Search results

The application of linked data by government has limited visibility due to complexities in publishing details of the analysis and interventions. The Victorian Department of Health and Human Services (DHHS) acknowledges the value of linked data in the department's strategic plan, and linked data has been used in many policy reform and service design activities over the past three years.
IntroductionIn 2016 the Centre for Victorian Data Linkage (CVDL), located in DHHS, developed the Victorian Linkage Map (VLM) of 20 plus health and human services datasets linked with births and deaths data. The VLM has since been expanded to include education, justice and police data. DHHS has been an "early adopter" in applying linked data to policy development, service reforms and departmental operations to improve the health and wellbeing of the Victorian population.
Objectives and ApproachThe presentation will provide an overview of multi-sector data linkage in Victoria, and the value of frequent collaborations between CVDL and DHHS staff in applying linked data to priority projects and reforms. It will consider the challenges of cross-jurisdictional linkage in Australia, highlighting initiatives which are expanding linked datasets available to DHHS.
ResultsDHHS has undertaken a broad range of linked data projects which have provided an evidence base for departmental activities, including provision of hospital, housing, health surveillance and child protection services. Linked data has also been used to develop an integrated demand model, which forecasts impact of investment in one program area on other parts of the service system.
Conclusion / ImplicationsThe active use of linked data by DHHS provides a model for other government departments to improve service design and delivery to vulnerable populations. DHSS is expanding linked-data use to additional areas, and further imbedding in departmental operations.

The COMPASS program supports young people to successfully transition from Out of Home Care to adulthood. COMPASS is a social impact bond partnership between Victorian government, not-for-profit organisations and investors. COMPASS uses linked data in every phase from design, to implementation, to outcome measurement.
IntroductionStudies demonstrates that young care leavers experience significantly poorer outcomes than their peers. COMPASS is a preventative program progressivelyproviding 200 care leavers with access to housing and individualised support. The Centre for Victorian Data Linkage (CVDL) developed the Victorian Linkage Map (VLM) in 2016, linking 20 plus health and human services datasets with births and deaths data. Linked data provides a critical evidence base for COMPASS.
Objectives and ApproachThe presentation describes application of linked data for COMPASS design, implementation and measurement. Linked data cohort analysis of post-care service use of 6000 young people informed the program design. Linked data also provides the basis for a stratification tool to measure the complexity profile of referrals. Health, housing and justice payable outcomes are measured by comparing relative performance of participants with a matched control group using linked data.
ResultsImplementation of COMPASS has highlighted the value of linked data in service design and measurement of payable outcomes. It has also highlighted the challenges of using linked data in a real-world environment, including the need for thorough documentation and testing of specifications, calculations and processes.
Conclusion / ImplicationsThe use of linked data for COMPASS provides a model for evidence-based service design and tests the use of linked data for robust and sustainable outcome measurement. The lessons from COMPASS are applicable to other social impact bonds and service implementation and outcome measurement more broadly.

ObjectiveFamilial relationships can provide researchers with important insight into genetic, environmental, and social influences across many domains of research. While most administrative datasets do not collect information about relationships, familial linkage is an approach which seeks to identify such relationships among individuals within a linked data environment. We sought to develop a familial linkage resource which permits researchers access to relationship information otherwise not available in unlinked disparate administrative collections.
ApproachLeveraging birth and marriage registration data, we sought to identify relationships between Victorians. Familial links were formed by summarising relationships that were either explicit in the data (e.g. Parent and Child), or implied (e.g. A parent of a parent is a grandparent). Resulting relationship data was stored in a data asset which interfaces with our linkage infrastructure for easy access and use by linked data end-users.
ResultsThrough familial linkage, we have identified over 8.8 million unique relationships for Victorians, spanning 24 relationship subtypes which include both biological and non-biological connections. This data can be linked to all administrative datasets within our linkage environment, however representation varies across sources. Overall, the highest coverage of known relationships is found in datasets which specialise in child services, while older Victorians remain a gap.
Conclusion and ImplicationsFamilial linkage offers new dimensions of insight to researchers than what is accessible in source data alone. This information enables our data end-users to gain critical insights into the complex interplay between biological and social influences on Victorians' health and well-being.

ObjectiveDemand for real-time data during the COVID-19 pandemic revealed a need to increase efficiencies in manual linkage processes to respond to events in near real-time. In response, our jurisdictional linkage agency transitioned from yearly to daily, weekly and monthly linkage practices through increasing automation and improving process flows.
ApproachOur linkage agency transitioned to a fully automated process utilising scalable cloud infrastructure. Source data is now provided directly to a common data platform. This data is split into linkage and content, cleansed and quality assured in Python and set to automatically run via Azure data pipelines. The data is then linked via a combination of deterministic and probabilistic criteria, with data quality checks automatically performed along the way. Researchers can analyse this data in a secure virtual machine that only they can access and retrieve data from.
Results This infrastructure expedites the data linkage process allowing daily linkage results to select datasets, and enables advanced research such as a predictive micro-simulation model, which leverages the platform to predict and intervene on outcomes influenced by governmental policies. This model relies on timely administrative data to build targeted interventions for groups with poor future outcomes, tests these interventions, and monitors outcomes in near real-time.
ConclusionOur linkage agency has transitioned from manual to automated linkage processes in response to increasing need for timely data. By embracing cloud infrastructure and leveraging automation, we have streamlined our operations, enabling responsive linkage depending on need and expediting the provision of de-identified, linked data to researchers.

