Sage Journals: Discover world-class research

Abstract

Data linkage is a powerful tool for understanding the multifaceted needs and priorities of mental health care from the perspective of users and providers. Its potential remains underutilised in Australian settings – the Productivity Commission Inquiry into Mental Health in 2020 highlighted a significant gap: routinely collected administrative data are seldom leveraged in mental health research and service evaluation. In this manuscript, we provide insights into how data linkage has been used in mental health research, the type of questions that can be addressed, the steps involved in conducting data linkage research and the benefits and limitations of the use of this methodology. We propose crucial recommendations for advancing this field including: enhancing education for stakeholders (including the public, data custodians, ethics committees and policy makers); fostering stronger collaborative relationships with individuals with lived experiences throughout the research journey; improving infrastructure and resources for data linkage activities and linking data across sectors to address complex meaningful research questions. Data linkage is not just a method but a critical strategy to transform mental health research and service evaluation, ensuring more informed, effective and holistic mental health care.

Keywords

Mental health Australia data linkage Big Data mental health services mental health outcomes psychiatric epidemiology

Introduction

Data linkage involves the merging of two or more discrete data records pertaining to an individual and/or family (Andrew et al., 2016). In Australia, data linkage has a long history, dating back to 1995 with the establishment of the first data linkage unit (DLU) in Western Australia (Boyle and Emery, 2017; see Supplementary Material for a more detailed history). Data linkage has been used for a range activities including follow-up of cohorts or registry studies (e.g. Iorfino et al., 2023; Iveson et al., 2024), epidemiological and disease surveillance research (e.g. Fahridin et al., 2024; Papalia et al., 2024), health services research (e.g. Bradley et al., 2010; Fahridin et al., 2024; Young et al., 2020), economic analysis (Fahridin et al., 2024) and clinical trials (Fahridin et al., 2024). Such research can inform clinical and policy changes. For example, in Australia, world leading data linkage research identified the association between folate deficiency and neural tube defects (Bower and Stanley, 1989) and later demonstrated that folate supplementation in pregnant mothers could mitigate these defects (Bower et al., 2002).

The uptake of data linkage in mental health has been much slower than other disciplines. According to the Productivity Commission, administrative databases are rarely scrutinised for service evaluation and mental health research despite substantial data collection; this is an enormous, missed opportunity (Productivity Commission, 2020). Here we argue that data linkage is potentially a useful tool for mental health research. Specifically, we present: (1) some applications of data linkage in mental health research; (2) the processes involved with data linkage; (3) the advantages and disadvantages of data linkage (including challenges that may prevent uptake despite clear benefits) and (4) provide recommendations for future directions.

Exemplars in mental health

The Australian and New Zealand Journal of Psychiatry has published a range of data linkage studies; these demonstrate the breadth of research questions that can be addressed. Table 1 provides exemplars of such research, covering diverse areas of interest including health and mental health service use patterns (Borschmann et al., 2017; Cvejic et al., 2022; Ducat et al., 2013; Meurk et al., 2022; Young et al., 2024, self-harm and suicidal behaviours (Borschmann et al., 2017), psychotropic medication prescription (Young et al., 2024) and mortality (Borschmann et al., 2017; Meurk et al., 2022). Examined populations have included those with severe mental illness (Cvejic et al., 2022) or history of self-harm (Meurk et al., 2022), forensic populations (Borschmann et al., 2017; Ducat et al., 2013), general community (Hafekost et al., 2016) and Indigenous youth (Young et al., 2024).

Table 1.

Examples of Australian data linkage studies focusing on mental health published in the Australian and New Zealand Journal of Psychiatry.

Study	Research question	Cohort description	Jurisdictions covered in linkage	Consent	Linkage variables	Linkage method	Linked databases
Borschmann et al. (2017)	What is the rate of self-harm after release from prison?	1967 prisoners who were within 6 weeks of release from index incarceration	State-based	Written informed	Not clear in paper	Probabilistic	Emergency Data Collection (EDC)Queensland Hospital Admitted Patient Data Collection (QHAPDC)Consumer Integrated Mental Health Application (CIMHA)
Cvejic et al. (2022)	What are the differences in service use behaviours of individuals with and without a psychotic disorder?	Individuals aged between 12 and 100 years who were admitted to either a mental health or non-mental health unit with a diagnosis of mental health disorder	State-based	Waiver	Name (first and surname), date of birth, sex	Not reported but assumed to be probabilistic	Admitted Patient Data Collection (APDC)Emergency Department Data Collection (EDDC)Mental Health Ambulatory Data Collection (MH-AMB)NSW Registry of Births, Deaths and Marriages (RBDM)
Ducat et al. (2013)	Are convicted arsonists more likely to experience mental illness, substance use, personality disorders, and greater mental health service usage compared to other offenders and community controls?	Convicted arsonists: 1328 convicted identified in the Sentencing Advisory Council of Victoria (SAC)Community controls: 4830 Victorian residents from the Electoral RollOffenders (non-arsonist): 429 from community controls with criminal charges	State-based	Waiver	Name (first, surname and aliases), date of birth, age range and gender	Deterministic then probabilistic	Victorian Psychiatric Case Register (VPCR)
Hafekost et al. (2016)	What are the prevalence estimates for mental health disorders and service use patterns of Australian youth?	Young Minds Matter: The Second Australian Child and Adolescent Survey of Mental Health and Wellbeing	National	Guardian	Not clear in paper	Not clear in paper	Medical Benefits Scheme (MBS)Pharmaceutical Benefits Scheme (PBS)National Assessment Programme – Literacy and Numeracy (NAPLAN)
Meurk et al. (2022)	What are the pathways and outcomes of suicide-related calls to emergency services in Queensland?	Suicide-related calls to Queensland Police and Ambulance Services between February 2014 and January 2017	Queensland	Waiver	NameDate of birthAddress	Deterministic then probabilistic	Emergency Data Collection (EDC)Queensland Hospital Admitted Patient Data Collection (QHAPDC) Consumer Integrated Mental Health Application (CIMHA)Alcohol, Tobacco and Other Drugs Information System (ATODS-IS)Perinatal Data Collection (PDC); and the Queensland Death Register (QDR)
Young et al. (2024)	What are the rates of mental health service and psychotropic medicine use in Aboriginal children?	892 Aboriginal children aged 0–17 years living in urban and regional areas of New South Wales. Children had to be on same Medicare Card as their carer.	National	Guardian	Medicare number	Not reported but assumed to be deterministic	Medicare Benefits Scheme (MBS)Pharmaceutical Benefits Scheme (PBS)

Such studies can have impact and reach. For example, Ducat et al. (2013) examined Victorian public mental health service usage of the entire Victorian population of convicted arsonists (n = 1328, between 2000 and 2009) and was able to generate two controls groups: 1328 matched community controls (based on electoral roll) and 421 non-firesetting offenders. Contacts with mental health services and rates of psychiatric disorders (viz., substance use, mood and personality disorders) were much higher in the firesetters (37%) compared to the two other control groups (other offenders 29.3%, community 8.7%). The need for early detection and intervention for individuals who experience psychiatric disorders and who are at risk of firesetting behaviours was highlighted as of great public benefit. The researchers also stressed that not all firesetters had contact with psychiatric services and/or had a diagnosis, and more research was needed to characterise firesetters. Even though this paper is over 10 years old, in 2024, it was cited across 55 news outlets regarding the rates of deliberately set bushfires within Australia (Altmetrics, database for tracking online attention for a study). Altmetrics has highlighted that is the ninth highest scoring paper in terms of receiving online attention of 2654 papers published in the Australian and New Zealand Journal of Psychiatry.

