Bip4Cast: Some advances in
mood disorders data analysisP Llamocca1, D Urgelés2, M Cukic3, V Lopez4
1pavellam@ucm.es, 2diego.urgeles@sjd.es, 3cukic@3ega.nl, 4vlopezlo@ucm.es
1,4www.ucm.es, 2www.sjd.es, 3www.3ega.nl
Dec-2019
The research team:
Complutense University-GRASIA research
group (Madrid, Spain)
1 full professor
1 associate professor
2 PhD students (+)
Ntra. Sra. de la Paz, Hospital (Madrid, Spain)
1 psychiatrist doctor
25 supervised patients
3EGA .nl (Amsterdam, Nederland)
Dr. Cukic
Monitoring devices
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 2
Introduction to Bip4Cast
Bip4Cast: Some advances in mood
disorders data analysis
Index
1. Introduction to Bip4Cast
2. Hypothesis and goals
3. Information sources
4. Data integration
5. Data Analysis
6. Conclusions and Future work
Motivation:
• Mood disorders
• Depression
• Bipolar disorder
• Others
Bipolar disorder and depression are chronic
and severe mental disorders and a major
mental health problem in Europe.
The Regional Office for Europe of the World
Health Organization, in 2019, reports that
25% of the population suffer from depression
or anxiety.
They have a high suicide index.
Introduction to Bip4Cast
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 4
Introduction to Bip4Cast
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 5
Signs and Symptoms of Depression
• Feelings of uselessness, hopelessness,
excessive guilt
• Loss of interest in work, school,
hobbies, people
• Social isolation
• Agitation and irritability
• Low energy and lethargy
• Sad mood
• Changes in appetite or weight – eating
too little or too much
• Oversleeping or insomnia
• Suicidal thoughts
Source:
Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
Introduction to Bip4Cast
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 6
Signs and Symptoms of Bipolar
Disorder
Depression – the ‘lows’ of bipolar
disorder
• Feelings of uselessness,
hopelessness, excessive guilt
• Loss of interest in work, school,
hobbies, people
• Social isolation
• Agitation and irritability
• Low energy and lethargy
• Sad mood
• Changes in appetite or weight –
eating too little or too much
• Oversleeping or insomnia
• Suicidal thoughts
Source:
Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
Introduction to Bip4Cast
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 7
Signs and Symptoms of Bipolar Disorder
Mania – the ‘highs’ of bipolar disorder
• Elevated, expansive mood
• Extreme irritability
• Rapid, unpredictable emotional
changes
• Racing thoughts, flights of fancy
• Overspending
• Sense of grandiosity, inflated self-
esteem
• Decreased need for sleep
Source:
Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
Introduction to Bip4Cast
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 8
Sign and Symptoms of Postpartum
Depression
• Feeling restless or slowed down
• Feeling sad most of the day
• Loss of interest or pleasure in all or
most things
• No interest or pleasure in your baby
• Crying for no reason
• Excessive worrying about your baby
• Scary thoughts about harming your
baby
• Anxiety or panic attacks
• No desire to be with family or friends
• Suicidal thoughts or frequent
thoughts of death
Source:
Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
Introduction to Bip4Cast
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 9
Signs and Symptoms of Anxiety
• Excessive and uncontrollable
worry
• Irritability
• Sleep disturbances
• Fear of losing control
• Dizziness or light headedness
• Fear of being in public
Source:
Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
Hypothesis and goals
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 10
Traditional indicators like Hamilton and
Yang are insufficient to predict and avoid
crises. This study adds up to 149 new
variables from data gathered from
alternative sources.
The results show the existence of a
relationship between biological,
psychological, physical indicators with
the appearance of a crisis of depression
or mania.
These relations are the base of the
predictive analytics that clinicians need
in order to make better decisions on the
future treatment plans.
Deduction of the euthymic state by movement
patterns
Deduction of the euthymic state by sound patterns
PROFILE: Worker with family
DEVICE: WITHINGS Steel HR sport
Goal: 10.000 steps per day
Example 1
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 11
PROFILE: Retired old woman (80 years
old)
DEVICE: GARMIN VIVOFIT 3
Goal: 15.000 steps per day
Example 2
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 12
Information Sources
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 13
Interviews (Practician Report). Data observed
periodically by a psychiatrist in an interview
session. 38 variables including “crisis
indicator” , HDRS scale (Hamilton Depression
Rating Scale) and YMRS scale (Young Mania
Rating Scale)
Smartwatches/bands (Monitor). Each patient
wear a medical smart band that collects
automatically a total of 108 variables and
indicators (heart rate, physical activity, sleep
quality…)
Fill-in Forms (Self-Report). Daily Patients
complete an electronic form of 41 variables
(coffee, tobacco, alcohol consumption,
anxiety, concentration, mood, etc. )
Data integration
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 14
Building a single structure
This structure is also scalable for allowing
to store variables coming from new
sources (as voice recording, images and
video). Currently, there are 187 variables
extracted from sources already
incorporated.
