Transcript New generation of Personal Health Systems enabling quality health data and information gathering and use Nicos Maglaveras Professor & Director Lab of Medical Informatics Aristotle University Thessaloniki, GREECE EMAIL.

New generation of Personal Health
Systems enabling quality health
data and information gathering
and use
Nicos Maglaveras
Professor & Director
Lab of Medical Informatics
Aristotle University
Thessaloniki, GREECE
EMAIL : [email protected]
Some background…..
• Health delivery is becoming more preventive,
continuous, and personalized
• A large scale of biomedical data and information
are becoming available for use in health delivery
• Biomedical data are produced by diverse sources
and technologies
• Medical decision support is directly related with
data quality, medical evidence and intelligent
semantics extraction and integration
Main Features of Personal Health
Systems (PHS)
• Personal health systems aim in the monitoring, intelligent
interaction between physicians and patients, implementation of
multiparametric information analysis, providing coaching and
intervention possibilities, enable medical decision support, and
personalize health services delivery.
• The main layers as it concerns information a PHS are the data layer,
the information layer, the knowledge engineering layer and the
output of information processing and analysis.
• In all the above mentioned layers, the primary concern is the
quality both in data and information, thus increasing the need for
filtering out noise and artifacts from wherever they originate.
• PHS lead to new R&D pathways as it concerns biomedical
information processing and management, as well as new pathways
in designing new intelligent medical CDSS enabling timely medical
interventions and quality health care delivery.
PHS – TELEHEALTH - AAL
• Telehealth is expected to become a reality in the near
future
• Today we are developing the third generation of
telemedicine systems, implementing closed loop
approaches such as e.g. in the HEARTCYCLE project
• Use of the above mentioned systems / modules is
expected to be used in the multimorbid patients
management arena and the AAL and healthy ageing
arena, involving users more and more
• In the end PHS, PGS, VPH, access platforms, are
expected to be integrated and work in an
interoperable and reconfigurable way
Major challenges in PHS data &
information gathering
• Embedding intelligence and medical knowledge
in personal health systems
• Multi-parametric modeling for clinical decision
support
• Contextualization of healthcare services
• Interoperability across personal health systems
and the clinical IT infrastructure (EHR, PHR, PGS)
• Patient and healthcare professionals’ acceptance
• Evaluation aspects of coordinated care (CC)
•
Data:
– are generated in huge volumes, fast & continuously
– vary in nature and complexity
– vary in structure; it can be even unpredictable! (=> ad-hoc solutions
required)
→ “Big data” characteristics: Volume, Velocity and Versatility (3V)
• “Big data” management:
– Cloud-based approaches have been proposed
– But still:
• efficient distribution of data and workload to support massive
parallel processing is a challenge
• data-intensive processing over a distributed network of computer
machines is required
6
• To generate “Value” out data, i.e. mastering the process to
derive insight from the data; requires:
– capturing data, aligning data from different sources,
transforming data for analytical processing, modelling
data and, finally, understanding the output as well as
visualizing and sharing the results
• Beyond “big data” management:
– Scalable Data Analytics (SDA) to empower organizations
in extracting knowledge from their data and support
decision making
– Deployment of flexible and open platform architectures
for data streaming, federated storage solutions and above
all robust and scalable data analytics
7
• Decentralization: Highly distributed across the healthcare
organization
• Wide range: Span from clinical information systems to research
information systems and personal health systems
• Technical heterogeneity, with differences in terms of:
– schematic and semantic information representation and
– data access (from legacy systems to well-defined / standardbased communication interfaces)
• Autonomy: Typically operate independently, but their linkage is
required
8
• “Big data” technologies are considered the cornerstone of
“Personalized Medicine”
• Medical information systems are “big data” producers,
especially when seen in an integrated fashion:
– e.g. Electronic Health Record + Personal Health Record +
Patient Monitoring System + Genetic Profiling (e.g. Next
Generation Sequencers)
• Exploiting value from data is a key quality procedure for
healthcare organizations:
– May help healthcare professionals in decision making
and patient treatment
9
Data complexity scale evolution
 Support wide-scale epidemiological studies by
managing and exploiting the wealth of sensor data
 Support the management and exploitation of data in
the context of wide-scale health-related studies (e.g.
concerning lifestyle) acquired via opportunistic sensing
by devices like smartphones
 Contribute in individualized care through scalable data
analytics
11
• Within the clinical environment a wide variety of
Information Systems operate
• The dominant Information System is the Electronic
Medical Record
• All actions and resources within the clinical
environment are expected to be captured and
annotated electronically
• The necessity for continuity of health/medical
records brings to the scene a key challenge:
interoperability!