In other journals, a range of Australian data linkage studies have been recently published including studies examining the relationship between maternal mental health and adverse birth outcomes in the Northern Territory (Dadi et al., 2024), suicide and mortality in Culturally and Linguistic Communities in Victoria (Pham et al., 2024), prediction of psychiatric hospitalisations in young people discharged from New South Wales criminal justice services (Akpanekpo et al., 2024) and follow-up of children and adolescents with a history of suicide-related contact with police or ambulance services in Queensland (Wittenhagen et al., 2024).

There are international studies that demonstrate how data linkage can address pertinent societal issues. The impacts of COVID-19 on mental health, health and mortality outcomes is one example (e.g. Yu et al., 2023a; Yu et al., 2023b). Another recent example is research examining the possible association between social media and poor mental health outcomes, particularly for young people (e.g. Bye et al., 2023; Bye et al., 2024). Such research has the potential to lead to changes in service delivery, resource allocation and policy.

Thus, using data linkage, a broad range of questions can be asked for either specific groups or the general population for a wide variety of outcomes (e.g. pathways through care, service use, medication use, self-harm and mortality).

Data linkage methods

Our research team is currently undertaking a range of data linkage studies in the youth mental health space. This includes focusing on population-level patterns of health and human service utilisation in youth people with mental health problems (NHMRC Partnership Grant GNT#1198696), follow-up of specific clinical cohorts (Cotton et al., 2022) and evaluations of specific care models (e.g. youth-specific Hospital-in-the-Home model, HCF Health and Melbourne Health). In undertaking these activities, we have operationalised the steps involved in planning a data linkage project. The steps for setting up a data linkage project include (1) study planning; (2) institutional ethics, governance and data custodian approvals; (3) linking and data preparation; (4) data cleaning and analysis and (5) reporting of findings (see Figure 1).

Figure 1.

Activities and processes involved in data linkage.

Planning

Clear aims and research questions should be identified and assessed for feasibility given constraints of data (e.g. missingness). Developing relationships with key stakeholders is critical for these activities (Bradley et al., 2010; Downs et al., 2019). DLUs can guide researchers in identifying strengths and limitations of data sets, assessing technical feasibility of the linkage, navigating approvals and statutory requirements and informing protocol content for ethics submissions. Engagement with consumers, clinicians, advocacy groups, health service planners and commissioners, as well as other government bodies is strongly recommended (Downs et al., 2019). Lived experience and consumer involvement is commonly overlooked but can improve the relevance of work and aid protocol development and selection of outcomes (Jewell et al., 2019). Meaningful and authentic engagement with consumers is becoming increasingly recognised by both researchers and Australian grant funding bodies as critical component to advancing medical research (National Health and Medical Research Council and Consumers Health Forum of Australia, 2016). In the early stages of our research, we have consulted with young people with mental health problems to ensure that our research questions are meaningful to them, to gauge their understanding of data linkage and the ethical use of personal, sensitive and health data.

Institutional ethics and governance

Data linkage research requires ethics approval from a local Human Research and Ethics Committee (HREC) as well as approvals from DLUs and each data custodian (often coordinated by the DLU). Some Commonwealth linkage authorities (e.g. Australian Institute of Health and Welfare, AIHW) require a separate, second ethical approval from their own in-house HREC. It is imperative that researchers do their groundwork early and engage with appropriate stakeholders to inform ethics and governance applications. Consideration of principles of research merit and integrity, justice, beneficence and respect is essential (National Health and Medical Research Council et al., 2023). There are also specific principles and legislative requirements for data linkage that are covered in the Principles for Accessing and Using Publicly Funded Data for Health Research (National Health and Medical Research Council, 2015; see Table 2). If a data linkage study involves Indigenous peoples and communities, it must be Indigenous-led or conducted in directed collaboration with Indigenous researchers. The following material should be reviewed: the Ethical Conduct in Research with Aboriginal and Torres Strait Islander Peoples and Communities (National Health and Medical Research Council, 2018) and the Australian Institute of Aboriginal and Torres Strait Islander Studies (Australian Institute of Aboriginal and Torres Strait Islander Studies (AIATSIS), 2020). More recently there has been a discussion paper focusing on Indigenous data governance and sovereignty (Lowitja Institute, 2024). This discussion paper highlights the rights of Indigenous peoples to control the way that their data are collected, accessed and used to ensure that their priorities, values, cultures, views and diversity are embraced. We recommend that researchers consult the relevant groups such as the Lowitja Institute, if their work involves Indigenous peoples and communities.

Table 2.

The three principles for data linkage research as outlined in the National Health and Medical Research Council’s publication on principles for accessing and using publicly funded data for health research.

Principle	Researchers’ considerations
Maximise the use of publicly funded health and health-related data for research.	It is essential to highlight the utility of data linkage projects aimed at serving the public interest and benefit, such as enhancing disease prevention and clinical care. However, the process must prioritise protecting individual privacy and confidentiality (e.g., applying procedures that ensure data cannot be used to identify or re-identify individuals) and maintaining public trust in the use of personal data for research. Additionally, ongoing consumer involvement is vital to maximise the public benefit of data linkage projects.
Data custodians should recognise their responsibilities and accountabilities when providing access to data for research.	When providing access to data for research, custodians should facilitate and support data access while ensuring privacy and confidentiality, specify access conditions and provide timely data access for approved applications. They were also recommended to harmonise overall processes and maintain transparency about ethics requirements, application processes, timelines and any associated charges. In data linkage projects, particularly those involving cross-jurisdiction data, researchers should be aware of the complex processes required by individual data custodians and DLUs and design their study timeline and budget accordingly.
Researchers should recognise their responsibilities and accountabilities when accessing and using publicly funded health and health-related datasets.	Researchers should consult with data custodians early, comply with ethical and legal requirements, ensure data security and confidentiality and meet all obligations under legislation and agreements. Researchers must acknowledge data sources, declare conflicts of interest and be transparent about data protection, funding, dataset quality and compliance with all relevant regulations. Data custodians may require reviews of research findings and reports prior to publicly release, and researchers should also meet those obligations.

Consent-related issues are a specific concern for data linkage research. Addressing these issues requires clear policies, robust privacy protection protocols, transparent communication and opportunities for individuals to exercise control over their data (da Silva Marinho et al., 2012). Data linkage involves access to personal identifiers, such as name, date of birth (DOB) and address. Accordingly, this has sparked debate regarding the need for formal prospective consent processes (Bradfield, 2022; da Silva Marinho et al., 2012; Palamuthusingam et al., 2019). When individuals present to health and human services, large amounts of routine and personal data are typically generated, collected and compiled into datasets. Often information about data use is mentioned in fine print, so even though individuals technically consent to their data being used, consent is not necessarily informed (Boyle and Emery, 2017; Bradfield, 2022). Often, the primary purpose of this data collection is for service monitoring, improvement and/or quality assurance. Data linkage, however, involves secondary data use beyond the original reason for collection, raising concerns of contextual integrity (how privacy is maintained when information flows consistently with social context’s norms and expectations) (Nissenbaum, 2009). There are various approaches for how to address the issue of informed consent. Two methods of informed consent include opt-in and opt-out methods, with the opt-in method being more frequently used (de Man et al., 2023). For data linkage involving registries, prospective informed consent is usually required. Opt-out informed consent can be used where individuals are included in the dataset unless they express their desire not to participate ((Andrew et al., 2016; Australian Commission on Safety and Quality in Health Care, 2008). However, this option is complicated in that some ethics bodies may require opt-in consent for registry data linkage (Andrew et al., 2016).

