The CKD.QLD data linkage framework: chronic kidney disease and health services utilisation in Queensland, Australia

Data linkage framework

Seven years of retrospective data for persons with any hospital admissions with a discharge date from 1 May 2011 to 30 June 2018 were sourced from Queensland Health (QH) for two cohorts of patients with CKD and one cohort without CKD. The specific data sources were the Queensland Hospital Admitted Patient Data Collection (QHAPDC), the Queensland Registrar General (RG) deaths and the Activity Based Funding (ABF) Model Output data for Queensland. The data linkage framework is illustrated in Figure 1.

Figure 1. The data linkage framework of the sub-study of CKD.QLD Registry.

The first cohort of people with CKD (Cohort 1) had consented and enrolled in the CKD.QLD Registry until January 2019. As for all admitted patients, the Data Linkage Team of the Queensland Department of Health assigned each a unique identifier within the QH system.^13,14

A second cohort of people with CKD (Cohort 2), but not in the CKD.QLD Registry, were identified by the QH Data Linkage Team through International Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM) codes related to their first hospital admission that were compatible with CKD (see Table 1). They were matched by HHS with Cohort 1 patients in a ratio of 3 to 1. The second cohort reflects a broader group of people with CKD than those in the CKD.QLD Registry and includes many managed by non-nephrologists.

The third cohort (Cohort 3) consisted of people without an ICD-10-AM diagnosis compatible with CKD in their first hospital admission, as identified by the QH Data Linkage Team, who were matched by age, sex and HHS to Cohort 1 in a ratio of 3 to 1.

Data were generated for the discharge from hospital admission period of 1 May 2011 to 30 June 2018. For all members of Cohort 2, and Cohort 3 the same scope and data items were requested from QHAPDC, Deaths and the ABF Model Output datasets, as for as Cohort 1. In addition, patient details including date of birth, sex, country of birth and Indigenous status were provided.

Settings and participants

CKD.QLD is an initiative of the University of Queensland’s Centre for Chronic Disease, in partnership with Queensland Health and the Queensland University of Technology.

There are 12 adult public renal services across the 16 regional Queensland Health Hospital and Health Services (see Figure 2). The renal services operate varied models of CKD community and satellite clinics. Participants are pre-dialysis CKD patients who were referred or within these services are eligible for inclusion in the Registry, via informed consent, and with ethics and governance oversight. Patients with CKD from both the Registry and the matched cohorts in Queensland are age 18 years and over.

Figure 2. Queensland Health Adult Renal Services by Hospital and Health Service area.

Data sources for data linkage

QHAPDC and RG deaths

The Queensland Hospital Admitted Patient Data Collection (QHAPDC) receives demographic and clinical information on all admitted patients separated (an inclusive term meaning discharged, died, transferred or statistically separated) from both public and licensed private hospitals and licenced day surgery units in Queensland.¹⁵

The linked data included the following QHAPDC data items and death Registry data from the Queensland RG deaths for participants in the three cohorts (see Table 2).

Queensland ABF model output data

QH allocate funds to the HHSs on their activity using the ABF Model.¹⁶ The model calculates number of episodes of care and case complexity to derive a weighted unit of activity (WAU). The methodology of deriving WAU standardises resource utilisation across episodes of care that are diverse in resource usage and geography. The ABF model calculates a Queensland Efficient Price using Queensland WAUs (QWAU) and the Queensland base price per WAU, noting the QWAU may differ from the national WAU (NWAU) reflecting State-Government specific costs and priorities.¹⁷ HHSs report their activities through the National Hospital Cost Data Collection (NHCDC) and the data are also used as inputs into the ABF model to derive output data.¹⁸ Person service level costs are determined using the methodology of the NHCDC and are reported in NHCDC summation and format. The data in scope for the NHCDC includes all patient level activity for all public hospital facilities across Australia, and the costs incurred by the hospital with the activity in each financial year. For this data linkage study, we had the following ABF data items (see Table 3), linked to the cohort file for each CKD.QLD Registry participant and participants in the matched cohorts 2 and 3.

Data linkage process

The data linkage process for this study is illustrated in Figure 3. The process is adapted from the “Overview of Application Process for Access to Confidential Health Information held by Queensland Health for Research Purposes” published by the State of Queensland (Queensland Health), January 2016.¹⁴ We adopted the process, contacting the Statistical Analysis and Linkage Team (SALT), Statistical Services Branch, Queensland Health for initial advice and consulted data custodians. The work was conducted with the ethical approval of the QH Health Innovation, Investment and Research Office (HIIRO) under the Public Health Act (PHA), a pathway if researchers are requesting identifiable or potentially re-identifiable health information but are unable to obtain participant consent. The Public Health Act 2005¹⁹ establishes the process for accessing health information held by Queensland Health for approved research projects.