12
• A wide range of data sources is available, capable to
facilitate medication safety: spontaneous reporting
systems, EHRs, patient reports, scientific literature
• Key challenges:
– effectively explore large volumes of heterogeneous
data
– capturing the dynamic features of the obtained drug
safety signals
13
 More Americans die from medical errors than from car
accidents, breast cancer, or AIDS annually, 44,00098,000 deaths/year
 Medication errors result in at least 1 death per day and
1.5 million people injured per year
 Estimated US annual cost of drug-related morbidity and
mortality is nearly $17 billion
 Preventable adverse drug events cost the healthcare
system $2.5 billion annually
Institute of Medicine, Preventing Medication
Errors, 2006
14
KOHN LT, CORRIGAN J, DONALDSON
MS. To err is human building a safer health
system. Washington, D.C.: National
Academy Press; 2000.
 The PSIP (http://www.psip-project.eu/)
approach:
 Identification of Adverse Drug Events (ADEs) via data
and semantic mining applied on large volumes of
patient data (EHRs), so as to obtain a better
knowledge of the prevalence of ADEs and of their
characteristics per Medical Department, Hospital,
Region, and Country
 Development of concepts, methods and knowledgebased modules to provide contextualized Clinical
Decision Support System (CDSS) functions for ADEs
prevention
16
 Identification of ADEs requires large volumes of
quality-controlled, longitudinal data to be
analyzed, e.g., millions of patient records
 Drug safety signals (typically expressed in the form
of rules) have a dynamic nature, i.e. their
importance (in terms of statistical significance)
varies according to local settings (e.g. a specific
hospital department/clinic) and time aspects
 Knowledge discovery has to be implemented as a
continuous process
 Decision support has to be provided in real-time:
Performance is an issue!
17
The “big data”
challenge
18
V. Koutkias, V. Kilintzis, G. Stalidis, K. Lazou, J. Niès, L. Durand-Texte, P. McNair, R. Beuscart and N.
Maglaveras, “Knowledge Engineering for Adverse Drug Event Prevention: On the Design and Development of a
Uniform, Contextualized and Sustainable Knowledge-based Framework”, Journal of Biomedical Informatics, vol.
45, no. 3, 2012, pp. 495-506.
19
30000
25000
# classifications
20000
15000
10000
5000
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Response time (seconds)
Scatter plot of the response time (ms)
and request size (89,340 classifications)
Histogram of the CDSS response
time (89,340 classifications)
20
 Wider spectrum of data exploration for ADE
identification
 More accurate identification & better
understanding of ADEs
→ Performance issues will occur at a larger scale:
Cloud or Grid computing necessary!
21
• Research Information Systems in biomedicine have been
recognized as a major source of new knowledge
• Typical examples involve data registries obtained through
specialized studies following application-specific protocols
• Large-scale studies require huge volumes of data to be
available and analyzed:
– data reuse is an option for acquiring the necessary
magnitude of samples
22
• The aim:
– Revealing evidence among factors contributing to
disease understanding/prognosis/evolution
• Challenges:
– Coping with heterogeneous data models
– The need for semantic integration of data
– The need for effective inference mechanisms
– Overcoming performance issues
– Reproducibility of the results obtained
23
Epidemiologica
l data
(e.g. lifestyle,
marital status,
sexual habits,
smoking habits,
etc.)
Clinical data
(e.g. clinical
tests, health
status)
Genetic data
(e.g. viral type,
variant, load,
personal and
family genotype
alterations)
Data Assembly
For: Hypothesis testing  Statistical evaluation 
Conclusions
Goal: Identify new markers of risk, diagnosis
and prognosis
24
An integrated environment that:
• virtually unifies multiple patient records
• facilitates the construction of study groups “on
demand”
• enables association studies combining
phenotypic and genotypic data
• automates the process of evaluating medical
hypotheses
25
Gynecology Clinic B
Medical
Format:
B data
Level of Detail: B
Format: B
Level of detail: B
Medical Data
Gynecology Clinic A
Medical data
Format: A
Level of detail: A
Medical Data
Format:
A
Level of Detail: A
A
I
SS
ST
Medical Data
Format:
C
Medical
data
Level of Detail: C
Format: C
Level of detail: C
Gynecology Clinic C
26
Hypothesis:
Combination of certain types of
p53, GSTT1 gene
polymorphisms and HPV
infection increases risk of
developing low-grade squamous
intraepithelial lesions.