Obtaining informed consent for the data use and linkage can be impractical, labour-intensive and costly (Bradfield, 2022; Palamuthusingam et al., 2019). For de-identified data, it would be difficult to justify re-identifying individuals to seek consent (Bradfield, 2022). Seeking consent is not possible for those who are deceased, or in situations where services do not have individuals’ most recent contact information. Obtaining informed consent, especially through opt-in methods, can also introduce response biases due to inherent differences between those who do and do not consent, which ultimately impacts the external validity of the research (Bohensky et al., 2011; Bradfield, 2022; de Man et al., 2023).

As an alternative to informed consent, a waiver of consent approach has been suggested (Bradfield, 2022; Downs et al., 2019). A HREC must approve a waiver of consent. In Section 2.3 of the National Statement of Ethical Conduct in Human Research (National Health and Medical Research Council et al., 2023), the guidelines for a waiver of consent are provided. For a waiver of consent to be granted, researchers must justify that the linkage study is of low risk (e.g. use of de-identified data), there are minimal harms associated with not seeking consent, that obtaining informed consent is impracticable and that it is not unreasonable to expect that individuals would have provided consent. There must also be no jurisdiction laws prohibiting a waiver and researchers need to demonstrate that there are effective processes in place for protecting individuals’ privacy and confidentiality (e.g. data de-identification, use of secure storage servers) (Bradfield, 2022; National Health and Medical Research Council et al., 2023; da Silva Marinho et al., 2012; Palamuthusingam et al., 2019). We highly recommend that researchers explicitly address each of these points in applying for a waiver of consent.

Linking and data preparation

The procedures and pipelines of data linkage research vary according to the project’s design and requirements (see Figure 2). The ‘separation’ principle and Privacy-Preserving Record Linkage (PPRL) are used to protect privacy by separating roles of DLUs, data custodians and researchers (Vatsalan et al., 2017). The Australian Bureau of Statistics (ABS) and the AIHW now adhere to the ‘Fife Safes’ framework which involves a multidimensional approach to balancing the interface between disclosure risk and the utility of data (https://www.abs.gov.au/about/data-services/data-confidentiality-guide/five-safes-framework). Safety and risk are evaluated in terms of people involved, project purpose and details, the access environment, data protection and outputs.

Figure 2.

Data linkage pipelines.

Researchers provide identifiable information (e.g. names, DOB, sex), but not content or clinical data, to data custodians or DLUs. In Australia, a unique identifier across datasets (e.g. Medicare number) would facilitate easier and accurate linkage; however, no such option is available across health and human services. As a result, DLUs must develop statistical linkage keys (SLKs) based on personal identifiers (e.g. letters of surname and given name, DOB and sex) (Andrew et al., 2016; Boyle and Emery, 2017; Harron et al., 2016). DLUs use the SLK to identify individuals across databases.

The goal of linkage is to achieve a true match between two records belonging to the same individual, family or event (Harron et al., 2016). It is not possible to quantify the number of true matches that occur, but a link status (linked or non-linked) can be delineated (Harron et al., 2016). Two different linkage methods, deterministic and probabilistic linkages, can be used depending on the requirements and characteristics of the project. Deterministic (or rule-based) linkage requires exact matches based on precise agreement between two records using predefined criteria and identifying variables (Doidge and Harron, 2018; Downs et al., 2019; Harron, 2022; Harron et al., 2016). Probabilistic linkage, involves the use of a weighting system based on statistical theory (e.g. odds that two records are from the same person) to delineate whether there is identifier agreement; higher weights signify greater agreement between identifiers (Doidge and Harron, 2018; Harron, 2022; Harron et al., 2016; Randall et al., 2013). Given unique identifiers are unavailable in Australia, dataset linkage mainly utilises probabilistic techniques (Tran et al., 2017). Once linkage has been finalised, the DLUs strip any identifiable or re-identifiable data fields from the datasets. The de-identified linked data are then returned to the project team. In this way, researchers are never able to view a person’s complete record across multiple data sources.

DLUs have a high degree of responsibility for protection, storage and use of data (Bradley et al., 2010). Datasets are stored on secure cloud-based servers to reduce data privacy and safety risks and enable regulatory control about data access (e.g. AIHW uses Secure Unified Research Environment [SURE]). Researchers complete all statistical analyses on these servers. Any outputs that are taken off the server are checked to ensure compliance with confidentiality and privacy requirements.

Data cleaning and analysis

Administrative datasets have been criticised for inconsistency in structure, format and content (Harron et al., 2017a; Harron et al., 2016). A systematic approach to preparing and cleaning the data is required to improve quality (Guo and Chen, 2023; Randall et al., 2013; Tran et al., 2017). This involves adjusting, eliminating or modifying data fields (Randall et al., 2013). Appropriate documentation and transparency in the adopted approaches are recommended (Gilbert et al., 2018; Tran et al., 2017). Data custodians, clinical experts and other individuals involved in data entry should be consulted to improve the robustness of data cleaning. Tran et al. (2017) provided a data cleaning and management protocol for cross-jurisdictional linkage of perinatal records from two states in Australia, linking other state-based data and national pharmaceutical dispensing data. They proposed 22 steps to examine uniqueness of records, consistency of reporting variables within and across datasets and identify cross-jurisdictional movement. Such procedures are important for identifying duplicates, mismatches, missing and inconsistent data. While data cleaning is time-consuming – estimated to take up to 70% of analysis time in a data linkage study (Christen and Goiser, 2007) – systematic approaches are essential to ensure statistical analyses and research findings are robust and generalisable.

Statistical analysis plans should be developed a priori, with appropriate documentation and record-keeping for replication. Code and algorithms should be openly available for how the target population was delineated, data files linked, whether a blocking step was used in matching (comparisons applied to records grouped together on blocking variables, increasing manageability of data and reducing number of comparisons), how the success of the data linkage was assessed (including rates of linkage biases), data cleaning, methods of analysis and any sensitivity analyses (i.e. impact of missing data, representativeness of the population, testing for choice of threshold or matching rules, controlling for linkage error) that have been conducted (Benchimol et al., 2015; Boyd et al., 2015; Gilbert et al., 2018; Harron, 2022; Harron et al., 2017a).

Research questions drive statistical technique choice. A range of frequentist statistics can be used such as simple descriptive analysis, regression models (e.g. linear, logistic, Poisson, LASSO regression models), survival models (e.g. Cox survival model and recurrent event survival model), longitudinal models (e.g. linear mixed effects model, Generalised Estimating Equation [GEE]) and Markov modelling (Haneef et al., 2022). However, given the size and breadth of linked data, advanced machine learning techniques including prediction models, clustering techniques, pattern mining and other advanced models such as large language models (LLM), deep learning and artificial intelligence algorithms can be exploited (Guo and Chen, 2023; Haneef et al., 2022). There has also been work related to statistical methods for causal inference with observational data, such as those applied to emulated trial designs (viz., mimics an randomised controlled trial [RCT] design but uses observational data to contrive groups) (Hernán et al., 2022; Szmulewicz, 2024).

