can be

unreliable, especially if the patient is not keeping regular notes or a diary.

The Help4Mood system is intended to support the treatment of people with ma-

jor unipolar depression in the community. In addition to monitoring through question-

naires and sensors, the patient interacts with Help4Mood through an avatar interface.

Help4Mood consists of three components, a personal monitoring system, a virtual

agent, which implements the avatar interface, and a decision support system.

K. Wolters M., Martínez-Miranda J., F. Hastie H. and Matheson C..

Managing Data in Help4Mood.

DOI: 10.5220/0003879400170024

In Proceedings of the 2nd International Workshop on Computing Paradigms for Mental Health (MindCare-2012), pages 17-24

ISBN: 978-989-8425-92-8

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

A. Five (or more) of the following symptoms have been present during the same

2-week period and represent a change from previous functioning; at least one

of the symptoms is either (1) depressed mood or (2) loss of interest or pleasure:

(1) depressed mood most of the day, nearly every day, as indicated by either subjec-

tive report (e.g., feels sad or empty) or observation made by others (e.g., appears

tearful).

(2) markedly diminished interest or pleasure in all, or almost all, activities most of

the day, nearly every day (as indicated by either subjective account or observa-

tion made by others)

(3) signiﬁcant weight loss when not dieting or weight gain (e.g., a change of more

than 5% of body weight in a month), or decrease or increase in appetite nearly

every day.

(4) insomnia or hypersomnia nearly every day

(5) psychomotor agitation or retardation nearly every day (observable by others,

not merely subjective feelings of restlessness or being slowed down)

(6) fatigue or loss of energy nearly every day

(7) feelings of worthlessness or excessive or inappropriate guilt (which may be

delusional) nearly every day (not merely self-reproach or guilt about being sick)

(8) diminished ability to think or concentrate, or indecisiveness, nearly every day

(either by subjective account or as observed by others)

(9) recurrent thoughts of death (not just fear of dying), recurrent suicidal ideation

without a speciﬁc plan, or a suicide attempt or a speciﬁc plan for committing

suicide

B. The symptoms do not meet criteria for a Mixed Episode.

C. The symptoms cause clinically signiﬁcant distress or impairment in social, oc-

cupational, or other important areas of functioning.

D. The symptoms are not due to the direct physiological effects of a substance

(e.g., a drug of abuse, a medication) or a general medical condition (e.g., hy-

pothyroidism).

E. The symptoms are not better accounted for by Bereavement, i.e., after the loss

of a loved one, the symptoms persist for longer than two months or are char-

acterised by marked functional impairment, morbid preoccupation with worth-

lessness, suicidal ideation, psychotic symptoms, or psychomotor retardation.

Fig.1. DSM-IV criteria for major depressive episode.

The sensors of the personal monitoring system assess sleep and activity patterns us-

ing sleep sensors and a wrist actigraph. The virtual agent asks questions, sets tasks, and

summarises the results of each session. Some of these tasks will yield cognitive data,

such as relevant negative automatic thoughts, others are designed to capture relevant

neuropsychomotor symptoms of depression, such as speech changes and slowed reac-

tion times [3]. The decision support system plans and controls sessions with the virtual

agent and converts data about the patient’s sleep, motor, speech, and other psychomotor

patterns into graphical, textual, and conceptual summaries that can be communicated to

clinicians, patients, and electronic health records.

In this paper, we describe our approach to data management in Help4Mood. We

focus on the high-level data structures that form the basis for communicating with clin-

icians, patients, and other stakeholders; the personal monitoring system and the virtual

agent have internal structures for storing the ﬁne-grained, detailed data which is anal-

ysed by the decision support system. The basic elements of the high-level Help4Mood

data structures are described in Section 2. Provisions for interoperability are outlined in

Section 4, and future work plans are summarised in Section 5.

