Study design
A retrospective cohort with de-identified linked hospital and administrative data.
Data sets
The study used data from the NSW Admitted Patient Data Collection (APDC). The data collection includes all hospital separations in public and private hospitals in NSW and includes discharges, transfers and deaths. Fact of death was provided by the NSW Registry of Births, Deaths and Marriages (RBDM).
Study cohort
The study cohort comprised patients who: were aged 18 years and older at the time of index admission; were admitted to a NSW public hospital; discharged from hospital to the community; and had one or more of the following selected ICD-10 defined ambulatory care sensitive (ACS) chronic diseases as a principle diagnosis: diabetic complications, asthma, angina, hypertension, congestive heart failure (CHF) and/or chronic obstructive pulmonary disease (COPD) (Additional File 1). These chronic conditions were selected as they are highly prevalent among Aboriginal people and an admission to hospital relating to these chronic conditions is considered potentially avoidable through health promotion, preventative measures, or timely access to non-hospital care such as through community health care.[2, 3]
Sampling
The Aboriginal sample included all APDC patients who met the above eligibility criteria, had at least one record during the study period, and were documented as Aboriginal and/or Torres Strait Islander on any APDC record. A non-Aboriginal comparison sample was selected by using an equal number of randomly sampled patients who met the eligibility criteria and were not documented as Aboriginal and/or Torres strait Islander on any records. RBDM fact of death pertaining to the sample were included in the final dataset.
Data preparation
The APDC and RBDM data were provided in a de-identified format by the Centre for Health Record Linkage (CHeReL).[19] Duplicate records were excluded. Periods of care were defined as overlapping episodes of care and sequential transfers were considered in order to define the start and end dates for the period of continuous hospital care. A period of care ended with discharge from hospital. If a patient was discharged and then readmitted the same day, this represented the next period of care. Periods of care in the year of an individual’s death were included in the analysis. Periods of care are referred to as admissions for the remainder of this paper. Two datasets were prepared for analysis: an un-aggregated database of admissions with a defined ACS ICD code (n=31,836) and an aggregated dataset of counts of the number of avoidable admissions for each patient by financial year, and whether they were planned or unplanned admissions (n=22,802).
Exclusions
Private hospital admissions were excluded from the cohort. It was a priori acknowledged that most private hospital admissions are planned as very few private hospitals have emergency departments, and private patients who experience frequent admissions would have a different sociodemographic profile to those in public hospitals. Planned admissions were excluded from the analysis.
Analysis variables
For each individual the following outcomes were used: 1) the number of avoidable admissions (defined as an unplanned admission with a principal diagnosis of an ACS chronic condition) for an individual in each financial year; 2) whether or not an individual experienced three or more avoidable admissions in each financial year they were observed over the study period (compared with one to two avoidable admissions). Unplanned admissions were coded as an “emergency status recode” in the APDC.
Patient demographics included in the final dataset were gender, age, Aboriginal status and marital status. The Accessibility/Remoteness Index of Australia (ARIA) and the Index of Relative Socio-economic Disadvantage (IRSD) quintile were calculated. ARIA is an Australian Bureau of Statistics measure of remoteness [20] and the IRSD is a measure of socio-economic status derived from the economic and social conditions within geographic areas.[21] The Charlson Co-morbidity Index (CCI) was also calculated [22] which provided a measure of the risk of mortality from comorbidity during the next 12 months. Length of stay was also included.
Statistical analysis
The denominator for the analysis was all avoidable admissions which met the eligibility criteria. At the admission level (unaggregated data), chi-square and t-tests were used to examine crude associations between Aboriginal status and sociodemographic, disease and admission factors. Then at the patient level (aggregated data), the yearly means of avoidable admissions were calculated by Aboriginal status and financial year. Chi-square tests were then used to examine associations of the proportion of individuals with three or more avoidable admissions compared with one to two avoidable admissions by Aboriginal status and financial year. Multivariable analyses were conducted using the aggregated data. Firstly, a Poisson regression model was used to examine the association of the number of avoidable admissions and Aboriginal status controlling for age, sex, marital status, financial year, IRSD, ARIA and restricted to patients aged ≤75 to account for the younger age structure of the Aboriginal patients. Secondly, a logistic regression model was used to assess the association of three or more avoidable admissions compared with one to two per financial year and Aboriginal status, controlling for age, sex, marital status, financial year, IRSD, ARIA and restricted to patients aged ≤75. To examine any differences in yearly trend between Aboriginal and non-Aboriginal people, an interaction term for Aboriginal status and financial year (as a categorical variable) was included in both final models, followed by a post estimation Wald test of the interaction term. The model was also fit without the interaction term and a post estimation Wald test was used to test the significance of the financial year term. A sensitivity analysis was used to determine any potential differences in results when indices ending in death were excluded. Stata software was used for all analyses.[23]
Ethics approval
The NSW Aboriginal Health and Medical Research Council Ethics Committee and the NSW Population & Health Services Research Ethics Committee provided ethical approval for the study.