Access
ASSIST for
more data
100 Gynecology Clinics join
ASSIST mutually allowing
access to patient data:
1. Data differ in degree of detail
2. Use different examinations
3. Have different ranges for normal
values
Medical
Researcher
X
Gynecology
Clinic A
ASSIST
Local Records:
Available p53 and GSTT1
tests for 35 patients
Group 1 -> 20
Group 2 -> 15
Gynecology
Clinic B
Gyn. Clinic C
ASSIST: 220 Cases
Group 1: 100 cases
Group 2: 120 cases
27
Encoding
AUTH
Charite
Un. of Ghent
0: Normal Inflammation
Hyperplasia Metaplasia
Carnification
Hyperkeratosis
PAPI
PAPII
0: normal
1: reactive changes
1: CIN1, ASCUS, LSIL,
AGUS
PAPIII
PAPIIID
LSIL
2: ASCUS
3: LSIL (CIN1 - koilocytosis mild dysplasia)
7: AGC-NOS
2
2: ASC-H, CIN2, CIN3,
Ca in situ, AIS
PAPIVa
PAPIVb
HSIL
5: HSIL (CIN2 - CIN3 - CIS moderate & severe dysplasia)
8: AGC-Favours neoplasia
4: ASC-H
9: AIS
3
3: Invasive cancer
PAPV
6: squamous cell carcinoma
10: Adenocarcinoma
0
1
28
Clinic 1
Clinic N
HPV Test
+/-
HPV Test
+/-
Pap Test
Class
1/2/3/4
Pap Test
Class
1/2/3/4
MTHFR
+/-
MTHFR
N/A
Colposcopy
Normal/LCIN/
HCIN
Diagnosis
CIN 1/2/3,
Ca
ASSIST
Virtual
Repository
Unification
HPV Test
+/-
Pap Test
WNL, LCIN,
HCIN, Ca
Diagnosis
Diagnosis
CIN 1/2/3,
Ca
Colposcopy
MTHFR
HPV test
Clinic 2
Colposcopy
Normal/LCIN/
HCIN
Test
Pap test
MTHFR
Colposcopy
+
-
Normal
WNL
LCIN
HCIN
Cancer
CIN2
CIN3
+
-
+/N/A
CIN 1/2/3,
Ca
Class1
Class2
CIN1
Class3
Class4
Relation <isMeasured>
Relation <subsumes>
29
RULE 1:
IF There is a Difference between any two diagnostic test
results (Colposcopy, Cytology, Biopsy) >= 2 ,
THEN The greater index should be used as <severity
index>
IF all test results are present and the difference between
the derived <severity index> and two test results is >=2,
then the validity of the derived <severity index> is LOW
IF all test results are present and the difference between
the derived <severity index> and 1 only test result is >=2
then the validity of the derived <severity index> is
MEDIUM
IF at least one test result is missing ...
30
31
for Heart
Diseases
The data engineering perspective
The knowledge engineering
perspective
for Breast
Cancer
for Cervical Cancer
for Colon
Cancer
32
Third generation telemonitoring
driven PHS
• Example case: A pervasive health care system, providing home
telemonitoring, surveillance, and educational services to chronic disease
patients
• Healthcare professionals monitor the patients’ status, based on their
regular interaction with the system, and accordingly regulate them
through medical interventions
• Persistent need for effectively managing and interpreting the large
volume of multi-parametric biomedical data collected during the patient
sessions / interactions with the system
→ Efficient information extraction and monitoring mechanisms are
required, capable of timely identifying cases where urgent attention is
necessary
PHS and Tele-monitoring of
chronic disease patients issues
• Big data are needed for Population Health
Management
• It is currently difficult to collect large data sets
in telehealth home monitoring
– Example from the HeartCycle experience
• The clinical trials process as well as the
modeling based CDSS, and Coordinated Care
Why do we need large data sets in
coordinated care?
• The growth in health expenditures is driven by multiple factors. One
critical factor is the rising incidence of chronic diseases, which account for
75% of the cost of medical care.
• Chronic patients have higher rates of unnecessary hospital admissions and
take many medications to manage their conditions.
• Traditional fee-for-service payment models that pay for treatment
transactions are ill-suited for serving patients that require close
monitoring and treatment tweaking and must be managed in a
coordinated fashion.
• The traditional providers of chronic disease management, primary care
physicians, only touch patients intermittently and rely on patients
themselves to comply with care plans and lifestyle recommendations.