2 Overview of Help4Mood

Help4Mood is structured around patients’ sessions with the Virtual Agent. Ideally, pa-

tients interact with their Virtual Agent daily. While content and length of a session

can be varied depending on the patient’s current mood and stamina, a default session

consists of ﬁve parts: welcome and daily mood check, diary, documenting negative

thoughts, speech task / game, summary feedback and closing.

The daily mood check is a validated four-item questionnaire, the CES-VAS-VA

[8]. In the diary entries, patients reﬂect on a speciﬁc prompt. These entries are stored

internally in the VA; patients will be able to discard them or save them for rereading.

All that is stored for the DSS is their length. Next, patients document negative thoughts

relating to this diary entry, and Help4Mood provides guidance for challenging these

thoughts. Then, patients perform a speech task or a cognitive game. Finally, the session

is brieﬂy summarised in a ﬁnal screen, and the virtual agent bids the patient goodbye.

While sleep data is collected every night, the wrist actigraph will only be worn for

72 hours at a time. Sessions with the virtual agent can include summaries of activity and

sleep patterns. In addition to the daily mood check, patients ﬁll in a formal screening

questionnaire, the PHQ-9 [7], every fortnight; this task is added to the session at the

appropriate time.

3 Core Data Structures

All data structures are described using XML. We chose this solution over a relational

database, because Help4Mood has a highly modular architecture, and almost all inter-

and intra-module communication is based on XML. Elements are extensively cross-

indexed to ensure ﬂexible access to data.

The XML elements that are used to store relevant data are summarised in Table 1.

They fall into four main categories, high-level tracking of patient and Help4Mood use,

storing monitoring results, managing the interaction between patient and virtual agent,

and storing the data collected during the interaction. Each set of elements is brieﬂy

explained below.

High Level Tracking. The three high-level tracking elements summarise relevant in-

formation about the patient and system usage and store the regular reports generated

by the system. Patient information includes basic demographics (occupation, gender,

age) as well as current depression scores. As for reports, only ofﬁcial reports that are

sent to the clinicians and can be discussed in patient/clinician meetings are stored. The

feedback given to the patient at the end of each session is not saved, because it can be

reconstructed deterministically from the data collected in each session.

Table 1. Basic Elements of the Help4Mood data structure. Each one is deﬁned using XML.

Element Description

High-Level Tracking of Patient and Help4Mood Usage

User Model high-level summary of information about the patient

Adherence adherence of patient to Help4Mood; can refer to sessions, tasks, and

monitoring schedules

Report summary report generated for clinician

Managing the Results of Monitoring

Monitor Data set of measures that are collected during a session

Measure high-level measure computed from a given set of monitoring data

Score score on a standardised questionnaire

Managing the Interaction with the Virtual Agent

Session content and results of a session with the Virtual Agent

Event event triggered by the decision support system during a session

Task task that is performed by the patient during a speciﬁc session

Emotion emotion used by virtual agent while patient performs task

Storing Information Collected During Interaction with the Virtual Agent

Diary information related to diary entries

Speech changes in relevant speech parameters

Games changes in reaction times and scores

Negative Thought frequency of speciﬁc negative automatic thoughts

Monitoring. The next two elements are used to describe high-level monitoring data.

While the measure element covers speciﬁc analysis results, the monitoringdata

element collects a set of measures obtained during a given session.

Managing the Interaction with the Virtual Agent. The Decision Support System con-

trols the Virtual Agent’s interaction with the user through events (event element).

Events are triggered when their preconditions are fulﬁlled. They are implemented as

interaction tasks (task element). Each task is associated with an emotion (emotion

element) that controls the affective behaviour of the Virtual Agent.

The sequence of of events and task/emotion pairs that occurred during a session and

the data that was generated during a session is stored in a session element for easy

reference.

Table 2 shows the structure of an event element. Each event is linked to a session,

a patient, and a speciﬁc time within the session. A range of auxiliary elements is used

to specify events. Descriptors link Events to a formal code that describes the under-

lying procedure and can be exported to external systems (c.f. Table 3). Preconditions

and postconditions are described using condition elements that consist of hProperty,

Operator, Valuei-tuples (c.f. Table 4).

