Diapositiva 1

1
Personality Diagnosis for Personalized eHealth
Services Marco Nalin, Angelica Morandi, Alberto
Sanna, eServices for Life and Health San
Raffaele Foundation Milan Fabio Cortellese,
Floriana Grasso Department of Computer Science,
University of Liverpool, UK
2
Context physical activity in prevention

• Research has shown that modifiable behaviors,
  including specific aspects of diet, overweight,
  inactivity, and smoking, account for over 70 of
  stroke and colon cancer, over 80 of coronary
  heart disease, and over 90 of adult onset
  diabetes.
• For diabetes in particular, intervention trials
  recently showed that a correct diet in
  combination with exercise programs can reduce the
  risk of developing diabetes by 60 in subjects
  with impaired glucose tolerance.
• Also, in the last two decades attention has
  shifted towards the quality of life in living
  with diabetes, especially in terms of physical
  activity.

3
PIPS (6th framework)
Personalised information platform for life and
health services FP6 Contract n. 507019,
January 2004 - June 2008 (continuing now in
clinical trial phase within San Raffaele
Hospital, Milan)
Investigation on the use of an eHealth platform,
combined with motivational tools, for health
promotion and life style targets A number of
scenarios identifying citizens needs, barriers,
and how technology can help
4
Diabetes and Physical Activity

• Motivation Only a minor proportion (20-25) of
  adults with type 2 Diabetes Mellitus exercise for
  at least 60 minutes per week

Regular physical exercise requires more time and
effort than dietary modifications and taking
medications, and diabetic patients often perceive
it as a significant and difficult change in their
lifestyle.

• Strolling and Motivation scenario
• Encouraging people to keep a daily regime of
  exercise
• Monitor the achievement of the daily target via
  a pedometer connected to the mobile phone
• Daily summary of activities, with a final
  personalised motivational message, sent to the
  mobile phone

5
Preferences/ Habits QuestionnaireExercise plan
6
Stage of Behaviour Change Questionnaire
Match counselling techniques to individual
motivational state ad hoc messages/ information
for each stage of behaviour change

• It was used for
• Motivation toward Physical Activity
• Attitude toward use of Technology

7
Exercise Target

• Metabolic Improvement patient safety
• Self-efficacy exercise target agreement
• Target adaptation to success/difficulties
• Target adaptation to health status/ physical
  parameters

8
Pedometer and Mobile phone
Technology Pedometer ? 3 axes
accelerometer Mobile ? Bluetooth from pedometer
? GPRS to server

• Monitoring and Motivation
• Compliance check
• Reciprocal interaction mechanism
  (individual-system)

9
Diary
Working day
Love/family
Weather
Mood
Correlation between internal/external factors and
physical activity level Self-Knowledge ?
Self-Efficacy
10
Feedback

• Results Visualization tables, charts, messages
• Uncompliance alert no data or poor target
  achievement
• Express walking difficulties suggestions and
  recalls

11
More on the Messages
Message of the Day Recover Message Failure
Message Friday Message Sunday Message Month
Report Messages per Year

• - 365
• - 0365
• - 0365
• - 52
• - 52
• - 12
• 4811211

In average more than 800 messages per year per
patient
12
Message composition

• Based on the analysis of a corpus of messages
• Tagged for evidence of SoC tailoring,
  argumentative/motivational messages, reference to
  personal factors etc.
• Ontology of messages message macrostructure
• A comment, aimed at providing a feedback on the
  users performance.
• An argumentation, aimed at supporting the final
  thesis that each message is set to maintain (that
  the physical activity is good for the user).
• An aid to introspection, aimed at relating the
  users performance with the diary values.
• A suggestion, aimed at providing both general and
  practical indication on how the users can improve
  their performance.
• An encouragement, concluding the communication.

13
Message Generation as ontology querying
Discourse ontology, an instance
14
Example Message of the Day
Mary is a Diabetic person in Likelihood of Change
her exercise habits. PIPS Communicative Goal
provide encouragement and knowledge on how to
increase self efficacy

• Exercise Plan
• User preferences for interaction with system
• Perceived walking Obstacles

15
Example Message of the Day
TECHNICAL INFORMATION
Today you have reached 45 of the target.
16
Example Message of the Day

• Stage of Behaviour Change
• Walking performance level

COMMENT

• Program progress

Today you have reached 45 of the target.
Not there yet, but you are on the right way!
Remember youve great potentialities.
17
Example Message of the Day

• Walking performance level

OBSERVATION

• Emotional Status

Today you have reached 45 of the target Not
there yet, but you are on the right way! Remember
youve great potentialities.
Health and mood are interdependent. Its worth
trying to do better.
18
Example Message of the Day

• Preferences and Habits
• Reason for Failure
• Perceived Walking Obstacles
• Location

SUGGESTION
Today you have reached 45 of the target Not
there yet, but you are on the right way! Remember
youve great potentialities. Health and mood are
interdependent. Its worth trying to better.
Having a dog is an asset for you. Walk her
frequently she will help you walk briskly. Give
it a try!
19
System Validation User involvement

