Medical Hypothesis Testing

1
Medical Hypothesis Testing

• July 27, 2006
• Bill Bushey
• Emily Jenkins

2
Motivation

• We would like to provide a way to aggregate the
  collective knowledge that has been gained by the
  performance and known outcomes of thousands of
  procedures so that doctors can tap into this
  collective knowledge to quickly obtain
  information on the experiences of their peers.

3
Goals

• Our goal is to provide doctors with a broader
  view of the success of treatment options, beyond
  their own experiences and the experience of their
  colleagues.

4
Goals

• Ability to access multiple databases to search
  for records
• Output shows the effectiveness of procedures on a
  continuum
• Only relevant procedures will appear in the
  output
• Initially have a choice of searching for any of
  10 specified diagnosis, with the potential to
  easily expand that option

5
What We Started With

• Nothing!
• No access to any databases that may or may not
  exist
• ICD-9-CM coding system exists
• NHDS 2004 data

6
Prepping the Data

• Appended random outcomes to the records
• Load records into 1 database
• Delete records without a procedure attached to
  them (1/3), leaving over 200,000 records
• Drop excess diagnoses and procedures
• Load records into an additional 4 databases
• Add test diagnosis and procedure to 1 database
• Create specialty hospitals to increase the
  number of minority patients

7
Our Databases
Upstate Medical Center
Harvard Medical Center
University Virginia Health System
Stanford Medical Center
University Texas Houston Medical Hospital
8
How It Works

• Connects to specified databases
• Builds a query using the user specified criteria
• Searches for and returns records that fall under
  the search criteria
• The most commonly used procedures are displayed,
  along with their percent occurrence, mean
  outcome, and mean standard deviation of outcome

9
Demonstration
10
Testing

• Use added test diagnosis and procedure to verify
  the mean and standard deviation logic, as well as
  the search
• Command line program

11
Problems

• Initial
• What would the real database(s) look like?
• What data is needed
• Issue of the interdependence of procedures
  (likely more than one procedure performed per
  diagnosis, or multiple diagnoses)
• During
• Failure of machine containing the database
• Hypothesis testing using fuzzy logic
• No java graph package
• Auto-complete difficulties

12
Recommendations to theMedical Community

• Medical community should have a standard format
  for electronic records
• Primary diagnosis and primary procedure performed
  for that diagnosis should be specified
• Standards for evaluating patient outcome
• Changes in ICD-9-CM codes
• Program Usable ICD-9-CM file

13
Relics

• What well leave behind
• 3 page UML diagrams
• 5 page User Manual
• 1 cd of database data
• Change log
• 6 pages of testing documentation
• 17 pages of API documentation

14
Future Work

• Able to handle multiple diagnosis
• Multiple procedures for the same diagnosis
• Ability to print table or graph
• Secure access to databases as well as login
  information
• Better error handling
• Tabbed results with support for multiple searches