Australia has a universal health system comprising a multifaceted network of government (public) and private providers. Medicare is the universal public health insurance scheme, which is funded by the federal government through a combination of general tax revenue and a Medicare levy based on taxable income. It is administered by Services Australia (formerly the Department of Human Services) and provides free or subsidized access to 3 main areas of health care provision: (1) medical and health services outside of the public hospital setting (administered through the Medicare Benefits Schedule [MBS]), (2) prescription pharmaceuticals (administered through the Pharmaceutical Benefits Scheme [PBS]), and (3) public hospital services (funded jointly with, and managed by, state and territory governments). In addition, the federal government pays 75% of the fee for services and procedures for private patients in public or private hospitals and subsidizes private health insurance. The EXTEND45 Study includes administrative data collected by the federal government (ie, MBS and PBS) as well as state-specific data collections of the NSW government.

Details of the 45 and Up Study have been published previously [21]. Briefly, the 45 and Up Study comprises adults aged ≥45 years sampled from the general population of NSW, Australia’s most populous state. Between January 2006 and December 2009, participants were randomly sampled from the Services Australia enrollment database, with individuals aged ≥80 years or living in remote areas oversampled by a factor of 2. Consenting participants self-completed baseline questionnaires in English, which included questions on demographic and socioeconomic characteristics, personal health behaviors, medical and surgical history, medications, and physical and psychological health [22]. Participants also consented to long-term follow-up, including linkage of their baseline responses to routinely collected health datasets. In the EXTEND45 Study, laboratory and administrative datasets have been linked at an individual level to the 45 and Up Study participants and their baseline questionnaire responses.

Record linkage brings together information on the same individual from different data sources. Strict privacy protecting protocols are mandated by the NSW Population and Health Services Research Ethics Committee (PHSREC), the data custodians, and the NSW Center for Health Record Linkage (CHeReL). The CHeReL is an intermediate body established in 2006 to provide expertise in data linkage methodology and maintain a record linkage system that protects data privacy. It is jointly managed by the NSW Ministry of Health and the NSW Cancer Institute.

The Sax Institute has an ongoing link with Services Australia, and as a result has processes in place to deterministically link MBS and PBS data to their 45 and Up Study participants once a year. This is done using a unique identifier provided to the Sax Institute by Services Australia. For all other aforementioned datasets, linkage to 45 and Up Study data is performed by the CHeReL using probabilistic linkage (Figures 2 and 3). Data custodians provide the CHeReL with an encrypted unique identifier and relevant personal information (full name, date of birth, address, gender, and, where available, country of birth) for all patients over the relevant time-period. No clinical data are provided to the CHeReL at any point in the linkage process. The CHeReL uses personal information to link records from different data sources, using probabilistic linkage in the ChoiceMaker software (ChoiceMaker Technologies, Inc; Figure 2). For records with doubtful matches, a clerical review is also conducted [24]. Once records are linked, the corresponding individual is allocated a project-specific person number (PPN). The CHeReL sends the PPNs for all individuals to each data custodian, together with the original unique identifier used in that custodian’s particular dataset (Figure 3). Each data custodian will now hold an EXTEND45-specific PPN for each individual in their respective dataset for whom linkage with at least one other dataset was achieved. The data custodian merges the PPN with the clinical variables that have been approved for use in the EXTEND45 Study and uploads the deidentified data to a highly secure server that is accessed by the EXTEND45-approved researchers. The PPN is used to combine records for the same person from different datasets.

The 2 initial diseases of interest are CKD and diabetes. For each individual, their diabetes and CKD status and, where appropriate, timing of incident disease, are determined using the various linked datasets. The precise methods used to ascertain disease status will be outlined in the relevant publications. In general, however, kidney function and CKD status are predominantly derived using data from laboratory records and, where appropriate, the ANZDATA Registry. Linked serum creatinine measurements are applied to the Chronic Kidney Disease Epidemiology Collaboration equation to calculate the estimated glomerular filtration rate [25]. Other available laboratory data related to kidney function include urine albumin-to-creatinine ratio and urine protein-to-creatinine ratio.

