Queensland Cardiovascular Data Linkage (QCard): A population-based cohort study

Study design

QCard is a longitudinal patient-level cohort linkage study of all CVD hospitalisations in 2010 with subsequent hospitalisations over a six-year period in Queensland, Australia. That is, the QCard will utilise secondary data that encapsulates all primary, secondary and tertiary healthcare used by all cardiac patients who were first hospitalised in 2010 and follow up utilisations until December 2015.

Study setting and population Queensland is the second-largest in the geographical area and the third most populated state in Australia and has an area of 665,615 square miles ranging from -10 to -30 latitude. The QCard data will include all patients who were admitted to hospitals with a diagnosis of CVD in 2010 with subsequent admissions until December 2015 in Queensland (QLD), Australia. The cohort will represent all CVD admissions in QLD during the study period. We will use ICD-10-AM (International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Australian Modification) codes from the admission data to define cardiovascular disease as follows: All cardiovascular diseases: I00-I99, Acute Coronary Syndrome (ACS): I200, I210-I214, Unstable Angina: I200, ST-Elevation myocardial infarction: I210-I213, Non-ST Elevation myocardial infarction: I214, Chronic Heart Syndrome (CHS): I20-I25, Heart Failure: I50, Cerebrovascular Disease (CeVD): I60-I69, and Chronic rheumatic heart disease: I05-I09. The study is expected to have a sample of 200,000 first-ever cardiac hospitalisations²², with a survival rate of approximately 71%⁴.

Inclusion criteria: Any CVD-related inpatient admission to a public health facility in 2010 for people aged 18 and above.

Exclusion crtieria: patients aged under 18 or admitted to private health facilities as private patients. Non-admitted patients were also excluded – that is those attending outpatients or the emergency department who were not admitted. In addition, participants were excluded if they had a CVD-related inpatient hospitalisation before 2010.

Ethics

Ethical approval for the QCard study was obtained from the Human Research Ethics Committee (Ref No: 2017/001) by Griffith University and Queensland Health. Data will be stored and accessed using the Secure Unified Research Environment.

Individual patient consent was waived for this study as it involved no direct contact with, or identification of, patients. Access to data was authorised under the Public Health Act of Queensland. Only un-identified, and non-reidentifiable, data were obtained.

Data sources

The data required for the study were submitted to data custodians in each of the 16 Hospital and Health Services in Queensland for approval. After receiving approval from the data custodians, the data were extracted and provided by the Statistical Unit of Queensland Health. The cohort data will be extracted from the Queensland Hospital Admitted Patient Data Collection (QHAPDC), Emergency Department Data Collection (EDDC), Registrar General Deaths Database (RGDD), comorbidity, and costs database. The cohort will also be linked with national data on the Medicare Benefits Schedule (MBS) and the Pharmaceutical Benefits Scheme (PBS) from the Australian Institute of Health and Welfare (AIHW). These data sets will contain information about types of medication and health services used, total expenditures and out-of-pocket contribution by patients for services, imaging and diagnostic procedures and prescription medicines received in the community. Table 1 summarises the data sources the expected variables available in them.

Environmental data

External data on environmental factors will be linked using residential postcodes of patients. All study participants have their postcodes recorded in the data; hence, we will be able to identify inequality at all geographical levels including the 78 local government areas (LGAs) in Queensland, Statistical Area Level 3 (SA3), Statistical Area Level 4 (SA4), State Suburb (SSC) and Postal Area (POA) levels. Specifically, environmental factors on air pollution will be sourced from Queensland Government Department of Environment and Science (QGDES), ambient temperature from Australian Bureau of Meteorology (BoM), and macroeconomic factors and socio-economic status (SES) from Australian Bureau of Statistics (ABS). These data will be used to investigate the impact of environmental shocks and conditions on healthcare use and survival time of cardiac patients. Effects of macroeconomic fluctuations and policy changes on health services usages and outcomes are also investigated.

Hospital characteristics

Data on hospital characteristics such as type of hospital (public/private), number of beds, locations and types of funding (block vs activity-based), peer group codes (acute, sub-acute and non-acute) and remoteness index will be collected. Such data will be sourced from Hospital and Health Services (HHSs), Independent Hospital Pricing Authority (IHPA) and Australian Institute of Health and Welfare (AIHW) and will cover 2010 and then annually thereafter.

Data linkage

Data sets will be linked using the individual identifier code, generated by the AIHW. The episode identifier code, generated by the Queensland Health Statistical Analysis Unit, and date of services will also be used to match data sets. In particular, individual and episode identifier codes will be used to link admissions data with the cost and sub-acute and non-acute data sets (see Figure 2). Data from the death registry were linked using only the individual identifier. Admission data will be linked with the morbidity data using the individual and episode identifiers.

Figure 2. The data linkage process.

The selected data sets will also be linked with the MBS and PBS data. The MBS has records of federally subsidised health services accessible to all eligible Australians. The PBS has records of federally subsidised prescription medications accessible to all eligible Australians. The MBS and PBS data contain detailed information of MBS schedule fee, MBS fee charged, MBS benefits paid by the Commonwealth government, PBS patient contribution, PBS benefit by the Commonwealth government as well as the quantities of drugs dispensed. We will also link external data, including socio-economic index, macroeconomic indicators and meteorological information to the main data set using residential postcodes.

Finally, postcode-level data on socio-economic and environmental variables will be linked with the main data using the individual postcodes and time (month and year) of admission or separation. Hospital level variables will also be merged unto the admission data using hospital names. These external data sets will not be at the individual levels but based on place of usual residence, hospitals of visits, and the time of hospital event.

To protect the privacy of patients, an individual probabilistic identification code, generated by the AIHW, will be used. The data will be stored and analysed using the Secure Unified Research Environment (SURE). Only authorised researchers can access the data remotely using password and security tokens with time-synchronised one-time passwords. No data will be exported outside the SURE. Also, results of data analyses will be screened by the SURE personnel to ensure no identifiable information is included before exporting to an external gateway.

Main outcome measures

The outcomes that will be investigated in this study are put into primary and secondary outcomes: The primary outcomes are mortality, time-to-death, readmission and days alive and out of the hospital. Mortality in the study will be measured as deaths after the first CVD hospitalisation. This outcome will be traced from the RGDD database. Using the QHAPDC and RGDD data, we will be able to measure survival time by looking at when a patient was first hospitalised and the date of death. Alive and out of hospital will adjust the patient’s survival time for the number of days spent in the hospital.

The secondary outcomes will be healthcare utilisation across all care sectors. More specifically, indicators such as general practitioners (GP) visits, use of cardiac-specialised services (which will be traced from the Medical Benefits Scheme (MBS) data), emergency department and inpatient hospital admissions, in-hospital adverse events, and pharmaceutical use from the Pharmaceutical Benefits Scheme (PBS) data.

Data analysis and statistical plan

The study will adopt two analytical approaches. The first analytical strategy will be to explore the extent to which socio-economic and spatial inequality permeates among cardiac patients, as shown in Figure 1. Techniques as Bayesian Spatial Analysis (BSA) will be used to investigate the extent to which spatial variations across the 78 local government areas in Queensland associate with variations in health service utilisation and mortality among cardiac patients. The number of deaths, admissions and GP visits in each LGA-period-sex-age group will be specified using a Poisson model. The Poisson model can estimate LGA level deaths and use of health services relative to that expected if all LGAs have the same deaths and use of health service as the national rates. In addition to the BSA, the Maximum Likelihood Estimation (MLE) techniques such as the logit and probit model will be employed to investigate the impact of socio-demographic factors (such as gender, age, and ethnicity) and cardiac-specific conditions (such as CHD, HF, AF etc.) on healthcare use and outcomes among cardiac. These techniques will enable us to identify most the most vulnerable patients who may require targeted public policies.

The second analytical approach will seek to explore the impact of environmental factors on health outcomes. Particularly, environmental effects on the survival time of cardiac patients and will utilise parametric and semi-parametric survival models such as the Cox proportional hazard model, Weibull, Gompertz, and Exponential. By using both parametric and semi-parametric survival models, the study will be able to provide reliable estimates on the impact of environmental factors on survival time under different assumptions placed on the baseline hazards of cardiac patients. In addition, generalised linear models with relevant link function and distributional assumptions will be used to estimate the impact of hospital characteristics and environmental factors on other outcomes (e.g., readmissions) and healthcare use such as hospital admissions, GP visits, healthcare costs and drug use. Unobservable characteristics across patients will be mitigated using a random-effects approach, while a fixed-effect approach will be used to address unobserved characteristics at community or health service levels. These approaches will reduce potential biases by unobserved or missing variables (confounders).

Data will be analysed using STATA 15.0²³ or subsequent upgrades and R 3.6.1²⁴.

Dissemination of information

Findings of this study will be disseminated via publication in peer-review journals, conference presentations, policy briefings and public lectures.

Study status

At the time of writing, all health service data have been obtained and linked. A baseline paper that describe characteristics of the cohort is in preparation. Also environmental data on air pollution and ambient temperature have been obtained from the Queensland Government Department of Environment and Science (QGDES) and Australian Bureau of Meteorology (ABM) respectively and are being analysed. Collaborations with three other institutions in Australia have been developed to investigate a wide range of research topics from this cohort.

Underlying data

No data are associated with this article.