Clinical Data Mart CRADA

1
Clinical Data Mart CRADA
2
Clinical Data Mart CRADAAGENDA

• Background
• Vision
• Development of the CRADA
• Data Model
• VA production model
• (Kimballs fact and dimension tables)
• Constraints
• Performance criteria/testing (Intel, 4-16 to
  5-31, 2008)
• Use Cases-Initial Test Summary (1-17 and 22,
  2008)
• Clinical-Video
• Research-Video

3
AGENDA cont.

• Future Potential
• Next Steps
• Questions/Discussion

4
Background

• Former Assistant Secretary of Defense (Health
  Affairs) charge develop a Clinical Data
  Warehouse model flexible, scalable and modular
  enough to provide end-user access to data in the
  Clinical Data Repository and future additional
  data feeds
• Tasked Military Health System Chief Information
  Officer to partner with industry to leverage best
  practices
• Cooperative Research and Development Agreement
  (CRADA) established with Microsoft, Intel and
  Hewlett-Packard
• Executed under a Memorandum of Agreement among
  TRICARE Management Activity, US Army Medical
  Research and Material Command and Madigan Army
  Medical Center

5
Vision

• Create model to support multiple user-friendly
  toolsets for use by providers, administrators,
  analysts, and researchers
• Leverages data in the Clinical Data Repository
  for analysis and data transparency
• Provide a solution that is secure, scalable and
  adaptable to present and future needs
• Support standards based, Business Intelligence
  solutions from multiple vendors

6
CRADA Development

• TATRC Lead to coordinate CRADA Partners
• Military Health System
• Microsoft
• Intel
• Hewlett-Packard
• Medical Research and Material Command (MRMC)
• Madigan Army Medical Center (MAMC)

7
CDM CRADA CONOPS

• Provide a solution that is secure and
  protects PHI
• Be able to scale and adapt to present and
  future needs
• Support standards based BI solutions from
  multiple vendors
• Rapid development and deployment

8
Microsoft Healthcare Technologies
Researchers
Commanders
Data Warehouse
9
CONOPS Specific Requirements

• Provider scorecards
• CONOPS reports, clinical and administrative
• Patient registries
• Decision support
• Readiness scorecards and reporting
• Geographic Information System presentation and
  analysis (mapping)
• Interactive data analysis
• Research queries
• Data mining

10
Research Hypothesis

• Research goal was to determine whether new data
  model has any advantage to the end-user in terms
  of increased functionality, performance, or
  scalability, over what exists in the production
  CDM at DISA Oklahoma
• New CDM Prototype has DUA and IRB Approval and
  uses de-identified data

10
11
Prototype Development

• Microsoft and its consultants (and with VA and
  MHS participants) loaded 5 extract-transform-load
  (ETLs) for clinical encounters, lab, pharmacy,
  radiology, and vital signs under the original
  CRADA
• New data model based on Kimballs fact and
  dimension tables and production warehouse used at
  VHA
• TATRC secured de-identified data for use in the
  prototype, from a Northrop Grumman AHLTA test
  environment
• Microsoft loaded de-identified data for Clinical
  Encounters and Pharmacy data models (followed by
  3 other ETLs)
• Team installed HP hardware platform at Madigan

11
12
HP CDM Prototype Hardware
13
Prototype Development - continued

• Intel conducted scalability testing and provided
  report on technical platform to support 50,000
  clinical users
• Under CRADA extension, Madigan and Microsoft
  completed loading data from three remaining ETLs
• Microsoft and Madigan have completed a new
  Graphical User Interface (GUI) for clinician use
• Madigan completed Initial evaluation on 17 Jan
  2008
• Prototype and design to get input from
  prospective users
• New CDM prototype initial evaluation was not a
  direct comparison against current production CDM

13
14
CRADA Deliverables

• Security whitepaper from Microsoft ACES team
  and Intel security experts
• Scalability extensive scalability testing at HP
  labs recently completed and documented
• Proof of Concept Madigan Army Medical Center
• 4 years of de-identified data (155GB)
• 5 data loads (Patient Encounters, Labs, Pharmacy,
  Immunizations, Vitals)- delivered to EIDS

15
DATA Model

• VA Production Model-Kimballs Fact and Dimension
  tables
• http//www.decisionone.co.uk/resources/KimballArti
  cles.htm
• The Best Care Anywhere 25 years of data,
• Constraints
• 4 year data extract ending 11/2006
• De-identified data-scrambled demographics
• Limited comparison to existing CDM

16
Performance Testing

• 2 million/4 million records (Entry/Superdome)
• 50 Million/100 Million Encounters/Diagnosis/Proced
  ures
• 35-45,000/147,000K Transactions/hour
• Refresh Threshold 24 hrs/Objective 15 minutes
• Scale out vs. up
• (VHA model processes gt700B rows in 4 min, 18 Sec)

17
Use Case Clinician

• Video Link

18
Use Case Researcher

• Video Link

19
Evaluation results

• Evaluation Status
• Initial User Testing at Madigan, 1-17-2008
• Walk through of use cases and data exploration
• AAR of clinical and research groups
• Research deep dive 1-22-2008
• Intent-determine if model as designed could
  address user needs and questions raised
• Conclusion-no material weaknesses identified
  many comments gathered on user needs

20
Future Potential

• Harness collected data to
• Drive clinical decision support
• Guide business patient disease demographics,
  patient utilization patterns, transition to VA
  care
• Converge capabilities with VHA analytics

21
VHA Setting the U.S. Benchmark for Many
Comparable Indicators
22
Next Steps

• Implement the Kimball Model with current
• users requirements
• Refine based on use/feedback
• Incorporate additional data feeds (M2, MDR,
  UCAPERs, DHMRSi, Essentris GDR, SADR etc)
• Provide feedback to users on data input to
  improve data and output quality

23
Questions/Discussion