Data Collected During Interaction. During most tasks, the system collects rich in-

formation about the patient’s cognition and current psychomotor functioning. Rele-

vant high-level data is encoded in the diary, speech, games, and negative

thought elements.

Table 2. The Event class.

Attribute Value Description

hTypei {1,2,3,4} Event type as classiﬁed by data source

hSessioni timestamp Session ID

hPatienti alphanumerical code Patient ID

hDescriptioni descriptor Formal description of the event

hGeneratedi timestamp time at which the event was generated

hPreconditionsi list of conditions pre-conditions that trigger the event.

hPostconditionsi list of conditions ﬁndings or observable entities

Table 3. Structure of Descriptors.

Attribute Values Deﬁnition

hCodei 9-digit integer Numerical identiﬁer. If the attribute is a concept taken form the

SNOMED-CT classiﬁcation, the code is its corresponding ID

hSnomedi Yes/No Yes if the attribute is a SNOMED-CT concept, otherwise no

hNamei String SNOMED-CT description if attribute is a SNOMED-CT con-

cept, otherwise internal description

Table 4. Structure of Conditions.

Attribute Value Description

hPropertyi Descriptor Property or action in the VA’s world that is tested in the precon-

dition.

hOperatori <, >, =, != A string that speciﬁes the operator used to determine the truth

value of the condition. The ﬁrst two operators are deﬁned for

numerical values, the last two are deﬁned for numerical and

strong values

hValuei descriptor Value used to compare the property to.

4 Communication and Interoperability

4.1 Clinician

Clinicians can access a patient’s history through a web interface to a special clinician-

side version of the decision support system, which generates textual and graphic sum-

maries to support treatmentplanning. In addition, Help4Moodgenerates a regular report

for each patient / clinician meeting. All relevant health care professionals as well as the

patient can see this report; clinician and patient may discuss speciﬁc ﬁndings when they

meet.

4.2 Patient

The patient receives textual and / or graphical summaries at the end of each session

that are generated dynamically. The Virtual Agent stores detailed information about

performance on speech tasks and games which can be exported for further reﬁnement

of measures. Patients can choose to save their diary entries for later perusal. The entries

themselves are not passed on to the clinician. Patients will be made aware that diary

entries are private every time they use the system. In addition to their diary entries,

patients have access to the shared reports, which are stored patient-side.

4.3 EHR Integration

As far as possible, we use the international Core Release of SNOMED CT [9] to de-

scribe our ﬁndings and observations. SNOMED CT is a highly complex, extendable

clinical vocabulary that can be integrated with standards such as HL7 [10], which

Help4Mood will support.

Most of the SNOMED-CT concepts used in Help4Mood come from the Clinical

Finding hierarchy. Clinical ﬁndings are the outcome of assessments, observations, or

judgements. For example, if the sleep sensor data indicate that the patient tossed and

turned frequently at night, this can be encoded as the Clinical Finding “restless sleep”.

Other concepts are Procedures, i.e., activities that occur at a speciﬁc time and in-

volve the patient. Procedures include education and administration activities. For ex-

ample, showing the patient a list of activities that were identiﬁed as comforting could

be a procedure. Another example of a procedure would be guiding the patient through

a relaxation exercise.

Most of the relevant information about a patient’s social context is modelled using

concepts from the Social Context hierarchy, such as occupation.

A question or a procedure that produces a result is an Observable Entity. For exam-

ple, “gender” is an observable entity, while “female gender” is a ﬁnding. Questionnaire

scores are modelled as observable entities; the associated ﬁnding is their interpretation.

The questionnaire instruments themselves are part of yet another hierarchy, Staging and

Scales. For example, the Beck Depression Inventory [11]

When extending Help4Mood to provide medication education, we may also include

concepts from the hiearchy Pharmaceutical/biological product, which correspond to

the type of medication being given.

We deﬁned our own codes only if the relevant ﬁndings, observable entities, or pro-

cedures were not included in SNOMED-CT. In all cases, these codes are linked to a

parent concept in SNOMED-CT. For example, unlike the Beck Depression Inventory,

neither of our depression measures are modelled explicitly in SNOMED-CT. There-

fore, we assigned the resulting scores system-speciﬁc codes and linked them to relevant

parent concepts. Hence, the PHQ-9 score is an Observable Entity which is in an is-a

relation with the SNOMED CT concept Mental state, behaviour / psychosocial function

observable.

If necessary, relevant data is mapped onto SNOMED CT categories by the Decision

Support System. While information such as questionnaire scores can be stored more or

less directly, concepts such as “restless sleep” will be derived from sensor data using

validated algorithms.

Table 5. Anchoring New Concepts in SNOMED-CT.

SNOMED CT

Type

Concept Description is-a relation to SNOMED-CT

Code Description

Procedure 999991011 Assessment using CES-D-

VAS-VS

445536008 Assessment using assess-

ment scale

Observable En-

tity

999991012 CES-D-VAS-VS score 363870007 Mental state, behavior /

psychosocial function ob-

servable (observable entity)

Procedure 999991021 Assessment using PHQ-9 445536008 Assessment using assess-

ment scale

Observable En-

tity

999991022 PHQ-9 score 363870007 Mental state, behavior /

psychosocial function ob-

servable (observable entity)

Finding 999992011 Change of Score on Cogni-

tive Game SIMON

248536006 Finding of functional per-

formance and activity

5 Future Work

The data management approach outlined here provides a detailed, systematic repre-

sentation of all of the relevant high-level information that Help4Mood collects about

a patient with major unipolar depression. It was designed for easy maintenance and

maximum interoperability with EHRs.

We are currently implementing the ﬁrst version of Help4Mood based on the data

structures outlined in this paper. In future versions, we plan to add HL7 integration,

reﬁne the elements and clinical vocabulary described here to provide a more detailed

ontology for interoperability, and extend Help4Mood with reminder functionality and

patient education through tailored health information presentation.

References

1. USDHHS, U.: Mental health: A report of the Surgeon General. Executive summary.

Rockville, MD (1999)

2. : DSM-IV-TR : diagnostic and statistical manual of mental disorders. 4th edn. American

Psychiatric Press (DC) (2000)

3. Schrijvers, D., Hulstijn, W., Sabbe, B.G.: Psychomotor symptoms in depression: a diagnos-

tic, pathophysiological and therapeutic tool. J Affect Disord 109 (2008) 1–20

4. Buyukdura, J.S., McClintock, S.M., Croarkin, P.E.: Psychomotor retardation in de-

pression: Biological underpinnings, measurement, and treatment. Progress in neuro-

psychopharmacology & biological psychiatry (2010)

5. Alpert, M., Pouget, E.R., Silva, R.R.: Reﬂections of depression in acoustic measures of the

patient’s speech. J Affect Disord 66 (2001) 59–69

6. Stassen, H.H., Kuny, S., Hell, D.: The speech analysis approach to determining onset of

improvement under antidepressants. Eur Neuropsychopharmacol 8 (1998) 303–310

7. Kroenke, K., Spitzer, R.L., Williams, J.B.: The PHQ-9: validity of a brief depression severity

measure. Journal of general internal medicine 16 (2001) 606–13

8. Moullec, G., Ma¨ıano, C., Morin, A.J.S., Monthuy-Blanc, J., Rosello, L., Ninot, G.: A very

short visual analog form of the Center for Epidemiologic Studies Depression Scale (CES-D)

for the idiographic measurement of depression. Journal of affective disorders (2010)

9. The International Health Terminology Standards Development Organisation: SNOMED

Clinical Terms User Guide. Technical Report July (2009)

10. Benson, T.: Principles of Health Interoperability - HL7 and SNOMED. Springer, New York,

NY (2009)

11. Beck, A.T., Ward, C.H., Mendelson, M., Mock, J., Erbaugh, J.: An inventory for measuring

depression. Archives of general psychiatry 4 (1961) 561–71