Why do we need large data sets in
coordinated care?
•
Need for coordinated care
– Patients with complex chronic diseases and multiple comorbidities may
see on average 11 different doctors a year, creating major challenges for
communication, information sharing, reconciliation of care plans and
patient follow-up
– Another key driver of high health care expenditures is waste. Studies
found that roughly 30 percent of health care spending is wasted due to
unnecessary or poor quality care and a general lack of coordination
between providers.
– New incentives are required to increase accountability and foster a
continuous improvement culture focused on longitudinal patient
outcomes.
– In addition, all participants in the health care system – patients, doctors,
hospitals, health systems, long-term and post-acute care providers,
insurers and government agencies – need to have usable information
available to make informed decisions.
The HEARTCYCLE APPROACH
• HeartCycle developed a personalised disease management care system,
integrating care at home with professional care in the hospital.
• The approach follows a closed-loop disease management concept that
consists of two loops. An inner home-based loop directly interacts with
the patient in his daily life, giving feedback, motivation, and tailored help.
An outer loop involves medical professionals, maintaining a personalised
care plan for optimal therapy.
• The HeartCycle project aims to improve the quality of care for coronary
heart disease and heart failure patients by developing systems for
monitoring their condition at home and involving them in the daily
management of their disease.
• One of the main goals is to motivate patients to adhere to treatment
regimes and adopt beneficial lifestyles, with the expectation that
survivability of heart disease is improved and the overall cost of care is
reduced.
The HEARTCYCLE Professional & User
Loops
HEARTCYCLE CHF Monitoring and
Intervention Functionalities
CHF Management System Innovations
• #1: Medication Titration
• #2: Diuretic Management
• #3: Decision Support Analysis (Patient & Professional)
• #4: Education & Coaching for Patient Self-Management
• #5: Health Maintenance
Motivation: feedback on measures and
trends, what they mean and what to
do about them
Health-Care
Provider
Secondary
Loop
Analysis
Intelligent,
integrated, multimeasure (time &
type) personalised
analysis
Education: on healthy lifestyle, reasons
for treatments, self management
Patient / Carer
Primary
Loop
70% of Care Decisions
Communication
System
‘Monitor’
Guided Exercise Innovation
• #1: Independence & Compliance
• #2: Safety & Confidence
• #3: Improved Treatment Delivery & Closer
Follow-up
Overall HEARTCYCLE Assessment
System
HEARTCYCLE Assessment Use Cases
(AUC) - Innovations
• #1: Novel Sensors to Enhance Patient Assessment
in the Home Environment
• #2: New Information Processing for Integration &
Interpretation of Sensor & Patient Data
• #3: Improved Decision Support to Maintain
Patient Closer to Their Ideal Personal
Haemodynamic Profile
Novel Sensors
• For Assessing
– Congestion
• Bio-Impedance Monitor (BIM)
– Cardiac Output / Function
• Multi-Sensor (Sensatron)
– Arrhythmia
• Sensatron
• Bed-Sensor
Heart Sounds
Impedance
• Observational Studies
– Incident Events - Longer-Term Observations
– Congestion Resolution
• Randomized Trials
– Clinical Calibration
– Haemodynamic Interventions
Photoplethysmography
GEx – Product Overview
Monitoring Service Platform; targeted to patients with CAD providing real-time guidance
while exercising, as well as support and motivation while following an exercise plan.
46
GEx – System Components
Use
Development
pending*
Shirt
(6 different sizes
for males and
females)
to attach sensors and keep them
in skin contact
Biocompatibility testing
for sensation and
irritation according to
ISO standard
Clothing+
Image sensor
(with a cradle)
records vital signs and exercise
related signals
CE certification & CE
certification as medical
device
CSEM
Patients portable
device, including
a mobile phone
PDA mobile phone for patient
interaction; performs sensor data
processing, provides feedback to
patient during exercise
Continua Certification
& CE certification as
medical device
UPM
Motivation &
Education content
Strategy and content to change
lifestyle
Patients station
main interface between the
patient and the GEx system;
Gateway between the Portable
Station and the Professional
System and is used as
motivational and educational
platform
Continua Certification
& CE certification as
medical device
UPM
Continua Certification
& CE certification as
medical device
TSB/ITACA
Component
Portable
Station
Patient
Station
Professional
System
Professional
System
for interaction with medical
professionals. Based on clientserver architecture, medical
professionals are able to access
to the central server using a
standard web browser
Integrated
47
System
Integrated
System
integration with main server that
stores and processes recorded
data
Exploiter**
NHS Heart
Manual
TSB
*Refers to
additional
development issues
required in order
for the system to
be marketed.
