Australian Burden of Disease Study: Methods and supplementary material 2018

The ABDS 2018 measured health loss using a summary measure of health called the disability-adjusted life years (DALY). One DALY represents 1 lost year of ‘healthy life’ due to premature death, illness or disability, or a combination of these factors. This measure quantifies the gap between a population’s actual health and an ideal level of health in the given year – that is, every individual living in full health for his or her ideal or potential life span – and includes both fatal and non-fatal components.

A broad overview of the process for estimating DALY is shown in Figure 2.1.

Figure 2.1: Overview of disability-adjusted life year estimation process

The fatal component is measured using years of life lost (YLL) – 1 YLL represents 1 year of life lost (due to premature death). YLL measures the years lost between the age at which a person dies and an ideal life span according to a reference life table. Total YLL are influenced by both the total number of deaths, and the ages at which those deaths occur.

In the ABDS 2018, the ideal remaining expectancy varied at each age, but started with a life expectancy at birth of 86.0 years for both males and females. This ideal life span was drawn from the reference life table used in the GBD 2010 and 2013 studies, and was based on the lowest observed death rates at each age group from multiple countries (Murray et al. 2012).

See Estimating the fatal burden for more detail on YLL estimation.

The non-fatal component is measured using years lived with disability (YLD) – 1 YLD represents 1 year of life lost (due to the disabling effects of ill health). YLD measures the number of healthy years of life lost due to disease in the reference year. This is calculated by estimating the amount of person-time spent with a condition, multiplied by a disability weight which reflects the severity of the condition. Total YLD are influenced by the number of people with each disease, the time spent in less than full health, and the disability weights defined for each disease consequence. The disability weights used in this study were drawn from the GBD 2013 study, hereafter referred to as the GBD 2013 (see GBD 2013 Collaborators 2015), and represented the health loss caused by the consequences of each disease. Disability weights are further adjusted for comorbidity.

See Estimating the non-fatal burden for further detail on YLD estimation and use of disability weights.

As they use time as a common currency, the YLL and YLD can be summed to measure DALY: 1 DALY represents the loss of 1 year of healthy life.

DALY = YLL + YLD

When DALY are used to measure the burden of disease in a population in a time interval, they can be calculated in various ways: from an incidence, prevalence, or hybrid perspective. Each method produces a measurement of a different quantity. This study used the hybrid perspective for calculating DALY consistent with the ABDS 2011, ABDS 2015 and recent global studies. This calculates YLL from an incidence perspective (see Estimating the fatal burden for details) and YLD from a prevalence perspective (see Estimating the non-fatal burden for details). The main advantage of this approach is that all data needed to calculate DALY can be measured in the period in question.

Constructed this way, DALY can be thought of as an index of population health in a given year, providing a summary measure of the overall population health for the year being reported. This enables diseases, population groups and points in time to be compared.

Reference years 2018, 2015, 2011 and 2003

Based on the availability of data at the start of the study, 2018 was considered the most suitable choice for the primary reference year. It should be noted that some data used in the ABDS (mainly from surveys or epidemiological studies) related to periods earlier than 2018 as this was when the most recent survey or the most relevant epidemiological study was done. In such cases, modelling was required to adjust the counts or rates to 2018.

Although 2018 was used as the reference year of the study, more than 1 year of data was compiled and analysed in some cases to overcome small numbers or to smooth variability. For some estimations, it was also informative to look at trends over time.

There have been 4 previous Australian burden of disease studies with estimates published in ABDS 1996, ABDS 2003, ABDS 2011 and ABDS 2015. While overarching methods for estimating disease burden remained unchanged from the ABDS 2011, revision of some disease-specific methods in ABDS 2015 and ABDS 2018 lead to estimates that differed considerably from the ABDS 2011. Therefore, revision of 2015, 2011 and 2003 estimates were required to provide comparable Australian burden of disease estimates to assess changes over time. These revisions reduce the risk of users making erroneous comparisons between previous 2003, 2011 and 2015 estimates with those produced in ABDS 2018.

Reference populations

All Australian population-based rates for 2018 and 2015 were calculated using populations rebased to the 2016 Census (released 27 June 2017) (ABS 2017).