Diabetes status is derived using the following criteria: (1) recorded dispensing of insulin (Anatomical Therapeutic Chemical [ATC] Classification: A10A) or oral blood glucose-lowering medication (ATC Classification: A10B) in the PBS dataset, (2) self-reported diabetes on the 45 and Up Study (answered Yes to Q24. Has a doctor EVER told you that you have diabetes?), (3) laboratory record of HbA_1c result >6.5% on one occasion, (4) laboratory record of fasting plasma glucose levels >7.0 mmol/L, and (5) laboratory record of serum glucose levels >11.1 mmol/L. At present, no distinction has been made between type 1 and type 2 diabetes.

The EXTEND45 Study is led by a collaborative group of physicians, epidemiologists, and statisticians coordinated by The George Institute for Global Health, Sydney. Similar to the 45 and Up Study, the EXTEND45 Study has been designed to facilitate cooperative projects that lie within its scope, with oversight provided by a designated steering committee. The steering committee includes representatives from The George Institute for Global Health, the Universities of NSW and Sydney, the Sax Institute, clinical experts from Australian hospitals, and nonvoting representatives from pharmaceutical sponsors. Current steering committee members are listed in Multimedia Appendix 1.

The steering committee is responsible for the general oversight and governance of the study and provides academic independence and integrity. It provides scientific advice regarding all aspects of study design, conduct, analyses, and publication of results. In addition, the steering committee is responsible for the approval of prespecified analysis plans. Analysis proposals from internal or external researchers are encouraged and may be submitted by academic investigators, health service providers, or commercial health care entities.

Data collection for the EXTEND45 Study was through data linkage only. Participants are included on the basis of their participation and consent to the 45 and Up Study, which included consent for linkage to their routinely-collected health records. The 45 and Up Study was approved by the University of NSW Human Research Ethics Committee (UNSW HREC). Ethical approval for the EXTEND45 Study was obtained from the NSW Population and Health Service Research Ethics Committee (HREC/13/CIPHS/69).

Participant confidentiality is protected in several ways. First, linkage methodology splits the processes of record linkage and data analysis, ensuring that participant identifiers are always kept separate from clinical data. Second, security measures in place at the CHeReL ensure that the risks of a breach of privacy are minimal. Third, as mandated by the Sax Institute, all data within the EXTEND45 Study is transferred and accessed via the Secure Unified Research Environment (SURE). SURE is a high-performance remote-access virtual computing environment designed specifically for secure access, storage, and analysis of anonymized health information [26]. To prevent data from being exchanged between different linked data studies, each study in the SURE facility exists within its own security perimeter.

Through a novel linkage between the 45 and Up Study, various administrative databases, and data from laboratory service providers, the EXTEND45 Study represents a unique and rich data resource that can be used to investigate a range of questions relating to chronic disease. The study’s mission is to provide much-needed evidence of the epidemiology, burden, progression, and clinical management of chronic diseases such as CKD and diabetes in the general Australian population and across the full disease spectrum. Multiple projects that use this data resource are already completed or underway.

Altogether, the datasets included in the EXTEND45 Study provide a more complete overview of the care provided to patients with chronic disease than is currently available. The 45 and Up Study baseline questionnaire included questions on a range of personal health behaviors (eg, smoking, alcohol consumption, physical activity, and sleep habits) as well as demographic, social, and economic characteristics (eg, marital status, country of birth, and education level). Although many of these factors are either known or hypothesized to be important confounders of chronic disease incidence and progression, they are typically poorly captured in administrative datasets. Conversely, the 45 and Up Study baseline data provide only a single snapshot in time, are exclusively self-reported, and are lacking in granularity. In the EXTEND45 Study, the combining of NSW APDC and other health datasets will allow outcomes, comorbidities, and complications of CKD and diabetes to be examined. Moreover, laboratory data are critical for identifying individuals at the early stages of their disease and allow more granular assessment of the achievement (or lack thereof) of clinical targets. Finally, the collection of postcodes of residences and health care facilities in the various datasets will enable the geographic evaluation of access and use of all tiers of health services.

