Upload
krist-wongsuphasawat
View
109
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Paper presentation at the Workshop on Visual Analytics in Healthcare in conjunction with the IEEE VisWeek 2011, Providence, RI, 2011.Abstract:Electronic Medical Record (EMR) databases contain a large amount of temporal events such as diagnosis dates for various symptoms. Analyzing disease progression pathways in terms of these observed events can provide important insights into how diseases evolve over time. Moreover, connecting these pathways to the eventual outcomes of the corresponding patients can help clinicians understand how certain progression paths may lead to better or worse outcomes.In this paper, we describe the Outflow visualization technique, designed to summarize temporal event data that has been extracted from the EMRs of a cohort of patients.We include sample analyses to show examples of the insights that can be learned from this visualization.
Citation preview
m
OUTFLOW
Krist Wongsuphasawat David H. Gotz
Visualizing Patients Flow by Symptoms & Outcome
IBM T.J. Watson Research Center
m
m
Congestive Heart Failure (CHF)
Electronic Medical Records
m
Patient #1
Time
Aug 1998 Ankle Edema
Jan 1999 Weight Loss
Oct 1998 Cardiomegaly
m
Ankle
Patient #1
Cardio. Weight
Ankle
Patient #2
Cardio. Rales
Time
Ankle
Patient #3
Cardio. Rales
Ankle
Patient #n
Cardio. Rales Weight
Many patient records
m
with outcome
Ankle
Patient #1
Cardio. Weight
Ankle
Patient #2
Cardio. Rales
Time
Ankle
Patient #3
Cardio. Rales
Ankle
Patient #n
Cardio. Rales Weight
Die (0)
Live (1)
Live (1)
Live (1)
m
information overload!
6,000 patients
6,000,000 medications 200,000 symptoms
m
consumable
m
Overview / Summary
Millions of records
m
Steps 1. Aggregation
2. Visual Encoding
3. Interactions
m
Step 1: Aggregation Patients Outflow graph
m
Patient #1
Patient #2
Patient #4
Patient #3
Patient #5
Patient #6
Patient #n
Patient #7
…
Outflow Graph
Patient records
m
Assumption • Symptoms are accumulative.
Ankle
Patient #1
Cardio. Weight
Patient #1
m
Assumption • Symptoms are accumulative.
Ankle
Patient #1
Cardio. Weight
Ankle
Patient #1
Ankle Ankle
m
Assumption • Symptoms are accumulative.
Ankle
Patient #1
Cardio. Weight
Ankle
Patient #1
Ankle Cardio.
Ankle Cardio.
m
Assumption • Symptoms are accumulative.
Ankle
Patient #1
Cardio. Weight
Ankle
Patient #1
Ankle Cardio.
Ankle Cardio. Weight
m
Assumption • Symptoms are accumulative.
Ankle
Patient #1
Cardio. Weight
Ankle [A]
Patient #1
Ankle Cardio. [A,C]
Ankle Cardio. Weight
[A,C,W]
State
m
Select alignment point Target patient’s current state
Ankle Cardio. Weight
[A,C,W]
m
Filter patients
[A]
Patient #1
[A,C] [A,C,W]
[A]
Patient #2
[A,W] [A,R,W]
[A]
Patient #3
[A,W] [A,C,W]
[A,C,R,W]
[A,C,R,W]
[A,C,D,W]
m
Select alignment point Target patient’s current state
What are the paths that led to ?
What are the paths after ?
Ankle Cardio. Weight
[A,C,W]
m
Outflow Graph
[A,C,W]
[A,C]
[A,C,D,W]
[A]
[ ]
Alignment Point
m
Outflow Graph
[A,C,W]
[A,C]
[A,W]
[A,C,D,W]
[A]
[ ]
Alignment Point
m
Outflow Graph
[A,C,W]
[A,C]
[A,W]
[A,C,R,W]
[A,C,D,W]
[A]
[ ]
Alignment Point
m
Outflow Graph
[A,C,W]
[A,C]
[A,W]
[C,W]
[A,C,R,W]
[A,C,D,W]
[A]
[C]
[W]
[ ]
Alignment Point
Average outcome = 0.4 Average time = 10 days Number of patients = 10
m
Step 2: Visual Encoding Outflow graph Outflow visualization
m
NOW Future Past
A!C!
A!
C!
A!C!W!
A!C!D!
Color is outcome measure.
Node’s height is number of patients.
Time edge’s width is duration of transition.
Node’s horizontal position shows sequence of states.
time edge
link edge
End of path
m
m
Step 3: Interactions Static vis. Interactive vis.
m
Interactions • Panning
• Zooming
• Brushing + Freezing
• Tooltip
• Highlight target
m
Sample Analysis What can we learn from it?
m
Analysis Demo • outflow_analysis_demo.mp4
m
Steps 1. Aggregation – Outflow graph
2. Visual Encoding – Sketch
– Visualization
3. Interactions – Details on demand
m
Future Work • Evaluation & Design Improvement
• Use outcome from predictive modeling
• Similarity measure to select similar patients
m
Conclusions • Electronic Medical Records
– Rich information
– Large
• Visualization: Outflow – Visual summary: overview
– Interactive exploration: zoom, filter and details
• Not specific to CHF, or medical domain
Contact me [email protected] @kristwongz
m
Soccer Results
2-1
2-0
1-1
0-2
2-2
3-1
1-0
0-1
0-0
Alignment Point
Average outcome = win/lose Average time = 10 minutes Number of games = 10
m
Acknowledgement • Charalambos (Harry) Stavropoulos
• Robert Sorrentino
• Jimeng Sun
m
Conclusions • Electronic Medical Records
– Rich information
– Large
• Visualization: Outflow – Visual summary: overview
– Interactive exploration: zoom, filter and details
• Not specific to CHF, or medical domain
Contact me [email protected] @kristwongz
m
THANK YOU ขอบคุณครับ