Figure 3. CKD.QLD data linkage process.

After receiving PHA approval, CKD.QLD provided a unique identifier – Registry ID – and associated patient key demographic variables (see Table 4) to the Data Linkage Unit in a password protected file. Using this unique identifier, data outlined in this framework were linked from data custodian sources to the affiliated individual.

Data security

The information on CKD.QLD participants to enable data linkage was provided to Queensland Health by the research Chief Investigator (WH) at the University of Queensland. The data were stored electronically with an encrypted and password protected file and sent to the Linkage Team for linkage to the requested cohort and extraction of their data. The data linkages were carried out in compliance with QH data management guidelines, under the governance of the Statistical Services Branch, the Queensland Department of Health.¹⁴ Participant privacy was paramount. Upon the collation of the data via the data linkage protocols, a zipped file was loaded by SALT onto a secure web portal and downloaded by one of the approved investigators in this study (JZ) who stored it in a secure database at the University of Queensland.

Data reporting

The linked data are being analysed by the CKD.QLD research team and by the health economics team using statistical software (Stata Statistical Software: Release 16. StataCorp LLC (RRID:SCR_012763)). (An open-access alternative that can provide an equivalent function is the R stats package (R Project for Statistical Computing, RRID:SCR_001905)). The analysis methods including descriptive statistics, comparations, and multivariable analysis, e.g., cox regressions are or will be applied. The results will be transformed into reports of de-identified data for user groups and collaborators, which will include, but are not be limited to, Queensland Health, the Australian Institute of Health and Welfare, The University of Queensland, the Australasian and New Zealand Society of Nephrology, the Renal Society of Australasia, and Kidney Health Australia. In addition, outcomes will be published in academic journals and on the CKD.QLD (www.ckdqld.org) and CKD.CRE website (https://cre-ckd.centre.uq.edu.au) addresses for professional and patient access and dissemination. Data will be published in a form that does not identify, or allow the re-identification, of any individual participant or health service provider.

Ethical approval and consent to participate

This study under the data linkage framework is considered a sub-study of the CKD.QLD Registry (HREC/15/QRBW/294), which has been approved by Metro North Human Research Ethics Committee and University of Queensland Medical Research Ethics (Number: 2011000029). The study of the QH held datasets of CKD patients, both those in the CKD.QLD Registry and the comparison patient cohort, listed in this study is through an ethically approved waiver of consent according to the National Health and Medical Research Council (NHMRC) National Statement on Ethical Conduct in Human Research.²⁰ In addition, a Public Health Act application (RD006802) was approved to access all hospitalisation information of the CKD.QLD Registry patients.

Database utility and applications

The framework linked 7,341 CKD patients from the CKD.QLD Registry (Cohort 1) to health service data including QHAPDC, RG Deaths and Queensland ABF Model Output Data. In addition, a cohort of 22,023 participants with CKD (Cohort 2) and a cohort of 22,023 without CKD (Cohort 3) were matched. This dataset embraces 51,387 persons and 974,700 admissions over seven years of retrospective data, with admitted persons divided into three groups, as follows.

In the CKD.QLD cohort (Cohort 1), 45.9% were female and 54.1% were males, and 8.4 % were Indigenous Australians. The mean age (SD; IQR) on their first admission was 65.2 (21.0; 18.8). They had 221,347 admissions, of which 77,822 admissions (35.2%) were for KRT for people who ultimately required that treatment. Of the 7,341 CKD.QLD patients who had at least one hospitalisation episode, 831 patients (11.3%) started KRT and 1,797 died (23.1%) during the study period. The person years of the study period (from their first admission to death or censor date) was 27,900.8 years.

Of the 22,023 people in the non-CKD.QLD CKD cohort (Cohort 2), 45.9% were female and 54.1% were males, and 4.3 % were Indigenous Australians. The mean age (SD; IQR) on their first admission was 73.8 (13.7; 17.0). They had 626,554 admissions, of which 277,994 admissions (44.4%) were eventually for KRT. Of the 22,023 participants in this cohort, 1,233 (5.6%) started KRT and 9,494 died (43.1%). The person years of the study interval was 128,570.6 years.

Of the 22,023 age and gender matched non-CKD cohort (cohort 3), 45.8% were female and 54.2% were males, and 1.8 % were Indigenous Australians. The mean age (SD; IQR) on their first admission was 63.5 (14.9; 18.6). They had 126,799 total admissions, only three eventually had dialysis, and 2,444 patients died (11.1%). The person years of the study interval was 149,439.3 years.