Due to data sensitivity, both data cleaning and statistical analyses must be conducted within the secure server environments of DLUs, often associated with significant costs. At the time of writing, fees for one year’s worth of access to the AIHW’s SURE database for a small-to-medium sized project (approximately 10GB in size) for one researcher is $2674 AUD per annum. One proposed solution is to construct a dataset based on a synthetic population (Nicolaie et al., 2023). A synthetic population has all the characteristics of the available population data, but does not comprise data on real persons (Nicolaie et al., 2023). Data analysts can develop code and algorithms on the synthetic data and apply them to real data on secure servers, maintaining privacy, reducing the need to access real data and therefore minimising costs (Nicolaie et al., 2023).

Reporting and dissemination of findings

Various guidelines have been developed for reporting of research findings for specific research designs and to improve the quality, completeness and rigour of research (Simera, 2014). The STrengthening of Reporting of OBservational studies in Epidemiology (STROBE) (von Elm et al., 2007) guidelines closely align with observational data linkage studies (Benchimol et al., 2015), but they do not cover issues pertaining to dataset access, appropriate characterisation of linked data and data linkage techniques (Nicholls et al., 2015). Specific guidelines for data linkage have been developed, including by Bohensky et al. (2011) who, using the Delphi consensus technique, developed guidelines to promote better understanding and transparency in the processes and possible limitations of data linkage. The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) guidelines aims are an extension of STROBE (Benchimol et al., 2015; Nicholls et al., 2015), adapted also for pharmacoepidemiology research (ROUTINE-PE; Pratt et al., 2020). The CONSORT-ROUTINE guidelines (Kwakkenbos et al., 2021) have been recommended for randomised controlled trials that use routine outcome data, while the GUidance For Information about Linking Data (GUILD; Gilbert et al., 2018) guidelines can be used to improve the quality of reporting of data processing. Methodological guidelines for the use of machine learning techniques with linked data to estimate population outcomes have also been developed (Haneef et al., 2022). Use of any of these guidelines in the reporting of findings is strongly recommended.

Advantages and disadvantages of data linkage

Data linkage is a rapidly growing field; with this growth there is better understanding of the advantages and disadvantages associated with this methodology. It is important to consider these factors in planning a data linkage study (see Figure 3 for a summary of these advantages and disadvantages).

Figure 3.

Advantages and disadvantages associated with data linkage.

Advantages

Complex and meaningful information

Based on the richness of data collected through data linkage, especially across multiple datasets, more complex, holistic and meaningful questions can be addressed than can be achieved using single-source data (Andrew et al., 2016; Bradfield, 2022; Bradley et al., 2010; Fahridin et al., 2024; Harron, 2022). Linking health-related data with sociodemographic data provides insights into social determinants of health and mental health (e.g. (Cybulski et al., 2024; Marino et al., 2022). Retrospective historical data can allow for evaluating long-term trends in service use and outcomes (Dibben et al., 2016). Control groups or counterfactuals can be determined post hoc (Dibben et al., 2016), and linked data can be more extensive than that available through prospective cohort studies (Andrew et al., 2016; Brownell and Jutte, 2013). Data linkage can also be used as an alternative to a RCT. For example, we are using an emulated trial design to examine the effectiveness of a hospital-in-home as an alternative care model to inpatient care for young people with acute mental illness (HCF Research Foundation, Melbourne Health). Because of logistic and safety issues a RCT would not be appropriate for addressing whether the two care models differ.

Reduced burden and intrusion

Linkage of routinely collected data and/or follow-up of cohorts can be burdensome and intrusive to the individual participant (Boyd et al., 2015; Bradley et al., 2010; Harron, 2022). A key benefit is that it avoids duplication of effort and has minimal impact on individuals (Andrew et al., 2016; Boyd et al., 2015; Bradley et al., 2010; Dibben et al., 2016).

Minimisation of certain biases

Longitudinal studies involving follow-up of participants, can lead to biases in the data and participant-related biases in those consenting and those retained at follow-up (Downs et al., 2019). If the research involves sensitive topics (e.g. child maltreatment, suicide), then there are risks of social desirability and recall biases due to stigma (Brownell and Jutte, 2013) ultimately threatening the validity of research findings (Brownell and Jutte, 2013; Downs et al., 2019). Data linkage studies reduce and minimises these biases as routine data collection ensures that most individuals within a service are followed up (Brownell and Jutte, 2013; Downs et al., 2019; Fahridin et al., 2024), thereby enhancing the generalisability of the research undertaken in this field. It is also worth acknowledging that data linkage does not reduce all biases as individuals can be lost to follow-up or missing for other reasons (e.g. moving states, death, not being in contact with services); some of these issues are described under Disadvantages.

Cost-effectiveness

Data linkage is a cost-effective alternative to prospective studies and large-scale surveys, as it does not necessitate the resources, labour and time required to recruit participants to research (Andrew et al., 2016; Boyd et al., 2012; Bradley et al., 2010; Brownell and Jutte, 2013; da Silva Marinho et al., 2012; Dibben et al., 2016; Downs et al., 2019). The time taken could be shorter than a 5 year trial and could improve time to knowledge translation. Costs are mainly associated with protocol development for ethics and governance applications, data matching and extraction, computation, data storage, access to secure environments, cleaning, data analysis and importantly, stakeholder engagement.

Disadvantages

Challenges of ethics, governance and legislation

Data linkage involves navigating complex legal and ethical frameworks, including consent requirements, data ownership rights and regulatory compliance (Andrew et al., 2016). These matters can be further complication by any research involving vulnerable populations (e.g. individuals with mental disorders) and when there is use of very sensitive data (e.g. sexual or child abuse, criminal activities). There is ambiguity in interpreting legislations, fragmented policies around data use and often outdated and cumbersome legal requirements (Andrew et al., 2016; Brownell and Jutte, 2013; Carson et al., 2020; Olver, 2014; Palamuthusingam et al., 2019).

Within the Australian context, datasets can be based in different jurisdictions (e.g. State vs Commonwealth), which means that there are varied mechanisms for seeking approvals, and distinct legislative and governance requirements (Andrew et al., 2016; Olver, 2014). Cross-jurisdiction data linkage adds ethical issues relating to privacy and confidentiality, including sharing information with third parties such as sharing identifiers and data between State and Commonwealth departments (Andrew et al., 2016). Resolving such issues, while adhering to ethical standards, can be time-consuming and resource-intensive. There is often duplication of effort and significant delays between obtaining approvals from data custodians and ethical bodies to eventually receiving the data for analysis (Andrew et al., 2016; Boyd et al., 2012; Fahridin et al., 2024; Ritchie et al., 2015). It has been suggested that ethics approval processes should fall in line with changes that have occurred with multisite clinical trials; there is one ethics committee providing oversight and the addition of sites is through amendments rather than new applications (Carson et al., 2020).

Use of secondary data and data quality

Data linkage involves the secondary use of administrative data and consequently, there are concerns about data quality (Boyle and Emery, 2017; Harron et al., 2017b; Harron et al., 2016; Palamuthusingam et al., 2019). First, administrative data are often referred to as inflexible as the researcher has no control over how or what data are collected (Boyd et al., 2012; Boyd et al., 2015). In most cases, many confounders are not collected as a part of administrative records (e.g. education level, gender and/or sexual orientation, race, ethnicity, smoking status). There can be important clinical and mental health related variables that may be also not collected (e.g. exposure to trauma, family history of mental illness). Data may also not be fully aligned with research questions being asked.