The effective use of routinely collected data and, in particular, laboratory data in chronic disease research is demonstrated by the many research and policy outputs of the ongoing Alberta Kidney Disease Network (AKDN). The AKDN is a collaborative research organization that holds a central repository of linked laboratory and administrative data from Alberta, Canada [28]. Research using this repository has ranged from a comparative risk assessment of coronary events in patients with diabetes and CKD [20] to factors associated with RRT initiation [29]. In the United Kingdom, the UK Biobank, established in 2006-2010, has recruited 500,000 people aged 40 to 69 years to provide detailed information about themselves, undergo various measurements, provide biological samples for analysis, and have their electronic health record data linked, to create a major data resource. To date, 989 papers have been published that used this resource. Similarly, the UK Clinical Practice Research Datalink (CPRD) links general practice records to secondary care, mortality, and other disease-specific databases. The CPRD is an excellent example of the utility of real-world data in generating evidence for drug safety guidance and clinical practice in particular. Research carried out as part of the EXTEND45 Study will complement these earlier studies.

A main strength of the EXTEND45 Study is the breadth and diversity of the data held within it, allowing analyses into a range of research questions to be conducted. Some of the potential analyses have already been described. Importantly, the inclusion of laboratory data allows researchers to use the EXTEND45 Study dataset to identify earlier stages of both diabetes and CKD and move away from reliance upon self-report for diabetes. This will provide essential information on the prevalence and progression of early disease in the community as well as health care delivery to those affected with mild disease. Indeed, previous studies have either relied on clinical populations or used sources that only include outcome measures such as hospitalization or mortality—in both instances, chronic disease is identified later in its disease course. Second, multiple data sources will be used to ascertain chronic disease status, which will improve case identification. The relative contributions of different data sources to the identification of cases will be reported in the respective publications. Finally, the combined infrastructure of the 45 and Up Study and EXTEND45 Study offers a cheaper and more efficient alternative to traditional longitudinal cohort studies for assessing the incidence of disease, health outcomes, and health service utilization in a large cohort.

There are important limitations that must be considered when using the EXTEND45 Study database, both at the design stages and during interpretation of results. Routinely collected and administrative data are primarily collected for purposes other than research. As a result, not all variables of interest will be available all of the time. For example, within the EXTEND45 Study, there is no information on blood pressure, an important clinical variable in both CKD and diabetes. Depending on the specific research question, surrogates for some of these missing variables may be deployed. For instance, for high blood pressure, a pharmaceutical dispensing of antihypertensive medication or hospital admission for hypertension-related problems may be used. A second limitation is that CKD is primarily defined using laboratory tests. As a result, disease identification is limited to individuals who have presented for medical review and had laboratory tests performed. Conversely, the absence of a linked laboratory test in the EXTEND45 Study dataset may be due to (1) the test being performed by a laboratory service provider not part of the EXTEND45 Study, (2) an individual having no indication for a test (and, therefore, healthier), or (3) an individual being indicated but not having access to the appropriate health services. Given the countervailing effects of these different possibilities, a comparison of individuals with linked laboratory data to those without linked laboratory data must be performed in individual analyses. The use of routine laboratory results over time means that inter- and intralaboratory variation may exist. The effects of this are minimized through deep cleaning of the data to identify potentially anomalous results or inconsistent measurement units as well as detailed discussions with laboratory service providers to construct a data dictionary that is specific to their dataset.

Finally, the response rate for the 45 and Up Study baseline questionnaire was 18%, which is similar to other comparable studies. The 45 and Up Study cohort may, therefore, represent a slightly healthier cohort than the general population, with implications for the generalizability of findings that stem from it. Unfortunately, most cohort studies are vulnerable to this healthy volunteer effect. However, we believe that the use of routinely collected data to follow up participants is an excellent alternative to active longitudinal follow-up for minimizing this potential bias over time. Moreover, internal comparisons within the 45 and Up Study cohort have been shown to be generalizable even in the presence of any selection bias [30]. Finally, although not necessarily a limitation, it is important to remember that the different data sources included in the EXTEND45 Study dataset provide varying levels of population coverage; although government datasets typically capture health service use in both the private and public health care sector, the presence of multiple laboratory service providers in NSW means that laboratory data for the participants are not comprehensive over time or across individuals.

The EXTEND45 Study is, to our knowledge, the first linked data study in Australia to incorporate data from community laboratory service providers with other administrative datasets for chronic disease research. The study represents a unique collaboration among academics, clinicians, primary care researchers, data linkage experts, and laboratory service providers to construct an invaluable data resource that can be used to answer a range of questions relating to chronic disease. Although CKD, diabetes, and CVD have been identified as initial conditions to be examined, consideration of other chronic diseases such as cancer, dyslipidemia, and musculoskeletal conditions is also possible.