Data linkage using the National Hospital Morbidity Database

1
Data linkage using the National Hospital
Morbidity Database

• Penny Allbon
• AIHW

2
Structure of this presentation

• About the AIHW
• What sort of data linkage work do we do?
• Linkage within and with the National Hospital
  Morbidity Databases

3
About the Institute

• Established through AIHW Act 1987 as a Statutory
  authority of the Australian Government
• Governed by a Board, under the Commonwealth
  Authorities and Companies (CAC) Act
• Board categories of membership prescribed
• by legislation Minster appoints the
• Director, on recommendation of the Board

4
AIHW Board Members

• Chair Hon Peter Collins AM,QC
• Ministerial nominees
• Mr Ian Spicer
• Prof Heather Gardner
• Prof Sandra Eades
• Dr Greg Stewart
• Ms Libby Davies
• David Kalisch (DoHA) Brian Pink (ABS)
• Penny Allbon (AIHW)
• State/Territory Reps Louise York (Staff)
• Peter Allen (AHMAC) Observers
• Peter Smith (HMAC) Warwick Anderson (NHRMC)
• Bette kill (CDSMAC) Robyn McKay (FACSIA)

5
Required under the AIHW Act to

• Collect and produce and coordinate health and
  welfare-related information and statistics
• By itself, with ABS, or in association with other
  bodies or persons
• Develop relevant specialised statistical
  standards and classifications
• Subject to privacy considerations, enable
  researchers to have access to health and
  welfare-related information and statistics

6

• Conduct and promote research into the health of
  Australians and their health services
• Develop methods and undertake studies to assess
  the provision, use, cost and effectiveness of
  health services
• and technologies
• Publish methodological and
• substantive reports
• Report on the state of the nations
• health ( welfare) every two years

7
Strong privacy, confidentiality provisions

• Criminal penalties apply
• Can be extended by arrangement to collaborating
  institutions

8
AIHW data linkage protocol to protect privacy

• All linkage is subject to Ethics Committee
  approval, and approval of data providers as
  required
• Linkage uses purpose-specific linkage data sets
  that only include data required for establishing
  and checking links
• Analysis files do not contain identifying
  information such as name, DOB
• Intermediate data sets and project-specific link
  and record identifiers are deleted at the end of
  the project

9
Data linkage at the AIHW broad types

• Identifier-based names or name-based SLKs,
  with demographics (eg National Death Index)
• Event date-based, with demographics eg
  hospital-residential aged care linkage
• Using provider organisation identifiers, eg to
  link hospital datasets

10
Data linkage at the AIHW specific types

• Within health or welfare programs to convert
  event-based records into person-based records
  eg disability services, juvenile justice pathways
• Within programs to link data on events with data
  on provider organisations (eg hospitals)
• Between program linkage eg hospitals and
  residential aged care cancer registry and
  deaths PBS MBS

11

• To assess data quality in the data sets
• Methodological or ethical projects/activities
• (eg hospital morbidity data linkage project
  DoHA-funded Indigenous ID assessment)
• For external clients, eg National Death Index
  (about 50 so far this year), National Cancer
  Database

12
Linking events within and with the National
Hospital Morbidity Database

• A comprehensive national data source for admitted
  patient care (public private)
• No unique identifiers, or names
• ( what are the barriers to achieving this?)
• Feasibility of further probabilistic linkage
  studies
• assessment results

13
Feasibility Assessment

• 4 scenarios assessed
• aiming for false positive rate no more than
  2.5.. (assuming DOB-sex-SLA-date exact linkage)
• Measures of effectiveness
• False positive rate (estimated using probability)
• False negative rate (estimated by data comparison
  or linkage comparison)

14
1. Contiguous episodes

• Linkage feasible if
• Source data set limited to
• mode of separation of Transfer to another
  (acute) hospital or Statistical discharge
• Target data set limited to
• mode of admission of Admitted patient
  transferred from another hospital or Statistical
  admission
• Data quality an issue 6 more transfers out
  than transfers in

15
2. Linkage of Non-admitted ED presentations and
admitted patient episodes

• Linkage feasible if
• Source data set
• presentations to public hospital EDs resulting in
  an admission
• Target data set
• hospitals that are in scope for ED episode
  data
• Event date matching over a three day period

16
3. Non-contiguous records - readmissions

• Linkage may be feasible if
• Tight specification of source and target data
  sets eg a focus on particular diagnoses/procedures
  
• A maximum number of days for readmission
  specified
• True readmission rate is high enough
• E.g Should be feasible for principal diagnosis of
  asthma (32,000 separations in 2006-07) and 28 day
  readmission rate 4 (Ringland et al 2006)

17
Minimum readmission rate to achieve an effective
linkage (false match rate of lt 2.5)
18
4. Linking hospital data with mortality data

• Where death occurs in hospital, linkage with
  mortality data is feasible for
• Deaths for admitted patients - separation date
  DOD
• Deaths for non-admitted ED patients - end of
  episode DOD
• False positives less than 2 false negatives can
  be estimated, because linkage rate should 1

19
Deaths post discharge from hospital -

• Linkage not generally feasible as DOD not related
  to discharge date and true match rate (ie post
  discharge death rate) is too low
• Exception may be records in the NHMD for selected
  diagnoses/procedures with high expected mortality
  post-discharge

20
Summary of assessment

• Critical determinants
• Size of the two datasets
• Rate of true matches (eg post-hospital mortality
  rate)
• Degree of segmentation obtained by linkage
  variables
• Degree of uniformity of the segmentation
• Underlying quality of the data

21
Issues for reflection

• State/territory and Australian Government data
  custodian clearance process
• Names or IDs on the NHMD?
• State/territory ethics approvals
• Stakeholder consultation a major aspect of any
  new work