**The GEx system
will be
commercially
exploited through
subcontracting an
external
manufacturer
under license.
GEx – Competitive Positioning
Classical Polar’s belt integrated into an Adidas shirt and bra
The competitive advantage of the GEx system is
that it is tailored specifically to patients with
CAD and their needs
The BioHarness BT belt by Zephyr
48
Sensatron – Product Overview
Around 10 million people in the
EU and 5.7 million people in US
with HF. There is an unmet
need of the cardiologists to be
able to frequently monitor
haemodynamics in order to
safely titrate medication and to
effectively target doses
Solution
Sensatron is an on-body-unit that
measures clinical haemodynamic
parameters with the aim to be
used in home based telehealth
scenarios by
patients with CHF
M
A
R
K
E
T
Targeting the home telemonitoring
(HTM) market with focus on systems
for
non-invasive haemodynamic
monitoring
BedSensor – Product Overview
Ballistocardiography (BCG) based system
for non-obtrusive monitoring during
sleep that can be easily used at homes
sensitive 8-channel foil sensor
embedded in bed mattress
Adresses the need for cost-effective and
.
reliable
solutions for the diagnosis and
follow-up of sleep disorders
Magnetometer – Product Overview
non-contact monitoring of cardiorespiratory activity; respiration and pulse rate
One-Channel Magnetometer
applied to chair back-rest
(a) measurement sensor sewn coil and
flexible electronic part
(b) male volunteer wearing a shirt including
the sensor
51
cECG– Product Overview
non-contact monitoring of cardiac related signals on the body surface
without the necessity to undress the user
Early stage prototype of a capacitive ECG
measurement system (left);
the according block diagram (right)
Capacitive electrodes
integrated into the driver
seat of a Ford S-Max
52
Modeling Approaches on Adherence
and Complications of Chronic
Cardiac Patients in HEARTCYCLE
Perspectives And Challenges Based On Continuous
Data Gathering And Personal Health Systems
PHS and chronic patients
• In the scope of PHS data analysis, two basic pillars that
may support medical decision, especially in complex or
multimorbid chronic patients, are:
– Assessment of the person’s response to treatment,
including complications, which links directly to the
needs for care plan updates by the health
professional.
– Assessment of adherence to treatment, which reflects
the patient’s responsibility in following the therapy as
agreed and prescribed by the professional.
Main concepts
• Medication effectiveness
• actual change of vital signs as a result of medication
treatment, with respect to the expected change
• Medication compliance
• patient’s behaviour, as regards the extent to which
the prescribed medicines are taken as agreed with
the medical professional
Assessment of Compliance – Why?
• Costs and treatment inefficiency
• Barrier to treatment optimisation
• 'less forgiving' drugs that, when missed, may
lead to an adverse event (e.g. withdrawal
symptoms) or disease exacerbation.
Compliance Prediction Approaches
• Known factors (self-efficacy, depression, health literacy,
medication knowledge) in patients with specific
profiles (e.g. elderly patients with chronic diseases)
– Predictors of medication adherence may use machine
learning
• Assessment of ability
– Drug Regimen Unassisted Grading Scale (DRUGS). The
DRUGS tool uses a performance-based measurement to
assess the individual’s ability to identify, access, and
determine the dosage and timing of their medications.
Assessment of Compliance
• Direct questioning “Did you take your medication?”
– unreliable
– judgmental
• Questioning
– through patient or caregiver interview using open-ended,
non-threatening and non-judgmental questions.
• Prescription refill records and pill counts
– overestimate true adherence rates.
– Temporal information ?
• Other devices
– Digital pill – still in its infancy.
Adherence and vital signs
• A time series of vital sign values of a patient is
expected, after a transitional period due to treatment
change, to statistically converge, unless specific causal
factors occur in that instance, which might include an
acute medical condition, a period of non-adherence,
etc..
• Patterns of deviation from the steady state values can
be quantify with data driven approaches
• This is advantageous due to the existence of vital sign
recording in various setups
Treatment
effect
population-based
medical evidence
• Amplitude of effect
• Time scale
• What are the factors
potentially causing a
temporal variation?
Example
Acute
illness
Bad diet
?
?
Increase
salt
Withdraw
diuretics
worsening HF
AFib
Example: for a
patient, we notice
that she has:
HR
•Increased BP
•Increased weight
What are the possible reasons?