Second, there are often inconsistencies in the structure, format and content across datasets and/or jurisdictions (Bohensky, 2015; Harron et al., 2016; Olver, 2014). Lack of standardisation in how variables are defined and collected (e.g. variability in how employment and education data are collected) can occur. Inconsistencies and inaccuracies in the recording of mental health data can also be common. In our own data linkage activities with a cohort of individuals who had received treatment of a first episode of psychosis, we reviewed Coroner’s reports where psychiatric status and/or diagnoses were not often reported, multiple improbable diagnostic combinations were assigned (e.g. both schizophrenia and bipolar disorder) or the diagnosis did not match data from other administrative databases. The accuracy of recording of psychotic disorder diagnoses in administrative data; however, has been noted to be higher than for anxiety disorders (Davis et al., 2016). Accurate recording of timing of events can also be a problem. For some medical events, such as stroke or cardiac events, one would anticipate that hospital records would accurately reflect the timing of the event. For mental health, the timing of events (e.g. disorder onset) is often clouded and difficult to ascertain in medical records (Clements et al., 2015).

Third, there can be missing data. Missingness can be due to non-response, data entry errors, data integration or corruption issues, temporal variations in the way data are collected and privacy and confidentiality rules forcing some elements to be deliberately anonymised. There may be some mental health events that are also not recorded because the individual has not presented to services; suicide attempt data may be missing as not all events result in hospitalisation (Adhikari et al., 2020). Intervention and treatments for mental health conditions are often poorly recorded across data sources (Yu et al., 2023b). Data can also be missing for minority groups such as Indigenous communities due to inequitable access to services or issues such as misspellings of names (Bohensky, 2015; Bradfield, 2022; Harron et al., 2017a). Under-identification of Indigenous status in administrative datasets can result in either under- or over-estimation of Indigenous health indicators especially for conditions that are related to significant stigma such as sexually-transmitted infections (Thompson et al., 2012). Therefore, missingness can introduce biases into the analyses especially if the missingness is associated with non-random events (Downs et al., 2019; Harron et al., 2017a; Harron et al., 2017b; Harron et al., 2016).

Fourth, there can be inaccuracies in the way that the data are entered, processed or collected. As most health service data (e.g. hospitalisation records) were collected for management, auditing and billing purposes, accuracy of clinical coding may be impacted by many factors, such as optimising reimbursement and quality of clinical documentation (Ryan et al., 2021). Fifth, there can be duplicates in the data which can affect analysis. Sixth, there can be issues with the timeliness of data. Sometimes there can be delays in which datasets are compiled, and linked datasets may not match temporally.

DLUs are driving ongoing work to improve the quality of data collected and to minimise the impact of these issues (Harron et al., 2016). Researchers need to carefully plan their approaches to these issues, conduct data cleaning and standardisation and provide appropriate documentation of their approaches (Tran et al., 2017).

Biases in matching

Related to data quality are potential biases in matching processes. Matching biases, even when minimal, can reduce confidence in the accuracy of data and generalisability of research findings (Bohensky et al., 2011; Downs et al., 2019; Harron, 2022). Errors in matching can result from discrepancies in how identifiers are recorded or due to missing data (Bohensky, 2015; Boyd et al., 2012; Gilbert et al., 2018; Harron, 2022). Variations in names (e.g. typographic errors, cultural and tribal names, maiden vs married surnames) can increase the risk of linkage errors (Tahamont et al., 2021; Tibble et al., 2018). The mismatching of ethnic minorities and Indigenous people is more likely to occur because of misspelling of names, inaccurately recorded birth dates, residential instability and geographic mobility (Downs et al., 2019; Gilbert et al., 2018; Thompson et al., 2012).

Biases in data matching can also be due to the choice of matching method (Bohensky, 2015). Because of the inflexibility of deterministic matching methods, there is a greater likelihood of mismatching and false negatives (Downs et al., 2019; Haneef et al., 2022; Lariscy, 2011). Probabilistic linkage techniques are more robust against errors and more adaptable when there are larger datasets (Randall et al., 2013).

The application of the ‘separation principle’ can also increase the likelihood of linkage biases (Gilbert et al., 2018). Because of the distinction between the roles of linkers and analysts and what information is available to them, it may be unclear as to whether specific groups are at greater risk of linkage error (Gilbert et al., 2018). DLUs should work on providing researchers with a detailed report on data linkage error so researchers can develop strategies to deal with them in the analyses. Assessing the potential for data linkage errors should occur during data cleaning.

Barriers with cross-jurisdiction linkage

Cross-jurisdiction linkage is crucial for understanding national health outcomes and reducing biases due to the increasing mobility of the Australian population, but limited infrastructure and legal/regulatory challenges hinder its use (Andrew et al., 2016; Boyd et al., 2012; Boyd et al., 2015; Fahridin et al., 2024). Currently, the environment and infrastructure to support cross-jurisdiction linkage is limited, and researchers face several challenges (Lloyd et al., 2024; Palamuthusingam et al., 2019).

These challenges first include variability in legal and regulatory compliances and in jurisdictions’ interpretation of laws; difficult issues to navigate can delay approvals for years (Andrew et al., 2016; Lloyd et al., 2024; Palamuthusingam et al., 2019). Second, duplication of effort is a problem, with different applications and procedures required in each jurisdiction (Andrew et al., 2016). Third, there are differences in data quality and lack of data standardisation (Olver, 2014). Fourth, technical and infrastructure issues may arise due to different software and platforms used across jurisdictions, requiring significant technical expertise and resources. Fifth, there may be privacy issues with release of data with identifiable information across jurisdictions (third-party) for linkage (Andrew et al., 2016; Boyd et al., 2015). Finally, data governance and ownership issues require careful consideration including clear governance frameworks and agreements to facilitate data access.

Despite these challenges, there are successful examples of cross-jurisdiction linkage such as Boyd et al. (2015) who linked over 44 million records from state-based hospital admission data across four Australian states to examine national mortality records. The accuracy of data linkage was determined to be high (Boyd et al., 2015). Andrew et al. (2016) have also shown how registry data (Australian Stroke Clinical Registry, AuSCR) could be linked across jurisdictions to hospital and mortality datasets.

Government agencies are also in the process of developing large data assets integrating data from across sectors and jurisdictions. Individuals are encouraged to monitor and check what mental health–related data are contained in these assets. The Person Level Integrated Data Asset (PLIDA) comprises health, education, government payments, income, employment and Census data as well as including survey responses to the 2022 National Health Study of Mental Health and Wellbeing. Another asset is the National Health Data Hub (NHDH) which is a national health data linkage system comprising data from states and territory as well as the Commonwealth. There are plans to further add other mental health datasets to NHDH including the National Community Mental Health Care Database and National Residential Mental Health Care Database.

Moving forward

Data linkage research is both necessary and for the public good (Bradley et al., 2010). Holman et al. (2008) predicted that ‘health data linkage systems may become normal infrastructure for health services research in nations like Australia within the next decade’ (p. 767); however, the rate of uptake has been slower, particularly in mental health (Productivity Commission, 2020). This is at the opportunity cost for Australian mental health research. The Productivity Commission (2020) suggested the imperative for existing data to be used, rather than extending resources to collect new, but similar data, which may be more likely to be incomplete and at greater burden to individuals. We also advocate that using data linkage allows for a broad range of research questions to be addressed that would otherwise remain unanswered. In advancing the adoption of data linkage in research activities, several approaches are needed.

