Hippocrates Archive

1
Hippocrates Archive

• John Brennan (jbren_at_wam.umd.edu)
• Hyunyoung Song (hsong_at_gmail.com)
• Nima Negahban (nimacn_at_gmail.com)

2
Introduction

• Patient Data Management Application
• Inherited application Pattern Finder
• Pattern finder was using fake data
• Conducted usability studies to make
  recommendations on the interface

3
Motivation

• Twenty percent of the nations medical spending is
  on managing patient data
• Paper Systems
• Potential to be an invaluable tool for doctors
  and researchers

4
Previous Works

• Lifeline
• One of the first applications in this field
• Very good at displaying large sets of data
• Lacked the ability to make complex queries

5
Previous Works

• KNAVE
• Robust query system
• Poor Visualization System

6
Pattern Finder

• Medical History Search Application built by Jerry
  Fails, Amy Karlson, Layla Shahamet in Spring 2005
  CMSC838S term project
• System Spec
• C program on top of .NET Framework
• Piccolo api for visualization

7
(No Transcript)
8
Dataset

• Test - Students
• Original Data
• Medication
• Tests
• Visits
• Values
• Imaginary values

• Test - Doctors
• National Washington Hospital Data Set
• Only tests (blood test)
• Real values with reference level

9
Usability Study 6 students

• 3 scenarios with 3 tasks each
• Participants used Pattern Finder with its
• default (phony) data
• Tasks created to test several key
• fundamentals of the application
• Understanding events
• Creating simple and complex queries using each of
  the tools provided
• Our study gave more focus to the query
• construction than visualization understanding

10
Observation
User Troubles

• 3 tier menu hierarchy
• Understanding full query when more
• than 3 events have been created
• Creating time spans between events
• Difficulty querying medication changes
• (relative values feature)

11
Post-Test Result

• Used Shneidermans QUIS model (1-9 semantically
  anchored radio buttons)

12
Recommendation

• Create text query window to display
• full query user is generating
• Change editable span text to NumericUpDown
  controls
• Change relative values text wording to
  increasing/decreasing values
• Those using custom data for the system should use
  care when developing the event drop down menu
  hierarchy (naming and placement)

13
(No Transcript)
14
Usability Test 2 Doctors

• No need for scenarios
• 6 Concrete Tasks
• Desc one from each query type
• Measurement Difficulty and Usefulness
• Pilot Profile
• Researcher who practiced Neurology 15 years ago
  gt Researcher point of view
• Gynecology and Obstetrics doctor who practiced
  over 30 years gt Pure doctor point of view

15
Task 1 E (Event)
Task Find Patients who have had 3 WBC (White
Blood Cell) measurement.
16
Task 2 E-FT(Event - Fixed Time span)
Task Find Patients whose HGB (Hemoglobin)
measurement was between 10 14 Gm/dl (normal
person reference level) but 2 day later it
increased over 14 Gm/dl.
17
Task 3 E-VT(Event Variable Time span)
Task Find Patients whose HGB measurement was
under 14 Gm/dl, but within 3 days it decreased
under 14 Gm/dl.
18
Task 4 E-VT(Multiple Event Variable Time
span)
Task Find Patients whose WBC measurement is
above 8.50 K/ul and BASO (basophilia ) lab test
after 2 days
19
Task 5 - EWC-VT (Multiple Event with Window
Constraint Variable Time span)
Task Find patients who had 5 or more HGB within
3 days followed by MCH measurement.
20
Task 6 (EWC)F-VT (Multiple Event with
Functional Window Constraint Variable Time span)
Task Find patients who had 2 or more HGB test
within 2 days that are less than the reference
level (1014 Gm/dl) within a range of 3 Gm/dl .
21
Results

• Components
• Most Intuitive (1)(7)
• Confusing but useful (2)(3)(6)
• Not much useful (4)(5)

22
Results
Query Type Difficulty Usefulness Usefulness
Query Type Difficulty Practice Research
E (11)/2 5 1
E-FT (34)/23.5 8 2
E-VT (35)/24 8 8
E-VT (35)/24 7 7
EWC-VT (99)/29 2 7
(EWC)F-VT (99)/29 4 7
23
Conclusion

• Visualization
• Combinatorial display of results are redundant
• Query Power
• 3rd, 4th query are the most useful
• Consumes exponential time when query becomes
  complicated
• Medical Use
• Potential for Patient monitoring system in ER

24
Futurework

• Need for centralized patient data
• Data from emergency room was only result of
  blood, urine and cerebral spinal fluid result for
  a month period.
• Add various filters (currently the system has
  just two) to the search system

25
Special Thanks To..

• Professor Ben Shneiderman
• Professor James Reggia, Doctor Yong Song for the
  Usability Study
• Doctor Mike Gillam for providing the emergency
  room patient dataset and useful information

26
The End

• John Brennan (jbren_at_wam.umd.edu)
• Hyunyoung Song (hsong_at_cs.umd.edu)
• Nima Negaban (nimacn_at_gmail.com)