Building an integration process.
Several tasks are performed: cleaning,
formatting, standardization,
interpolation, etc.
Group of Variables Type Nº of
Variables
Source Frequency
Patient Id CHAR 1 All Daily
Date DATE 1 All Daily
Fill-in Forms
Parameters
NUMERIC 39 Fill-in Forms Daily
Smartwatch
Parameters
NUMERIC 108 Smartwatches Daily
Interviews
Parameters
NUMERIC 37 Interviews Each 2 weeks
Crisis Indicator NUMERIC 1 Interviews Each 2 weeks
New Sources TBD TBD Voice, video TBD
Table 1. Integration Structure
Preliminar Data Analysis
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 15
PREVIOUS DIAGNOSYS AND TREATMENT
Patient’s data was individually analysed
Tradional index and the evolution of the
medical treatment is part of the study
Data about medication (lithium and
others) and diagnostic detection play still
a very important role.
R programming and packages for
regression analysis.
Data Analysis
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 16
In the subsequent analysis patient’s data
are also individually analyzed .
Pearson’s correlations help to find the
relation between pairs of daily-form
features
a. Visualization ggplot
b. R programming and packages.
The correlation coefficient some times
reaches 85%
Data Analysis
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 17
Correlations show the relation among
accelerometer variables.
Conclusions about:
- Steps: daily activity
- Sleep activity
The relation among all types of features:
Due to daily habits, coffee is not as
relevant as smoking.
Data Analysis
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 18
ABC balance: A+B+C=1
A- importance of the doctor report
indexes
B- importance of the accelerometer data
(objective)
C- importance of the self report data
(subjective component)
Rate example: (A,B,C)=(.3,.4.2)
FR final ratio
x, y, z are normalized selected variables
t, f, g are transformations
w, w’, w” are weight of each feature with
in the system
𝑖=1
𝑛1
𝑤𝑖 ∗ 𝑡(𝑥𝑖)
Doctor report index (RED)
𝑖=1
𝑛2
𝑤′𝑖 ∗ 𝑓(𝑦𝑖)
Accelerometer index (BLUE)
𝑖=1
𝑛3
𝑤"𝑖 ∗ 𝑔(𝑧𝑖)
Self Report index (GREEN)
FR= A* ( 𝑖=1
𝑛1
𝑤𝑖 ∗ 𝑡(𝑥𝑖) ) + B* ( 𝑖=1
𝑛2
𝑤′𝑖 ∗ 𝑓(𝑦𝑖) ) +C*( 𝑖=1
𝑛3
𝑤"𝑖 ∗ 𝑔(𝑧𝑖))
Conclusions & Future work
Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 19
The better information, the better
diagnosys and treatment.
Computer-Aid system /recommender
system to help practitians and patiens to
get useful information.
Design of a recommendation system for
medical diagnosis with fuzzy implications
(Nas – outliers tolerance).
Bayesian analysis with the new
incremental data.
Bip4Cast: Some advances in
mood disorders data analysisP Llamocca1, D Urgelés2, M Cukic3, V Lopez4
1pavellam@ucm.es, 2diego.urgeles@sjd.es, 3cukic@3ega.nl, 4vlopezlo@ucm.es
1,4www.ucm.es, 2www.sjd.es, 3www.3ega.nl
Dec-2019

Alan turing uva-presentationdec-2019

  • 1.
    Bip4Cast: Some advancesin mood disorders data analysisP Llamocca1, D Urgelés2, M Cukic3, V Lopez4 1pavellam@ucm.es, 2diego.urgeles@sjd.es, 3cukic@3ega.nl, 4vlopezlo@ucm.es 1,4www.ucm.es, 2www.sjd.es, 3www.3ega.nl Dec-2019
  • 2.