Increase
weight
BP
symptoms
Clinical Data Protocols
Test1
HF & HTN patients
The difference between MD-MG days in
cycle1 relates to incompliance effect
SBP-DBP-HR
lying supine, standing, sitting,
exercise task preparing
The difference between MD-MG days in cycle2
relates to relates to deviations under medication
Test 1 - What differs in HF
• Morning
between day
differences vs
• Afternoon
between day
differences
• Morning
deviation due to
medication
omission
• -relate to salt
• -activity?
Test 1 - What differs in HTN
• Standing/sitting
differs due to
medication omission
• Also due to salt
Classifying medication effect &
compliance
o Single features and pairs of features were tested via linear classifiers with leave-one-out cross-validation
o The classification of the incompliance detection seems better in HS
o In HTN, the standing to sitting differences (*diff) appear important for classification
o In HF, systolic pressure seems more important in LS and diastolic pressure more important in HS
Clinical Data protocols
Test2
HF & HTN patients
In Normal diet
• Reference day
• Medication given same
morning
• Medication omitted for
48hrs, and taken after the
two cycles of
measurements
SBP-DBP-HR
Medication & Activities
• how vitals change with activity & medication
in HF
155
med taken
med omitted
150
145
Medication
omission
mean
140
135
130
Among
activities
125
120
115
SBP-semiR
SBP-su
SBP-Leg
SBP-st
SBP-si
SBP-nit
SBP s6
25
med taken
med omitted
20
15
10
mean
5
0
-5
-10
-15
-20
-25
SBP-semiR
SBP-suSBP-LegSBP-st SBP-si SBP-nit SBP s6
DBP-semiR
DBP-suDBP-LegDBP-st DBP-si DBP-nitDBP s6
HR-semiRHR-suHR-Leg HR-st HR-si HR-nit HR s6
Constructing Personal models of vital
signs
• daily vital sign data may
vary due to medication,
activity, time of the day,
lifestyle and diet,
as well as other reasons
related to health, stress
or challenges of life.
• non-linear mixed effect
modeling (NLME)
SBP=f(salt,time,medication,activity)
• each parameter is the
sum of a fixed and a
random effect, which
expresses personal
variations per patient
The model
Low salt SBP
High salt SBP
Med*+
0.4975
0.3006
Time*+
0.5104
-0.1739
Activity*
1.8415
1.3527
Ref
0.8676
0.9290
Ref bias
-0.2041
-0.1511
Correlation
0.5333
0.4455
Y(i)=
beta(1)*sign(time-24h)*time from last med+
beta(2)*time of day+
beta(3)*activity index+
• * the random effects in LS
beta(4)*mean value+
• + the random effects in HS
beta(5)*(mean-thres); thres=140 for SBP
Model results
• Low salt SBP
12
20
High salt SBP
10
15
8
10
6
4
5
2
0
0
-2
-5
-4
-10
-6
-15
0
2
4
6
8
10
-8
12
0
2
4
6
8
10
12
SBP, Low Salt
150
real
axis
model
140
130
120
A subject in low salt
110
100
90
start bedstand1sit1stand2Ex start bedstand1sit1stand2Ex start bedstand1sit1stand2Ex start bedstand1sit1stand2Ex
In black: model
In red: real
Test model with the other datasets
• We need to generalise descriptions
• Patient
 Problems and
– Disease
– Diet
– Medication
– Personal baseline values
– Recordings
• Time
• Activity
• Value
challenges




Missing values
Daily Variation
Different profiles,
comorbidities,
medication dosages
How to encode
different activities?
From controlled experiments to
continuous home data
• Extending these methods to an uncontrolled
PHS case would be extremely exciting
– for actually understanding patient’s health
condition in real life, including medication effect,
compliance, worsening
– for guiding patients in achieving health goals
– for retrospective data repurposing: the discovery
of new patterns and knowledge based on
longitudinal continuous physiological data
From controlled experiments to
continuous home data
• New challenges, including:
– How can large scale annotated data be requested and
acquired
– How can the various daily life conditions be mapped
to unambiguous and quantitatively processable
factors,
• Use contextual information , activity, place, lifestyle and out
of regular events, medication, stress, symptoms, diet
– How can these factors be combined to a model that
may help interpret data, and guide patient/health
professional
Steady
properties
Cloud
Cloud Storage
Storage of
of
PHS
PHS data
data
For example
in IndivoX
Treatment
plans and
schedules
Activity, e.g. 1 value
per min
…
…
Day y
Modifiable
health
parameters
Patient
Day x
•
Personal Health Systems and
Continuous Data
Activity
HRV, e.g. 1Hz
HRV
BP, e.g. 1 value per
min
BP
Diet, as reported
Diet
Stress, as reported
Stress
Assuming
a basic
local
processing
Sleep, e.g. 1/30
hypnogram
Event/Symptom, as
reported
How much data?