ObjectivePoor quality data confounds efforts to link clients across datasets. To combat this, we have trialed an approach which seeks to identify linkage candidates using associations with related services. We utilised a machine learning (ML) approach to query linkage candidates' dataset associations and make predictions about whether candidates are likely to be a client of the service being integrated for linkage.
ApproachUtilizing a K-nearest neighbors algorithm, we trained a model using 15 variables to predict whether a linkage candidate is a client of a family service dataset. We then evaluated the model's success, and modified its native performance to minimise false positive matches. Subsequently, we tested the validity of these predictions as linkage criteria by employing blocking strategies and linking the service dataset to our linkage spine. The trained model was then used to identify correct links against the spine.
ResultsThe evaluation of the machine learning model yielded promising results, with high accuracy (88.5%) and precision (95.5%). Testing the predictions as linkage criteria resulted in highly accurate links ranging from 96.0% to 98.7% across different blocking strategies. Despite some records failing to establish any links to the spine, rates of false positive matches remained low (0.7% to 3.1%).
ConclusionMissingness and inaccuracy in data remains a key problem for data linkage, and a robust approach is required to resolve complex linkage cases. However, these findings suggest that machine learning can present novel options for a toolbox of many approaches to link problem records to a linkage spine.

IntroductionRecent advances in data linkage infrastructure in Australia mean that data can be linked based on various identifiers across datasets. In a first for Australia, we tested the feasibility of linking data between a clinical quality disease registry with Australian and state government health data across multiple jurisdictions.
Objectives and ApproachTo determine whether high quality linked data for stroke can be obtained using a non-government managed registry (Australian Stroke Clinical Registry, AuSCR), national death registry data (Australian government), and hospital admission and emergency presentation data (state governments) to assess the accuracy of consistent variables across the different datasets. We used a cohort design with probabilistic data linkage to merge patient-level records. Descriptive statistics presented for matching concordance and Cohen's kappa for concordance across demographic variables. The sensitivity and specificity of in-hospital deaths collected in the AuSCR was assessed against national death registrations.
ResultsThere were 16,214 registrants in the study cohort. Their identifiers in the AuSCR from 2009-2013 were linked with death, emergency department and hospital discharge data from April 2004 to December 2016. In total, 99% of the AuSCR registrants were linked to one or more datasets; 98\% were linked with emergency presentation (80%) and/or admission (95%) data. Linkage to national death registrations identified 4,183 death; 1440 of these were identified as in-hospital deaths in both data sets demonstrating that in-hospital death classification in AuSCR had a 98.7% sensitivity and 99.6% specificity. Concordance between common demographic variables was excellent (kappa 0.84 for aboriginal status and kappa 0.99 for sex).
Conclusion/ImplicationsThe majority of AuSCR registrants were accurately linked to the Australian and state government datasets. Linkage quality was excellent and there was high concordance between common variables. The ability to reliably merge the datasets assures future comprehensive analyses of stroke care, ongoing health care resource utilisation and patient outcomes.

OBJECTIVES: To determine the feasibility of linking data from the Australian Stroke Clinical Registry (AuSCR), the National Death Index (NDI), and state-managed databases for hospital admissions and emergency presentations; to evaluate data completeness and concordance between datasets for common variables. DESIGN, SETTING, PARTICIPANTS: Cohort design; probabilistic/deterministic data linkage of merged records for patients treated in hospital for stroke or transient ischaemic attack from New South Wales, Queensland, Victoria, and Western Australia. MAIN OUTCOME MEASURES: Descriptive statistics for data matching success; concordance of demographic variables common to linked databases; sensitivity and specificity of AuSCR in-hospital death data for predicting NDI registrations. RESULTS: Data for 16 214 patients registered in the AuSCR during 2009-2013 were linked with one or more state datasets: 15 482 matches (95%) with hospital admissions data, and 12 902 matches (80%) with emergency department presentations data were made. Concordance of AuSCR and hospital admissions data exceeded 99% for sex, age, in-hospital death (each κ = 0.99), and Indigenous status (κ = 0.83). Of 1498 registrants identified in the AuSCR as dying in hospital, 1440 (96%) were also recorded by the NDI as dying in hospital. In-hospital death in AuSCR data had 98.7% sensitivity and 99.6% specificity for predicting in-hospital death in the NDI. CONCLUSION: We report the first linkage of data from an Australian national clinical quality disease registry with routinely collected data from several national and state government health datasets. Data linkage enriches the clinical registry dataset and provides additional information beyond that for the acute care setting and quality of life at follow-up, allowing clinical outcomes for people with stroke (mortality and hospital contacts) to be more comprehensively assessed.