Linking established large datasets is a powerful approach to identifying patterns and interaction between multiple variables without bias whilst limiting the costs of data acquisition. Our first objective in establishing this framework is to better understand hospital admissions and outcomes among persons with preterminal CKD seen in public renal services across the Australian state of Queensland. The second objective is to compare them with persons from the same regions who had CKD documented in one or more hospital admissions who were not in the CKD.QLD cohort, to understand the characteristics and outcomes of CKD in other community and practice settings. The third objective is to compare these findings with persons admitted to hospital without a diagnosis of CKD, but who are matched age, gender, place of residence and HHSs with the CKD cohort in order to assess the degree which a diagnosis of CKD and its accompaniments might influence hospital usage.

The data on CKD.QLD patients will allow evaluation of health service utilisation for individual patients, covering all admissions, not limited to renal admissions, both public and private, across all disciplines. Many facts can be illuminated, including assessment of the extent to which CKD-related conditions drive admissions vs conditions not clearly related to CKD, characterisation of high cost users of hospital services, potentially avoidable hospital admissions and readmissions, trajectory over time of persons of different baseline characteristics, processes, costs and outcomes with specific interventions, different models of care and changes in policy, for example, strengthening supportive/palliative care pathways for patients with end stage kidney failure, and for persons who ultimately go onto KRT, the changes in resource utilisation and costs. Additionally, patterns of hospital utilisation can be inspected by postcode of residence, by HHS, by ethnic group, and other variables, to identify gaps and opportunities for service modification to improve outcomes.

The data from the comparator CKD group show the broader view of persons with CKD interacting with all hospitals in Queensland. These include patients in private renal practices, in nonrenal public specialty practices, in primary care and general practice, in Indigenous health services, and people in whom CKD was not recognised prior to their hospital admission within the study period.

The data from the non-CKD matched cohort will help clarify to what extent the presence of CKD influences the frequency, causes, costs and outcomes of hospital admissions in persons who are otherwise matched for age, gender, region and hospital services.

There is extensive information on the health and outcomes of patients on KRT in Australia and elsewhere, but hospital usage by preterminal CKD patients is not well defined. This study is the first in Australia aiming to define this issue in preterminal CKD in public renal practices, to compare it with CKD patients in other practice and community settings and to estimate how much CKD amplifies hospital usage over the level of matched persons without CKD. The Queensland population is similar to the aggregate Australian populations in demographics, ethnic mix,²¹ and rates of CKD,^1–4 so our findings are likely to be generalizable across Australia. In addition, hospital admission data include all (licensed) hospital facilities across the state of Queensland, so that all episodes of care were captured even if patients used both private and public facilities.

There are several limitations to this study. First, whilst most public renal practices participated in CKD.QLD, patient enrolment depended on availability of discretionary time of study coordinators who were busy with other clinical duties, so not all patients were captured. Second, the final CKD.QLD cohort does not include one major renal specialty practice, for which all 998 enrolees were dropped from the database after discrepancies in outcomes data could not be resolved. Third, renal patients who received care exclusively in private specialty practices are not included; patient profiles and service utilisation in private renal services might be somewhat different than described here. Finally, people with CKD who were not admitted to the hospital during the observation interval were not included. For example, 11.6% of CKD.QLD patients did not have hospital admissions after their recruitment to the registry. Finally, the framework does not capture encounters and resource consumption in the outpatient setting. This usage is under study in another framework.²²

We have already used this framework to describe hospital admissions for patients at each individual registry site (ongoing), to define impacts of anaemia in CKD patients,²³ frequency and consequences of acute kidney injury in patients with CKD,²⁴ the characteristics of high-cost users,^25,26 and to explore factors involved with readmissions by Machine Learning,²⁷ and several other studies are planned. Beyond these, the datasets can inform CKD health practitioners, Queensland Health, the Federal government, the national and international research communities, and local and national advocacy groups like Kidney Health Australia. The outcomes of this linkage study can assist the Australian Institute of Health and Welfare in defining the broader implications of CKD in terms of service consumption and its associated costs.

Underlying data

The CKD.QLD Registry is managed according to legislative requirements including the Information and Privacy Act 2009, and as per the National Statement on Ethical Conduct in Human Research.²⁰ This study utilises an established data linkage methodology to ensure the security and confidentiality of all participants’ data, and to minimise any risk to participants, family members and the community.

Researchers (including PhD students) interested in using the data, should contact Jenny Zhang (jenny.zhang@uq.edu.au) and an amendment PHA will be submitted to Queensland Health adding the researcher’s name in the application. Researchers must write a letter requesting access to specific data and attach their CV. Once the amendment PHA is approved, a policy agreement regarding data access must be signed by the researchers and then a deidentified dataset will be sent.