    The research team: ComplutenseUniversity-GRASIA research group (Madrid, Spain) 1 full professor 1 associate professor 2 PhD students (+) Ntra. Sra. de la Paz, Hospital (Madrid, Spain) 1 psychiatrist doctor 25 supervised patients 3EGA .nl (Amsterdam, Nederland) Dr. Cukic Monitoring devices Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 2 Introduction to Bip4Cast
  • 3.
    Bip4Cast: Some advancesin mood disorders data analysis Index 1. Introduction to Bip4Cast 2. Hypothesis and goals 3. Information sources 4. Data integration 5. Data Analysis 6. Conclusions and Future work
  • 4.
    Motivation: • Mood disorders •Depression • Bipolar disorder • Others Bipolar disorder and depression are chronic and severe mental disorders and a major mental health problem in Europe. The Regional Office for Europe of the World Health Organization, in 2019, reports that 25% of the population suffer from depression or anxiety. They have a high suicide index. Introduction to Bip4Cast Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 4
  • 5.
    Introduction to Bip4Cast SeminarSocial & Health Data Analytics, Victoria López, Dec. 2019 5 Signs and Symptoms of Depression • Feelings of uselessness, hopelessness, excessive guilt • Loss of interest in work, school, hobbies, people • Social isolation • Agitation and irritability • Low energy and lethargy • Sad mood • Changes in appetite or weight – eating too little or too much • Oversleeping or insomnia • Suicidal thoughts Source: Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
  • 6.
    Introduction to Bip4Cast SeminarSocial & Health Data Analytics, Victoria López, Dec. 2019 6 Signs and Symptoms of Bipolar Disorder Depression – the ‘lows’ of bipolar disorder • Feelings of uselessness, hopelessness, excessive guilt • Loss of interest in work, school, hobbies, people • Social isolation • Agitation and irritability • Low energy and lethargy • Sad mood • Changes in appetite or weight – eating too little or too much • Oversleeping or insomnia • Suicidal thoughts Source: Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
  • 7.
    Introduction to Bip4Cast SeminarSocial & Health Data Analytics, Victoria López, Dec. 2019 7 Signs and Symptoms of Bipolar Disorder Mania – the ‘highs’ of bipolar disorder • Elevated, expansive mood • Extreme irritability • Rapid, unpredictable emotional changes • Racing thoughts, flights of fancy • Overspending • Sense of grandiosity, inflated self- esteem • Decreased need for sleep Source: Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
  • 8.
    Introduction to Bip4Cast SeminarSocial & Health Data Analytics, Victoria López, Dec. 2019 8 Sign and Symptoms of Postpartum Depression • Feeling restless or slowed down • Feeling sad most of the day • Loss of interest or pleasure in all or most things • No interest or pleasure in your baby • Crying for no reason • Excessive worrying about your baby • Scary thoughts about harming your baby • Anxiety or panic attacks • No desire to be with family or friends • Suicidal thoughts or frequent thoughts of death Source: Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
  • 9.
    Introduction to Bip4Cast SeminarSocial & Health Data Analytics, Victoria López, Dec. 2019 9 Signs and Symptoms of Anxiety • Excessive and uncontrollable worry • Irritability • Sleep disturbances • Fear of losing control • Dizziness or light headedness • Fear of being in public Source: Mood Disorders Society of Canada (MDSC, https://mdsc.ca/)
  • 10.
    Hypothesis and goals SeminarSocial & Health Data Analytics, Victoria López, Dec. 2019 10 Traditional indicators like Hamilton and Yang are insufficient to predict and avoid crises. This study adds up to 149 new variables from data gathered from alternative sources. The results show the existence of a relationship between biological, psychological, physical indicators with the appearance of a crisis of depression or mania. These relations are the base of the predictive analytics that clinicians need in order to make better decisions on the future treatment plans. Deduction of the euthymic state by movement patterns Deduction of the euthymic state by sound patterns
  • 11.
    PROFILE: Worker withfamily DEVICE: WITHINGS Steel HR sport Goal: 10.000 steps per day Example 1 Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 11
  • 12.
    PROFILE: Retired oldwoman (80 years old) DEVICE: GARMIN VIVOFIT 3 Goal: 15.000 steps per day Example 2 Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 12
  • 13.
    Information Sources Seminar Social& Health Data Analytics, Victoria López, Dec. 2019 13 Interviews (Practician Report). Data observed periodically by a psychiatrist in an interview session. 38 variables including “crisis indicator” , HDRS scale (Hamilton Depression Rating Scale) and YMRS scale (Young Mania Rating Scale) Smartwatches/bands (Monitor). Each patient wear a medical smart band that collects automatically a total of 108 variables and indicators (heart rate, physical activity, sleep quality…) Fill-in Forms (Self-Report). Daily Patients complete an electronic form of 41 variables (coffee, tobacco, alcohol consumption, anxiety, concentration, mood, etc. )
  • 14.