Greater awareness and education across stakeholder groups regarding data linkage and its benefits is required. The development of a social licence for data linkage has been suggested and involves (1) ensuring there are shared values of the importance of such work across stakeholder groups; (2) empowering individuals with control over their own data and how it is utilised; (3) building public trust that safeguards are in place to ensure that data is well-protected and (4) making sure that there is genuine transparency and accountability of the processes involved in data linkage (Muller et al., 2021; Productivity Commission, 2020). Without a social licence for data linkage, there may well be enduring challenges and contestation from the public (Carter et al., 2015). However, there are no guidelines to support how a social licence would be developed and implemented.

While raising public awareness is crucial, the education of ethics committees and policy makers is equally important (Bradfield, 2022; Brownell and Jutte, 2013; Tan et al., 2015). Training programmes have been established and successfully implemented to build confidence among ethics committee members in their ability to review data linkage applications (Tan et al., 2015). Researchers are encouraged to share their knowledge and experiences in the literature, focusing on topics such as risk management (Ritchie et al., 2015). In addition, educational toolkits can be developed and used for advocacy and policy discussions (Ritchie et al., 2015).

Building strong relationships between stakeholder groups is crucial for successful data linkage research (Bradley et al., 2010; Ritchie et al., 2015). A multidisciplinary team including clinical researchers, data custodians, DLUs, epidemiologists, statisticians and data scientists is required. Consumer involvement in data linkage research is generally overlooked but essential for ensuring meaningful outcomes (Jewell et al., 2019).

More infrastructure and resources to support data linkage research are needed. This includes more funding for DLUs to bolster their capacity to meet growing demands (Andrew et al., 2016), as well as greater investment from funding bodies for data linkage research (Andrew et al., 2016; Bradley et al., 2010; Downs et al., 2019). Data linkage expenses are increasing including DLUs expenses (linkage and storage of data), costs for analysts and stakeholder engagement. Researchers, health policy makers and data custodians need to remove barriers to accessing linked health datasets (Olver, 2014) and promote cross-jurisdiction research ((Boyd et al., 2015; Lloyd et al., 2024). Streamlining and harmonising approaches and processes for ethics and governance also needs to be addressed (Andrew et al., 2016; Carson et al., 2020; Lloyd et al., 2024). Such work is necessary to mitigate lengthy delays and resource duplication in research activities.

Researchers should adhere to various reporting guidelines specific to data linkage. Many of the guidelines have been published in recent years and have not been used or referenced in studies. However, such guidelines are designed to improve research quality, transparency and replicability.

Researchers should consider the full breadth of administrative data collected across sectors and explore how cross-sectoral data linkage can address complex and meaningful questions. Examining sensitive and difficult-to-investigate research topics using data linkage, including homelessness, family violence, self-harm, suicidality, mortality and health service utilisation, has clear benefits (Borschmann et al., 2017; Cvejic et al., 2022; Ducat et al., 2013; Meurk et al., 2022; Young et al., 2024. However, the impacts of mental illness are far broader. Consideration of other sectors such as education, disability, family, child protection, homelessness and housing support, substance use and forensic services will provide insights into gaps between sectors, improve estimation of service use and health outcomes, identify those individuals who are socially excluded from health services and allow the examination of social determinants of mental illness (Pearce et al., 2023).

There is also the need to look at how and what mental health-related data are collected across databases, sectors and jurisdictions. Can consensus be obtained regarding core routine outcome measures for mental health? Are these measures meaningful to key stakeholders such as those with a lived experience, their caregivers, clinicians, service providers and policy makers? Do they have appropriate psychometric properties? Can primary health and mental health care data be integrated to national data assets such as PLIDA?

Finally, a community of practice for data linkage research in mental health in Australia is needed. Efforts should focus on capacity-building via sharing and publishing experiences, methodologies and approaches (e.g. development of more sophisticated models such as foundation models, LLMs) and research translation for policy and practice.

Conclusion

In this manuscript, we have provided a comprehensive and up-to-date overview of data linkage methodologies, from the planning stages of research to reporting of research findings. We have also highlighted diverse research questions that have been addressed using data linkage in mental health research in Australia. To advance the field of data linkage research in mental health, there needs to be greater education across stakeholder groups, better engagement with lived experience, improved infrastructure and funding and further exploration of benefits and utility of cross-jurisdictional and cross-sectoral linkage. Such work will enable us to address complex and meaningful questions regarding mental health that other research designs cannot adequately address.

Supplemental Material

sj-docx-1-anp-10.1177_00048674251333574 – Supplemental material for Using data linkage for mental health research in Australia

Supplemental material, sj-docx-1-anp-10.1177_00048674251333574 for Using data linkage for mental health research in Australia by Sue M Cotton, Jana M Menssink, Matthew Hamilton, Kate M Filia, Shu Mei Teo, Mengmeng Wang, Dan ZQ Gan, Wenhua Yu, Amity Watson, Katrina Witt, Melissa Hasty, Carl Moller, Alison Yung and Caroline X Gao in Australian & New Zealand Journal of Psychiatry

Footnotes

Declaration of Conflicting Interests

The author (s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author (s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Sue M Cotton

Jana M Menssink

Matthew Hamilton

Kate M Filia

Shu Mei Teo

Mengmeng Wang

Dan ZQ Gan

Amity Watson

Katrina Witt

Melissa Hasty

Carl Moller

Alison Yung

Supplemental Material

Supplemental material for this article is available online.

References

Adhikari

Metcalfe

Bulloch

AGM

, et al. (2020) Mental disorders and subsequent suicide events in a representative community population. Journal of Affective Disorders 277: 456–462.

Akpanekpo

Kariminia

Srasuebkul

, et al. (2024) Psychiatric admissions in young people after expiration of criminal justice supervision in Australia: A retrospective data linkage study. BMJ Mental Health 27: e300958.

Andrew

Sundararajan

Thrift

, et al. (2016) Addressing the challenges of cross-jurisdictional data linkage between a national clinical quality registry and government-held health data. Australian and New Zealand Journal of Public Health 40: 436–442.

Australian Commission on Safety and Quality in Health Care (2008) Operating principles and technical standards for Australian clinical quality registries. TRIM record 20308. Sydney, NSW, Australia: Australian Commission on Safety and Quality in Health Care.

Australian Institute of Aboriginal and Torres Strait Islander Studies (AIATSIS) (2020) AIATSIS code of ethics for aboriginal and Torres strait islander research. Final report, Australian Institute of Aboriginal and Torres Strait Islander Studies, Canberra, ACT, Australia.

Benchimol

Smeeth

Guttmann

, et al. (2015) The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) Statement. PLOS Medicine 12: e1001885.

Bohensky

(2015) Bias in data linkage studies. In: Harron

Goldstein

Dibben

(eds) Methodological Developments in Data Linkage. West Sussex: Wiley, pp. 63–82.

Bohensky

Jolley

Sundararajan

, et al. (2011) Development and validation of reporting guidelines for studies involving data linkage. Australian and New Zealand Journal of Public Health 35: 486–489.

Borschmann

Thomas

Moran

, et al. (2017) Self-harm following release from prison: A prospective data linkage study. Australian & New Zealand Journal of Psychiatry 51: 250–259.

10.

Bower

Stanley

(1989) Dietary folate as a risk factor for neural-tube defects: Evidence from a case-control study in Western Australia. The Medical Journal of Australia 150: 613–619.

11.

Bower

Ryan

Rudy

, et al. (2002) Trends in neural tube defects in Western Australia. Australian and New Zealand Journal of Public Health 26: 150–151.