In one day/one subject
• HR: 86400 samples
• BR: 86400 samples
• Act: 1440 samples
• SBP/DBP: 1440x2 samples
• Hypn: 2880 samples
180000 samples per person per day
(not to mention context information)
In more intensive recordings…
• Heart activity monitored by
electrocardiography (ECG), I KHz
– -> 86.4 million readings a day per patient.
– -> heart rate and respiration rate, 86,400 readings
a day per patient.
• Impedance between each of the three ECG
leads attached to the patient’s chest
– chest wall movement to assess breathing rates
– ->5.4 million data points a day per patient.
Some HEARTCYCLE Conclusions
• The overall HEARTCYCLE CHF and Gex telemonitoring
system has produced the reference data set in its trial
phase
• AUC are currently used, with new sensors and intelligent
modules so as to permit the production of big data coming
from PHS in cardiology and other chronic disease domains
• The overall high level architecture of the HEARTCYCLE
assessment system is defined
• The closed-loop system is realised and tested for
performance optimisation and assessment of its
performance in clinical environment
• A third generation of telemonitoring systems is launched
and currently in use
Coordinated Care, Big Data and the
ACT project
What lies ahead in telemonitoring
and ICT for health?
• Consolidation of the processes for the acquisition of big data
in telemonitoring health applications
• Apply ICT for health telemonitoring systems in multi-morbid
patients and healthy young individuals for disease
management, prevention and behavioral patterns assessment
• Identify and use key performance indicators (KPIs) in regional
settings to assess CC in telemonitoring
• Simplify the interfacing layer by using new I/F technologies
• Integrate clouds of BAN big data with social media big data
and semantics
• Address regulatory and ethical constaints
The ACT project
Chronic illness poses a huge individual, social and economic burden. In the EU, heart failure alone accounts for over €10 billion per
year in direct healthcare costs. Mortality rates are over 30% in patients whose condition is not controlled.
Co-ordinated care and telehealth services have the potential to deliver quality care to chronically ill patients. These systems can
both reduce the economic burden of chronic care and maximise the delivery of clinical support, despite the shortage of skilled
professionals within European healthcare systems.
The value of telehealth services has been highlighted in several clinical studies and test cases. However such services haven’t yet
progressed substantially beyond pilots because they require new behaviors and ways of working directed at improving health
outcomes, administrative efficiency, cost effectiveness and user (patient and health professional) experience. Translating evidence
into practice is complex and requires significant organisational change.
The Advancing Care Coordination and Telehealth (ACT) programme aims to overcome these barriers. This international consortium
of key stakeholders (companies, universities, hospitals and healthcare authorities) investigates best practices in areas such as
patient risk stratification, patient adherence, professional engagement, optimisation of organisation and workflow structures, and
efficacy and efficiency factors.
The objective of this EU co-funded project is to overcome structural and organizational barriers. This will:
•
Gather data and best practices from 5 regions
•
Determine a baseline for how care coordination and telehealth works in these regions
•
Conduct an iterative evaluation of care structures and procedures
•
Create a best practices “cookbook” to ensure that the findings can be replicated in other EU health regions
. The ACT programme is fully aligned with the European Innovation Partnership in Active and Healthy Aging objectives to deploy
integrated care for chronically ill patients.
Picture from Philips
Communications
Link to EIP AHA B3:
https://webgate.ec.europa.eu/eipaha/act
iongroup/index/b3
ACT expected Outcomes
• Outcomes:
Principal outcome of ACT is the evaluation of key drivers & effective
deployment of CC&TH services in the five participating regions, a best practice
Cook Book specifying how the drivers can be leveraged to expedite
deployment of CC&TH services in other European healthcare regions. Cook
Book will provide explicit recommendations & examples regarding key drivers.
Recommendations will take into account the regional/national boundary
conditions & other diversity aspects.
• CC&TH drivers summarised in the Cook Book.
– Optimising CC&TH organisation & workflow structures.
– Effective Patient Stratification
– Improving Patient Adherence and Staff Engagement / Education
– Improving Care Provider efficacy and efficiency
ACT Objectives
•
•
The ACT programme is the first of its kind, specifically designed to help overcome the barriers
hindering care coordination and telehealth deployment.