• Experts evaluation (diabetologists,
  psychologists, )
• Interviews in Diabetes Unit ? 50 patients
• System demonstration focus groups ? 10 patients
• Pilot Study 3 months use ? 10 people (5 internal
  to the development team and 5 external)
• Randomized Clinical Trial (Control vs
  Intervention group) 12 months study ? 60 patients

20
San Raffaele Clinical Trial

• Control Group
• PIPS Pedometer and Mobile Phone (monitor)
• Standard care diet -exercise

• PIPS Group
• PIPS Pedometer and Mobile Phone
• (monitor and motivation support)
• Personalized Walking Target Path (steps/min
• and total minutes)
• Information and Motivation Feedback

21
San Raffaele Clinical Trial

• The primary objective of the study is to
  demonstrate the effectiveness of PIPS, which
  integrates a technological platform and a
  personalized motivation strategy, to achieve a
  personalized exercise target and to improve
  patient compliance.
• As secondary outcomes is considered the effect
  of physical activity on patients clinical and
  metabolic status (BMI, blood pressure, abdominal
  circumference, HbA1c, glycaemia, cholesterol,
  triglycerides, adiponectin)

22
First results

• The Motivation Strategy is currently under a
  formal evaluation
• 30 patients are in contemplation stage, 20 are
  in preparation, 20 are in action, and 30 are in
  maintenance
• For technology use, 20 is unaware of e-health
  potentialities, 30 find technology difficult to
  use, 10 have a positive attitude and 40 are in
  maintenance stage.
• First results report
• An improvement in the metabolic profile of the
  patients after 3 months of physical activity
  supported by PIPS strategy
• Positive feedback from users, claiming that
• By having a tool which monitors and provides
  feedback, they are more willing to walk
• They find it useful to receive feedback on their
  achievements.
• Messages were relevant to the actual target
  achievement
• Negative feedback
• Sometimes patients felt that messages were
  repetitive, so they tended to read them with less
  interest as the study progressed.

23
Personalization
Personalized Motivation Strategy
24
Revisiting the Personalization Strategy

• Static Clustering pre-determine clusters of
  populations, with a specific strategy targeting
  each cluster
• Dynamic Clustering strategy inspired by
  techniques used in Recommendation Systems
• Hybrid a combination of the two ? currently
  experimenting on Recommendation Systems coupled
  with Motivation Strategy

25
Dynamic Approach Recommender Systems
System predicts users preferences on the basis
of information acquired on the community of the
systems users.
Rating
Item
User
Hypothesis diabetic patients are the users,
while the set of items is the set of all the
possible motivational messages. The Rating is
given by a Success function.
Success
Patients
Messages
26
Determining Success factor
Day (N)
Day (N1)
Walking results
Diary
Walking results
Diary
Generation
Evaluation of Message success
Message of the Day
27
The success function

• A first approximation of the success function to
  be used as rating by the collaborative
  filtering algorithm
• PIPSImpact Target - R susc(p,m) f1
  fn
• Target is the percentage of achievement of the
  target
• R is a repetitiveness factor between zero (if the
  message has not been proposed for the past 3
  days) and 5 (addressing the criticism that a
  repeated message is less affective)
• susc(p,m) is a personal patient value which
  accounts for the greater or lesser susceptibility
  of the patient to that message, it derives from
  the initial questionnaire and it increases or
  decreases if that message influenced more or less
  the patient. It is not more than 20 of Target
  value
• f1 fn are the influence factor of some
  variables from the diary. They are not more than
  30.

28
Possible hybrid solutions

• Start off with Static Clustering then switch to
  Dynamic Clustering
• Pro
• Reduced cold start of the system
• Dynamic adaptation in long terms with patient
  behaviour
• Cons
• No control over the coherence of the message sent
  with the actual results and context
• Use the Static Clustering to shortlist a set of
  messages, then use Dynamic Clustering to predict
  the most likely to have impact
• Pro
• Messages sent to the user are in line with the
  Motivation Strategy designed by psychologist
• Messages are selected based also on the patients
  behaviour
• Cons
• Potentialities of the Recommendation System not
  fully exploited

29
Conclusions

• The Motivation Strategy is under formal
  evaluation. Preliminary results shows it is an
  effective personalized eHealth service to support
  patients in improving their physical activity,
  and their clinical conditions.
• A formal evaluation of Recommendation Systems
  approach is harder due to the high number of
  patients/messages needed to give significant
  results.
• The proposed hybrid solution is meant to combine
  the advantages of the static and dynamic
  approaches.
• Possible future developments can benefit from the
  Clinical Trial data as a training data set to be
  used by learning algorithms.

30
Thanks for your attention!!
For more technical questions Marco
Nalin e-Services for Life and Health nalin.marco_at_h
sr.it
Special issue of the UMUAI journal on
Personalisation for eHealth