12.

Boyd

Ferrante

O’Keefe

, et al. (2012) Data linkage infrastructure for cross-jurisdictional health-related research in Australia. BMC Health Services Research 12: 480.

13.

Boyd

Randall

Ferrante

, et al. (2015) Accuracy and completeness of patient pathways – The benefits of national data linkage in Australia. BMC Health Services Research 15: 312.

14.

Boyle

Emery

(2017) Data linkage. Australian Journal for General Practitioners 46: 615–619.

15.

Bradfield

(2022) Waving away waivers: An obligation to contribute to ‘herd knowledge’. For data linkage research? Research Ethics 18: 151–162.

16.

Bradley

Penberthy

Devers

, et al. (2010) Health services research and data linkages: Issues, methods, and directions for the future. Health Services Research 45: 1468–1488.

17.

Brownell

Jutte

(2013) Administrative data linkage as a tool for child maltreatment research. Child Abuse and Neglect 37: 120–124.

18.

Bye

Carter

Leightley

, et al. (2023) Observational prospective study of social media, smartphone use and self-harm in a clinical sample of young people: Study protocol. BMJ Open 13: e069748.

19.

Bye

Carter

Leightley

, et al. (2024) Cohort profile: The Social media, smartphone use and Self-harm in Young People (3S-YP) study–A prospective, observational cohort study of young people in contact with mental health services. PLoS ONE 19: e0299059.

20.

Carson

Jewell

Downs

, et al. (2020) Multisite data linkage projects in mental health research. The Lancet Psychiatry 7: e61.

21.

Carter

Laurie

Dixon-Woods

(2015) The social licence for research: Why care.data ran into trouble. Journal of Medical Ethics 41: 404–409.

22.

Christen

Goiser

(2007) Quality and complexity measures for data linkage and deduplication. In: Guillet

Hamilton

(eds) Quality Measures in Data Mining. Berlin: Springer, pp. 127–151.

23.

Clements

Jones

Morriss

, et al. (2015) Self-harm in bipolar disorder: Findings from a prospective clinical database. Journal of Affective Disorders 173: 113–119.

24.

Cotton

Filia

Watson

, et al. (2022) A protocol for the first episode psychosis outcome study (FEPOS): ⩾15 year follow-up after treatment at the early psychosis prevention and intervention centre, Melbourne, Australia. Early Intervention in Psychiatry 16: 715–723.

25.

Cvejic

Srasuebkul

Walker

, et al. (2022) The health service contact patterns of people with psychotic and non-psychotic forms of severe mental illness in New South Wales, Australia: A record-linkage study. Australian and New Zealand Journal of Psychiatry 56: 675–685.

26.

Cybulski

Chilman

Jewell

, et al. (2024) Improving our understanding of the social determinants of mental health: A data linkage study of mental health records and the 2011 UK census. BMJ Open 14: e073582.

27.

da Silva Marinho

Medina Coeli

Ventura

, et al. (2012) Informed consent for record linkage: A systematic review. Journal of Medical Ethics 38: 639.

28.

Dadi

Brown

, et al. (2024) Association between maternal mental health-related hospitalisation in the 5 years prior to or during pregnancy and adverse birth outcomes: A population-based retrospective cohort data linkage study in the Northern Territory of Australia. The Lancet Regional Health – Western Pacific 46: 101063.

29.

Davis

KAS

Sudlow

CLM

Hotopf

(2016) Can mental health diagnoses in administrative data be used for research? A systematic review of the accuracy of routinely collected diagnoses. BMC Psychiatry 16: 263.

30.

de Man

Wieland- Jorna

Torensma

, et al. (2023) Opt-in and opt-out consent procedures for the reuse of routinely recorded health data in scientific research and their consequences for consent rate and consent bias: Systematic review. Journal of Medical Internet Research 25: e42131.

31.

Dibben

Elliot

Gowans

, et al. (2016) The data linkage environment. In: Harron

Goldstein

Dibben

(eds) Methodological Developments in Data Linkage. West Sussex: Wiley, pp. 36–62.

32.

Doidge

Harron

(2018) Demystifying probabilistic linkage: Common myths and misconceptions. International Journal of Population Data Science 3: 410.

33.

Downs

Ford

Stewart

, et al. (2019) An approach to linking education, social care and electronic health records for children and young people in South London: A linkage study of child and adolescent mental health service data. BMJ Open 9: e024355.

34.

Ducat

Ogloff

McEwan

(2013) Mental illness and psychiatric treatment amongst firesetters, other offenders and the general community. Australian & New Zealand Journal of Psychiatry 47: 945–953.

35.

Fahridin

Agarwal

Bracken

, et al. (2024) The use of linked administrative data in Australian randomised controlled trials: A scoping review. Clinical Trials 21: 516–525.

36.

Gilbert

Lafferty

Hagger-Johnson

, et al. (2018) GUILD: GUidance for Information about Linking Data sets†. Journal of Public Health 40: 191–198.

37.

Guo

Chen

(2023) Big data analytics in healthcare. In: Nakamori

(ed.) Knowledge Technology and Systems: Toward Establishing Knowledge Systems Science. Singapore: Springer, pp. 27–70.

38.

Hafekost

Lawrence

Boterhoven de Haan

, et al. (2016) Methodology of young minds matter: The second Australian Child and Adolescent Survey of Mental Health and Wellbeing. Australian and New Zealand Journal of Psychiatry 50: 866–875.

39.

Haneef

Tijhuis

Thiébaut

, et al. (2022) Methodological guidelines to estimate population-based health indicators using linked data and/or machine learning techniques. Archives of Public Health 80: 9.

40.

Harron

(2022) Data linkage in medical research. BMJ Medicine 1: e000087.

41.

Harron

Dibben

Boyd

, et al. (2017a) Challenges in administrative data linkage for research. Big Data & Society 4: 2053951717745678.

42.

Harron

Doidge

Knight

, et al. (2017b) A guide to evaluating linkage quality for the analysis of linked data. International Journal of Epidemiology 46: 1699–1710.

43.

Harron

Goldstein

Dibben

(2016) Introduction. In: Harron

Goldstein

Dibben

(eds) Methodological Developments in Data Linkage. New York: John Wiley & Sons.

44.

Hernán

Wang

Leaf

(2022) Target trial emulation: A framework for causal inference from observational data. JAMA 328: 2446–2447.

45.

Holman

Bass

Rosman

, et al. (2008) A decade of data linkage in Western Australia: Strategic design. Applications and Benefits of the WA Data Linkage System Australian Health Review 32: 766–777.

46.

Iorfino

McHugh

Richards

, et al. (2023) Patterns of emergency department presentations for a youth mental health cohort: Data-linkage cohort study. The British Journal of Psychiatry Open 9: e170.

47.

Iveson

Ball

Doherty

, et al. (2024) Cohort profile: The Scottish SHARE Mental Health (SHARE–MH) cohort – linkable survey, genetic and routinely collected data for mental health research. BMJ Open 14: e078246.

48.

Jewell

Pritchard

Barrett

, et al. (2019) The Maudsley Biomedical Research Centre (BRC) data linkage service user and carer advisory group: Creating and sustaining a successful patient and public involvement group to guide research in a complex area. Research Involvement and Engagement 5: 20.

49.

Kwakkenbos

Imran

McCall

, et al. (2021) CONSORT extension for the reporting of randomised controlled trials conducted using cohorts and routinely collected data (CONSORT-ROUTINE): Checklist with explanation and elaboration. BMJ: British Medical Journal 373: n857.