It is specifically designed to concentrate the efforts on the key drivers:
Workpackages
Objectives
Optimisation CC&TH Organisation
and Workflow Structures
Effective deployment of CC&TH requires modification of the care provider eco-system, ensuring
seamless, integrated delivery of patient-centric care. During the Baseline phase, CC&TH concepts
and IT solutions in the regions are evaluated.
Patient stratification
Ensure CC&TH services can be tailored to needs of individual patients, accurate/easy to apply
patient stratification tools are required: evaluate solutions (Baseline), optimise/enhance tools
during Iteration phase, identify best in class solutions.
Improving Patient Adherence and
landscaping of staff engagement
concept
Evaluate current staff engagement and educational packages used within the region to support
Care Coordination & TeleHealth, and provide this overview as part of the overall ACT evaluation
package.
Investigate how patient anxieties can be overcome & services effectively implemented to improve
patient empowerment & adherence. After interactive phase (M24) service & solutions have been
evaluated & concepts improving outcomes listed.
Improving care provider efficacy &
efficiency
Clinical staff shortage mandates that existing resources are effectively utilised & deliver quality
care. Assess guideline based CarePathways & algorithms targeted at HF,COPD, diabetes &
comorbidities. Improve nurse/patient ratios.
WELCOME project – Multimorbid
patient integrated care model
Biosensors recording evolution enabling higher
dimensionality in biodata acquisition
passive probes
classical approach
standalone
sensors
active probes
(sensors)
+ in-situ electronics
(allow optical and
acoustical sensing)
+ simple connection
(no cable)
2-wire bus for:
ECG, EIT, isochronous synchronization, communication and
simultaneous recharge
state-of-the-art sensors (developed by CSEM)
top: standalone sensor for one-lead ECG
bottom: active probe for ECG and SpO2 (at chest)
frontend
analogue
electronics
outer ring for
current injection
(EIT)
microcontroller
for data
acquisition &
processing
isochronous
synchronization
mechanism and
communication
intermediate ring made
of electret for sound
measurement
simultaneous
recharge
mechanism
inner ring for voltage
measurement
(EIT and ECG)
SPLENDID project – Prevention and
coaching of healthy young people
Smart Dietary Monitoring
Sub-System Components
EAR-CH-SW SENSOR
(EARCS)
CHEWING-SWALLOWING SENSOR
INTEGRATED TO EARPHONE SET
ON-BODY DEDICATED DEVICES
- EARCS (embedded sensor in earphone set)
- Electrodes chewing / swallowing detection
- Embedded in standard fancy earphone set
- Sensor & audio wiring embedded
- Signal / Audio connector to ADL
Wireless Local Area
Network (WIFI)
Wireless Wide Area
Network
(GSM, GPRS, UMTS)
- ADL (in-pocket/belt)
- Sensor input analog front end / data acquisition
- Chewing/Swallowing detection algorithm
- Embedded accelerometer (ACC)
- Activity classification and energy expenditure
- User presence / absence detection
- Wireless DATA communication to SP
- Wired (standard) AUDIO connection to SP
PERSONALIZED
MONITORING AND
GUIDANCE PLATFORM
DATA TRANSMISSION
MOBILE PHONE
MSGS, MUSIC, ETC.
USER PRESENCE/ABSENCE
CHEWING &/OR SWALLOWING DETECTION
ACTIVITY CLASSIFICATION
ENERGY EXPENDITURE
ACCELEROMETER
DATA LOGGER (ADL)
SMARTPHONE OR
EQUIVALENT (SP)
Smartphone (User interface/Gateway)
MEAL
INTAKE
- GUI application
(PMGP-SP)
- Monitoring / Feedback
- Gateway
Bluetooth 4.0 or ANT+
Bluetooth
WiFi
EDGE ; GSM/GPRS ; UMTS
Mandometer (MMT)
- Food consumption during a meal
- Communication bridge to SP (tbd)
Conclusions
• Large scale data are a function of sensor recording capacity,
ICT barriers (e.g. network availability, communication means),
energy scavenging, knowledge management & engng, patient
empowerment, regulatory & ethical constraints, adaptability
of clinical trial protocols, integration of social network
information and semantics, regional KPIs, virtual physiological
human models, computational resources……..
• We are currently still in the beginning of a long journey to the
unknown world of big data
• We need tools and means to be able to navigate so as to
achieve efficient CDSS, patient adherence, personalized
health and production of solid evidence based medicine