    Data integration Seminar Social& Health Data Analytics, Victoria López, Dec. 2019 14 Building a single structure This structure is also scalable for allowing to store variables coming from new sources (as voice recording, images and video). Currently, there are 187 variables extracted from sources already incorporated. Building an integration process. Several tasks are performed: cleaning, formatting, standardization, interpolation, etc. Group of Variables Type Nº of Variables Source Frequency Patient Id CHAR 1 All Daily Date DATE 1 All Daily Fill-in Forms Parameters NUMERIC 39 Fill-in Forms Daily Smartwatch Parameters NUMERIC 108 Smartwatches Daily Interviews Parameters NUMERIC 37 Interviews Each 2 weeks Crisis Indicator NUMERIC 1 Interviews Each 2 weeks New Sources TBD TBD Voice, video TBD Table 1. Integration Structure
  • 15.
    Preliminar Data Analysis SeminarSocial & Health Data Analytics, Victoria López, Dec. 2019 15 PREVIOUS DIAGNOSYS AND TREATMENT Patient’s data was individually analysed Tradional index and the evolution of the medical treatment is part of the study Data about medication (lithium and others) and diagnostic detection play still a very important role. R programming and packages for regression analysis.
  • 16.
    Data Analysis Seminar Social& Health Data Analytics, Victoria López, Dec. 2019 16 In the subsequent analysis patient’s data are also individually analyzed . Pearson’s correlations help to find the relation between pairs of daily-form features a. Visualization ggplot b. R programming and packages. The correlation coefficient some times reaches 85%
  • 17.
    Data Analysis Seminar Social& Health Data Analytics, Victoria López, Dec. 2019 17 Correlations show the relation among accelerometer variables. Conclusions about: - Steps: daily activity - Sleep activity The relation among all types of features: Due to daily habits, coffee is not as relevant as smoking.
  • 18.
    Data Analysis Seminar Social& Health Data Analytics, Victoria López, Dec. 2019 18 ABC balance: A+B+C=1 A- importance of the doctor report indexes B- importance of the accelerometer data (objective) C- importance of the self report data (subjective component) Rate example: (A,B,C)=(.3,.4.2) FR final ratio x, y, z are normalized selected variables t, f, g are transformations w, w’, w” are weight of each feature with in the system 𝑖=1 𝑛1 𝑤𝑖 ∗ 𝑡(𝑥𝑖) Doctor report index (RED) 𝑖=1 𝑛2 𝑤′𝑖 ∗ 𝑓(𝑦𝑖) Accelerometer index (BLUE) 𝑖=1 𝑛3 𝑤"𝑖 ∗ 𝑔(𝑧𝑖) Self Report index (GREEN) FR= A* ( 𝑖=1 𝑛1 𝑤𝑖 ∗ 𝑡(𝑥𝑖) ) + B* ( 𝑖=1 𝑛2 𝑤′𝑖 ∗ 𝑓(𝑦𝑖) ) +C*( 𝑖=1 𝑛3 𝑤"𝑖 ∗ 𝑔(𝑧𝑖))
  • 19.
    Conclusions & Futurework Seminar Social & Health Data Analytics, Victoria López, Dec. 2019 19 The better information, the better diagnosys and treatment. Computer-Aid system /recommender system to help practitians and patiens to get useful information. Design of a recommendation system for medical diagnosis with fuzzy implications (Nas – outliers tolerance). Bayesian analysis with the new incremental data.
  • 20.
    Bip4Cast: Some advancesin mood disorders data analysisP Llamocca1, D Urgelés2, M Cukic3, V Lopez4 1pavellam@ucm.es, 2diego.urgeles@sjd.es, 3cukic@3ega.nl, 4vlopezlo@ucm.es 1,4www.ucm.es, 2www.sjd.es, 3www.3ega.nl Dec-2019

Editor's Notes

  • #2  Hello everyone. This presentation is about the project Bip4Cast where we work in analyzing data on mood disorders. The talk is Seminar on Social and Health Data Analytics, which is the second of a series in Amsterdam University and the 5th in the series of Spain + Amsterdam, December 2019
  • #3  We are abut 4-6 persons among professors, phd students doctors and experts. A hospital in Spain and a spinoff in Nederland is also involved.
  • #4  I am going to introduce the research, with definition, hypothesis and goals,, then i will speak about the data sources and its integration, and finally about the analysis conclusions and future work
  • #5  There are several mood disorders but we focus specially in depression and bipolar disorders. Both of them are chronic and a big mental health problem in Europe. According to the main health organizations, the 25% of the population suffer depression or anxiety which is a sign of a mood disorder. Also this patients have a very high suicide index.
  • #6  The first problem we find is how to detect the problem. Generally this persons lie about their feelings, stop with their medication and hide their situation. That’s why we study other ways of collecting this information with non intrusive methods. Analysing the symptoms we found many that can be collected with sensors like smartphones or smartbands. In depression for example low energy, oversleeping are very relevant.
  • #7  In the case of Bipolar disorder, we must distinguish between the lows and de highs, when the crisis is in depression, symptoms are very similar than in depression disorder.
  • #8  In the Mania stage, signs are expansive mood, extreme irritability, overspending, no need for sleeping, and so on. All these signs can be detected with monitoring the activity and behavior of the patiens in their daily life
  • #9  Also Other related disorders are also in the same way as postpartum depression or anxiety
  • #10  Then the study is much more interesting because we can cover more mood disorders
  • #11  Traditional indicators are indexes like Hamilton and Yang that because of the lack of information are not enough for making a prediction of the crisis. The idea is that a prediction of one or two weeks is time enougth to update the medication and avoid the crisis. So at first instance, we try to model the three stages of the patien, euthymia, mania and depression and use the data collected to guess the evolution of the distribution of the real stage.
  • #12  However we must face several problems. One of them is the personalization. Data about mood disorders is nothing to do one person to another. In this example, data comes from a smartwatch, the profile is a worker with family. Data reveals many different situations, maybe because the life of this person is in variety, bussiness man for example.
  • #13  This second example that shows a very well pattern of movement during the day corresponds to an 80 years old woman, who is making close to double steps tan the business man, but its daily routing is very clear.
  • #14  After these observations, we started colleting the following data sets: Traditional data, reported by the doctor in the interview with the patient, data collected automatically from the smartwatch, we used actigraph medical watches. And dayly self reports that patients fill about relevant cuantitative and cualitative information as number of cigarrettes or concentration
  • #15  All these data sets are combine in a single structure where the data is formatted, cleaned and normalized. We have a total of 187 variables of interest and next steps are about reducing the dimensonality if posible before the analytics.
  • #16  Preliminar analysis reveals the impact that the medication presents on the evolution of the patients. So the information about the medication is very relevant within the study (for the moment in a different repository for privacy). I want to point here that the datasets are anonimized for the study which means to have the permission for collecting data from the Spanish agency, for example.
  • #17  When we compare the varialbles from the self-report we find some interesting dependencies, for example, animus and motivation with a 85% of correlation.
  • #18  Some studies about the variables from the accelerometer reveals that the number of variables can be reduced a lot. We are interested specially in steps during the day (no aggregated) and sleepy activity, for example number of wake ups during the night is a very important variable. When we cross the variables of all cathegories we also find some interesting conclusions, for example, coffee is not representative at all (we think because is more a habit than a impulsive behaviour/necesity) On the contrary, smoking or drugs consumption are variables that change with the behaviour and the mood changes of the patients and then they are essential in the analytical process
  • #19  A first approach to the crisis evaluation can be made by weighting the selected variables within each group. A different aggregation is done according to the sense of the report: doctor report is closer to objective than self-report and accelerometer data is totally objective. Coefficients ABC are weithing that point. Each of these three studies return a measure of the eutimia as we can see in the graph. This graph corresponds to syntetic data to show the working way of the system. Purple bars are the confidence interval for the euthimic stage meanwhile red, blue and Green are the evolution of the three sources of information: doctor, accelerometer and patient. The black line is the final ratio. When the final ratio is below the lower purple line, indicates a depression. On the contrary if the black line lies on top of the high purple line, the sign is of a mania crisis.
  • #20  We conclude that the better information the better diagnosis and treatment. This comes with a good cleaning of data, a correct choice of the variables and features and a good construction of the indexes. A recommendation system for diagnosis is due. To make it better with missing data and outliers, we think in introducing a fuzzy set of inference rules. Also as future work and in order to face with new data, we study a bayesian analysis to update the distribution of the data.