50.

Lariscy

(2011) Differential record linkage by Hispanic ethnicity and age in linked mortality studies: Implications for the epidemiologic paradox. Journal of Aging and Health 23: 1263–1284.

51.

Lloyd

Nicholson

Strange

, et al. (2024) The burdensome logistics of data linkage in Australia – The example of a national registry for congenital heart disease. Australian Health Review 48: 8–15.

52.

Lowitja Institute (2024) Taking control of our data: A discussion paper on indigenous data governance for aboriginal and Torres strait islander people and communities. Discussion Paper, Lowitja Institute, Collingwood, VIC, Australia, 30 January.

53.

Marino

Tait

Straker

, et al. (2022) Health, social and economic implications of adolescent risk behaviours/states: Protocol for Raine study Gen2 cohort data linkage study. Longitudinal and Life Course Studies 13: 647–666.

54.

Meurk

Wittenhagen

Bosley

, et al. (2022) Suicide crisis calls to emergency services: Cohort profile and findings from a data linkage study in Queensland, Australia. Australian and New Zealand Journal of Psychiatry 56: 144–153.

55.

Muller

SHA

Kalkman

van Thiel

GJMW

(2021) The social licence for data-intensive health research: Towards co-creation, public value and trust. BMC Medical Ethics 22: 110.

56.

National Health and Medical Research Council (2015) Principles for Accessing and using Publicly Funded Data for Health Research. Canberra, ACT, Australia: National Health and Medical Research Council.

57.

National Health and Medical Research Council (2018) Ethical conduct in research with aboriginal and Torres strait islander peoples and communities: Guidelines for researchers and stakeholders. Research Paper, National Health and Medical Research Council, Canberra, ACT, Australia, 31 July.

58.

National Health and Medical Research Council and Consumers Health Forum of Australia (2016) Statement on Consumer and Community Involvement in Health and Medical Research. Canberra, ACT, Australia: National Health and Medical Research Council.

59.

National Health and Medical Research Council; Australian Research Council and Universities Australia (2023) National Statement on Ethical Conduct in Human Research. Canberra, ACT, Australia: National Health and Medical Research Council.

60.

Nicholls

Quach

von Elm

, et al. (2015) The REporting of Studies Conducted Using Observational Routinely-Collected Health Data (RECORD) statement: Methods for arriving at consensus and developing reporting guidelines. PLoS ONE 10: e0125620.

61.

Nicolaie

Füssenich

Ameling

, et al. (2023) Constructing synthetic populations in the age of big data. Population Health Metrics 21: 19.

62.

Nissenbaum

(2009) Privacy in Context: Technology, Policy, and the Integrity of Social Life. Stanford: Stanford University Press.

63.

Olver

(2014) Linking data to improve health outcomes: Routinely collected data, when linked, are a rich source of sound evidence for making health decisions. The Medical Journal of Australia 200: 368–369.

64.

Palamuthusingam

Johnson

Hawley

, et al. (2019) Health data linkage research in Australia remains challenging. Internal Medicine Journal 49: 539–544.

65.

Papalia

Simmons

Trood

, et al. (2024) Police-reported family violence victimisation or perpetration and mental health-related emergency department presentations: An Australian data-linkage study. BMC Public Health 24: 131.

66.

Pearce

Borschmann

Young

, et al. (2023) Advancing cross-sectoral data linkage to understand and address the health impacts of social exclusion: Challenges and potential solutions. International Journal of Population Data Science 8: 2116.

67.

Pham

TTL

O’Brien

Liu

, et al. (2024) Suicide and mortality following self-harm in Culturally and Linguistically Diverse communities in Victoria, Australia: Insights from a data linkage study. Frontiers in Public Health 12: 1256572.

68.

Pratt

Mack

Meyer

, et al. (2020) Data linkage in pharmacoepidemiology: A call for rigorous evaluation and reporting. Pharmacoepidemiology and Drug Safety 29: 9–17.

69.

Productivity Commission (2020) Mental Health: Productivity Commission Inquiry Report. Melbourne, VIC, Australia: Productivity Commission.

70.

Randall

Ferrante

Boyd

, et al. (2013) The effect of data cleaning on record linkage quality. BMC Medical Informatics and Decision Making 13: 64.

71.

Ritchie

Green

Webber

, et al. (2015) Enabling Data Linkage to Maximise the Value of Public Health Research Data. London: Welcome Trust.

72.

Ryan

Riley

Cadilhac

, et al. (2021) Factors associated with stroke coding quality: A comparison of registry and administrative data. Journal of Stroke and Cerebrovascular Diseases 30: 105469.

73.

Simera

(2014) The equator network: Supporting editors in publishing well-reported health research. Science Editor 37: 15.

74.

Szmulewicz

(2024) Target trial emulation in psychiatry: A call for more rigorous observational analyses. The Lancet Psychiatry 11: 492–494.

75.

Tahamont

Jelveh

Chalfin

, et al. (2021) Dude, where’s my treatment effect? Errors in administrative data linking and the destruction of statistical power in randomized experiments. Journal of Quantitative Criminology 37: 715–749.

76.

Tan

Flack

Bear

, et al. (2015) An evaluation of a data linkage training workshop for research ethics committees. BMC Medical Ethics 16: 13.

77.

Thompson

Woods

Katzenellenbogen

(2012) The quality of Indigenous identification in administrative health data in Australia: Insights from studies using data linkage. BMC Medical Informatics and Decision Making 12: 133.

78.

Tibble

Law

Spittal

, et al. (2018) The importance of including aliases in data linkage with vulnerable populations. BMC Medical Research Methodology 18: 76.

79.

Tran

Havard

Jorm

(2017) Data cleaning and management protocols for linked perinatal research data: A good practice example from the Smoking MUMS (Maternal use of medications and safety) study. BMC Medical Research Methodology 17: 97.

80.

Vatsalan

Sehili

Christen

, et al. (2017) Privacy-preserving record linkage for Big Data: Current approaches and research challenges. In: Zomaya

Sakr

(eds) Handbook of Big Data Technologies. Cham: Springer International Publishing, pp. 851–895.

81.

von Elm

Altman

Egger

, et al. (2007) The strengthening the reporting of observational studies in epidemiology (STROBE) statement: Guidelines for reporting observational studies. Lancet 370: 1453–1457.

82.

Wittenhagen

Hielscher

Meurk

, et al. (2024) A cohort profile of children and adolescents who had a suicide-related contact with police or paramedics in Queensland (Australia). Emergency medicine Australasia 36: 520–526.

83.

Young

Burgess

Falster

, et al. (2024) Mental health–related service and medicine use among a cohort of urban Aboriginal children and young people: Data linkage study. Australian & New Zealand Journal of Psychiatry 58: 787–799.

84.

Young

Borschmann

Heffernan

, et al. (2020) Contact with mental health services after acute care for self-harm among adults released from prison: A prospective data linkage study. Suicide and Life-threatening Behavior 50: 990–1006.

85.

Tang

EYH

(2023a) Health disparities and comparison of psychiatric medication use before and after the COVID-19 pandemic lockdown among general practitioner practices in the North East of England. Int J Environ Res Public Health 20: 6034.

86.

Vale

McMeekin

, et al. (2023b) Investigating changes in mental health services utilisation in England and their impact on health outcomes and wellbeing during the COVID-19 pandemic: Protocol for a health data-linkage study. PLoS ONE 18: e0283986.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB