eventflow presentation

21
Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/ EventFlow Megan Monroe Krist Wongsuphasawat Catherine Plaisant Ben Shneiderman Jeff Millstein Sigfried Gold

Upload: madeyjay

Post on 19-Jan-2015

11.373 views

Category:

Technology


1 download

DESCRIPTION

 

TRANSCRIPT

Page 1: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

EventFlow

Megan MonroeKrist WongsuphasawatCatherine PlaisantBen ShneidermanJeff MillsteinSigfried Gold

Page 2: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Temporal Event Sequences

Item Identifier Type of Event Time (Start/End)Patient #1 Stroke 11/3/2011Patient #2 Heart Attack 7/1/2011Patient #1 Diagnosed 11/5/2011

Data of the following format:

Examples:• Electronic Health Records• Process Flow• Web Logs

Page 3: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Big Data Data Mining

Post-marketing surveillance has become a viable alternative to clinical trials (example: medical insurance databases).

Advantages: Cost effectiveNo direct effect on patients

Disadvantages: Data is messy and impreciseMust use very large datasets

Page 4: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

US Army Pharmacovigilance Center, OTSG

14 million EHR’s(8 billion observations)

+Standard Statistical Software

(PVDAS, SAS, Systat, etc.)

Difficult to: Develop new questionsFind unusual event

patterns Understand dataset

holisticallyCommunicate findings

Page 5: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

EventFlow Motivation• Allow researchers to evaluate their data holistically

- Have immediate visual access to data- Develop new questions/hypotheses- Make rapid adjustments to view and

dataset

• Facilitate communication about the data and findings

• Integrate easily into existing systems (open source Java coding)

Page 6: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Point-Based EventsSTROKEADMITTEDDIAGNOSED

ASPIRIN

DIET

STEROIDS

Interval-Based Events

Page 7: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Individual Patient Records

Patient #1

Time

Patient #2

Patient #3

Patient #4

….

….

Page 8: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

EventFlow Display

Page 9: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Aggregated Patient Records

Page 10: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Aggregated Patient Records

Page 11: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Interval Merging

Page 12: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Interval Querying

Page 13: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Query Scope: Presence EventsPoint/Point Point/Interval Interval/Interval

Page 14: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Query Scope: Range Constraints

Point/Point Point/Interval Interval/Interval

Page 15: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Query Scope: Absence Events

The absence of a point event necessarily implies that the absence spans the period of time between the previous presence event (or the beginning of the record) and the subsequent presence event (or the end of the record).

Point Absences

The absence of an interval event can occur for both a span of time, or at a point in time.

Interval Absences

Page 16: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Query Scope: Absence EventsPoint/Point Point/Interval Interval/Interval

Page 17: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Basic QueryingSubsequence:

Overlap:

Page 18: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Advanced Querying

Page 19: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

LABAs Demo

Page 20: EventFlow Presentation

Research supported by NIH and Oracle http://www.cs.umd.edu/hcil/eventflow/

Next Steps

• Complete advanced query implementation

• User tests for advanced query interface

• Begin analysis of new datasets (Children’s Hospital)

Page 21: EventFlow Presentation

EventFlow

Contact: [email protected] http://www.cs.umd.edu/hcil/eventflow/