Transcript AssistMe projektet

AssistMe
Project leaders
Ankica Babic, Urban Lönn, Henrik Casimir Ahn
Problem solving 1
• Start with clinical questions that should be
supported by decision support and data mining.
• Distinguish levels of decision support: from user
driven to structured procedures for knowledge
mining:
– Cluster analysis, Case Based Reasoning (CBR),
statistical reports
– More, specialized reports?
Problem solving 2
• Actively involve the physicians in design,
implementation, and evaluation of our web
based system.
• Clinical evaluation of extracted knowledge.
System overview
Start page
Homepage for patients
Questionnaires
Homepage for physicians
Add patient cases
Case based reasoning
(result)
Case based reasoning
(patient case)
Cluster analysis - introduction
Cluster analysis
Cluster analysis
• Calculates the equality/difference between patients
Example: Calculation of difference
using age and weight:
c = “distance” between patients
a = age difference = 40 years
b = weight difference = 30kg
kg
c2  a 2  b2
90
c  a 2  b2
c
b
c  402  302  1600  900 
 2500  50
60
a
20
The difference is:
60 years
50
Cluster analysis
• Calculates the equality/difference of patients
• Places “similar” patients in the same groups
(clusters) and “different” patients in different
groups.
• The user can choose what variables to use for
comparing the patients when the population is
divided into subgroups. The number of groups
must also be specified.
• Additional information, such as the survival
percentage, is provided for the different groups.
Clusters (former page)
Age
2
80 80
6
4
70 70
60 60
50 50
5
3
1
40 40
30 30
20 20
10 10
Higgins
0 0
0 0
2 2
4 4
6 6
8 8
10 10
12 12
14 14
Outcome
Age
0,67
80 80
0,63
0,5
70 70
60 60
50 50
1,0
1,0
0,87
40 40
30 30
20 20
10 10
Higgins
0 0
0 0
2 2
4 4
6 6
8 8
10 10
12 12
14 14
What is w and b in the summarization table?
•w is short for “within distance”
•b is short for “between distance”
Large within distance
Small between distance
W/b=Large
Not a
good result!
What is w and b in the summarization table?
•w is short for “within distance”
•b is short for “between distance”
Large within distance
Small within distance
Small between distance
Large between distance
W/b=Large
Not a
good result!
w/B=Small
The desired
result!
Homogenization
In order to be able to compare different variables which have different magnitude of values.
Age
Patient 1: Age 61; Higgins 7
Patient 2: Age 72; Higgins 14
100
1
0,82
78
72
c
0,32
61
0,50
0,57
c  0,57 2  0,32 2  0,65
44
0,43
1
0
0
Higgins
0
4
7
11
14 15
Automatic cluster
Automatic cluster - setup
Automatic cluster – results
Design of user interface
Design for usability
• The design process is a constant shifting
between the following three abilities
– The ability to understand and formulate the
design problem
– The ability to create design solutions
– The ability to evaluate those solutions
How to create premises for the
design
• Initial understanding – What? Who? Where?
Why?
• Studies of literature
• Fields studies
• Increased understanding of What? Who?
Where? Why?
Field studies
•
•
•
•
Contextual research
Create scenarios
Design/ Style studies
Task analysis
Qualities in use
What is “good” for this type of system, these
users in this context?
Important qualities and what they are based on
• Aesthetic values: the feeling of a trustworthy system
• Practical values: easy to learn, effective use, possibility to
abort actions
• Psychological values: cognitive ease of use, psychological
support
• Autonomic values: Freedom of choice
• Social values: facilitate consent, supporting ”the team
mind”
Design phase
•
•
•
•
•
•
Sketch, evaluate, comment
Create paper prototype
Test paper prototype
Create computerized prototype
Test computerized prototype
Implementation
“The
doctor’s information tool of the
future might be some sort of combination
between the patient record and the
Internet, with the doctor and the patient
positioned together at the intersection but
not having to pay attention to the
technology.” (Smith 1996)
Database design
Layered structure
Application (AssistMe)
Database interface
Database manager / system
Layered structure
Java code of AssistMe
Database interface in Java
Patient cases
Meta Archive
database database
...
...
Old database design
• Flat structure (little or no relations)
Data
Data
Data
Data
Data
Data
Data
Patient case
Data
Data
Data
Data
Data
Data
Data
Data
Data
New database design
• Relational database design
Data
Data
Data
Data
Data
PreOp
Rel
Data
Rel
PostOp
Data
Rel
Discharge
Data
Demografi
Data
Data
PerOp
Rel
Data
Data
Data
Data
Database design
• Structured Query Language, SQL
– Standard for commercial database managers
– Easy to transfer information to and from the
database.
Database design
• Dynamical structure
– Should be easy to change the type of data that is
stored in the database
• Support for more than one database in the
system at once
– The system can be used in parallel for different
purposes.
Database interface
• Database interface specially developed for
the system
– Easy to read and write information in the
database.
– Easy to add new tools (Cluster, CBR, …) that
utilizes the databases.
LVAD Outcomes
• Overview of the area: functionality, clinical
use (bridge or destination therapy, continued
care), types/families of LVAD, short
technical descriptions and pictures.
• Scenario from start to end. QoL (including
cost consideration).
• This is focused on the aspects of morbidity
and mortality. Literature studies.
Mortality
• Definitions, surgical perspective on it, heart
transplant specific aspects and reflection
over the follow up and waiting time prior to
transplantation.
• Accepting the 30 days survival as standard.
All mortality is registered including cause of
death.
Morbidity
• Complications. Technical and clinical
complications with reference to device
related problems.
• Definitions of complications (clear cut
and/vs. working definitions), motivating the
definitions used in this research. Addressing
verity and complexity of definitions.
Morbidity
• Motivation or/and pragmatic reasoning
about the morbidity.
• Research vs. clinical thinking.
• Give better understanding of mechanisms
involved in order to reduce the incidence
(Piccione Jr. W. 2000).
Risk Factors
• Overview of risk factors used within the
LVAD domain and their usage to assess
morbidity and mortality.
• Higgins, Euro scores, other systems for risk
stratification.
• Outlines we have accepted in our research.
Patient Selection
• In terms of indications, demographic data,
selection criteria in use, ethics around it.
• It is of paramount importance to choose
patient that is ‘appropriate’ for treatment to
succeed.
• (See Left Ventricular Assist, Fraizer, 1997)