Making a Complex Data Repository Work: Nuts and Bolts

1
Making a Complex Data Repository WorkNuts and
Bolts

• Leslie L. Roos

1
2
Technological Innovations

• Technological developments have set the
• scene for information-rich environments
• Increased government data collection
• Powerful computers and programming languages
• Statistical methods for addressing large amounts
  of data
• Potential of these environments comes at a cost
  in complexity

3
Research Centres-Innovations

• Registry
• Record linkage
• Government cooperation
• File complexity
• Knowledge management

4
Manitoba Population Health Research Data
Repository
-standards tests -high school marks -graduation -r
etention
-income assistance -in care
-birthweight, gestation -injuries -chronic
diseases (asthma, diabetes)
Healthy Child Program Data
-diagnosis
-marital status
-residence
-family size

• meds
• dosage

Census Data at area level
National surveys
Key health databases start in 1970
5
Complexity

• Complexity increases with each file
• Each ministry has its own standards
• Both policy and coding change periodically
• Data may be missing for some fields and some
  years

6
Knowledge Management

• Knowledge management activities create readily
  retrievable information
• Enable these innovations to transform academic
  and policy-oriented research
• Support of researchers at Manitoba Centre for
  Health Policy website

7
Three Sections

• Research Protocol Critical steps in using
  Manitoba Repository
• Glossary Terms commonly used (coordinated with
  Centre deliverables)
• Concept Dictionary Detailed operational
  definitions of variables

8
Knowledge Resources and Mission
Project Initiation, Project Management
Knowledge Repository, Documentation, Internal
Training
Knowledge Sharing, Marketing
9
MCHP Pillars

• Try to manage knowledge so that more than one
  pillar is supported by any given activity
• Need for cost-effectiveness in documentation
• Help initiate and manage research

10
Repository Pillar

• Need to have organizational memory in
  considerable detail. May want to look up material
  after several years hiatus.
• Need for reflection on what has and has not
  worked.
• Synthesis of multiple sources of information over
  time.

11
Knowledge Translation

• Help build community of practice despite other
  differences
• For teaching and sharing information with other
  centres
• SAS videos

12
Knowledge Communities

• Great diversity of disciplines and collaborators
• Different data
• Different modeling techniques
• Knowledge sharing is essential

13
Manitoba Centre for Health Policy

• Magnet for new data sets
• Research partnerships maximize advantages
• Partnerships add to complexity

14
Example Complexity in Collaboration(6
Disciplines Represented)

• Local Investigators
• Noralou Roos, Les Roos, Randy Fransoo
• Local (but based in Ontario)
• Marni Brownell
• U of T / OISE
• Ben Levin
• U of T / School of Public Policy
• Marc Stabile
• Columbia University
• Janet Currie
• University of California, Berkeley and SF
• Doug Jutte

15
Complexity/Change Imposed on Organization

• New data every year
• Coding changes of varying magnitude
• Educationpolicy changes re data collection
• Educationmajor curriculum change in mathematics
• Intellectual advances (statistical techniques,
  research designs)
• New deliverablesusing data differently

16
Complexity re Longitudinal Information

• Health
• Education
• Social Assistance
• Family information
• Siblings
• Mother
• Family structure

17
Concepts

• Provide the basics
• Clarify choices of a particular strategy
• What can and can not be done
• Cross-reference for different ministries and
  topics

18
Activities Complexity and Change

• Research Protocol
• Users (particularly new users)
• New Statistical Methods
• Change
• Coding (ICD-9-CM to ICD-10-CA)

19
Ongoing Issues

• Longitudinal studies are great but tricky. Such
  research often involves several shifts in
  reporting conventions.
• Shift in perspective for survey researchers used
  to cleaned up, well-maintained data sets.

20
Knowns and Unknowns

• Knowns

• Problems approachable by sensitivity testing
• Example Biases inhospital collection of ER data

21
Knowns and Unknowns

• Communication of Known (by Some)

• Changing Formats over Time
• Example Variables miscalculated in complex
  cohort study using hospital and physician claims

22
Knowns and Unknowns

• Unknown

• Blending Information from Different Ministries
• Example Fiscal year (Health) and Calendar year
  (Education) when education data were weak

23
Ongoing Issues

• Costs -- How much is enough?
• Cost-sharing of a public good Project funding
  creates problem of free-riders

24
Ongoing Issues

• Business model must find ongoing balance among
• Bringing in new types of data/research
• Updates of data sets already held
• Documentation of all types
• Funding and new projects

25
M
Manitoba Centre for Health Policy
C
H
P
www.umanitoba.ca/centres/mchp/