Population-based rates for 2011 were calculated using final population estimates from the 2011 Census (released 15 December 2016).

The Australian 2001 standard population (published 15 December 2016) was used for all age‑standardisation, as per the Australian Institute of Health and Welfare (AIHW) and ABS standards (ABS 2016).

Age groups

Analysis was done using as fine an age disaggregation as was supported by the data. For fatal burden, YLL were calculated using single year of age. For non-fatal and total burden, construction of YLD (and hence DALY) estimates were based on 5-year age groups of 0, 1–4, 5–9, …, 100+ for the national estimates. Where the available data could not directly support 5 year age groups, modelling was used to derive estimates at the required level of age disaggregation.

The reporting age groups were aligned to fit with existing reporting practices by age and sex to enable comparisons with other data, within the constraints of the quality of the underlying data.

Selection and classification of diseases

The list of diseases and injuries (referred to as the ABDS disease list) – and their organisation into disease groups – forms the analytical framework of the ABDS 2018, and underpins all estimates of deaths, YLL, YLD, DALY and risk-attributable burden. As the burden of each disease is estimated relative to every other disease specified in the study, this list forms the foundation of all analysis and reporting.

The ABDS disease list uses the following hierarchical framework:

Disease groups: 17 disease groups of related diseases or conditions – such as cardiovascular diseases, gastrointestinal disorders, or injuries – and one alternative reporting disease group (nature of injury instead of injury by external cause).

Diseases: 219 specific conditions or sets of conditions such as coronary heart disease, appendicitis, or poisoning, for which estimates of deaths, YLL, YLD, DALY and risk‑attributable burden were produced. These conditions are mutually exclusive (non‑overlapping) including two perspectives for reporting injuries: by external cause or nature of injury.

The ABDS disease list is collectively exhaustive, meaning it covers the full spectrum of disease and injuries (ABDS 2018 list of diseases, conditions and injuries and ICD-10 codes).

Selection of diseases and injuries

The ABDS disease list is an Australian-specific disease list developed to reflect the needs of health reporting and monitoring in Australia. For this study, the ABDS 2015 disease list was reviewed, and modifications made based on a set of inclusion criteria originally developed and applied in the ABDS 2015.

For inclusion in the ABDS 2018 disease list, the condition or injury must meet at least one of the following guiding principles:

Included in other studies’ disease (or cause) lists

Have been included in:

• the GBD study for 2017 or the ABDS 2015 (AIHW 2016a) unless its inclusion in the ABDS 2018 conflicted with other criteria.

Significant burden

• Be of significant burden to at least 1 age group or sex – defined as either more than 25 deaths or more than 500 inpatient events averaged annually over a 4-year period, or as having a ‘significant’ primary care impact, as determined by expert judgement (ensuring the list is not overcome with very minor conditions, for which it might be difficult or costly to assemble data).

Policy interest

• Be of substantial Australian or Indigenous health policy interest – defined as being the focus of current policy or professional attention, or thought to be increasing substantively in impact (which might be signalled by large increases in incidence or prevalence), or
• be the subject of an existing health monitoring activity within Australian or Indigenous populations, or
• be required for the analyses of risk factors that are of high policy interest.

Be able to be measured

• High-quality, relevant and recent epidemiological data needed to be available for at least 2 out of these key epidemiological variables: incidence, prevalence, survival or mortality of/from the condition.

Using these criteria, a final list of 219 diseases, conditions and injuries (including residual conditions – see ‘Residual conditions’ section) were selected and agreed on by the Australian Burden of Disease Expert Advisory Group to form the basis of the ABDS 2018. This includes 13 conditions describing the nature of injury used for alternative reporting (see Injuries for more detail).

As such, the ABDS 2018 disease list will differ from that used in other studies.

Residual conditions

The disease list is collectively exhaustive. Conditions that could not be individually specified are included in a residual category for each disease group. For example, the residual category ‘other musculoskeletal conditions’ are those musculoskeletal conditions not included in arthritis, gout, rheumatoid arthritis and back pain and problems. There are 32 residual (‘other’) categories distributed across the 17 disease groups and another 2 in the alternative reporting group for injuries (nature of injury). In the ABDS 2018, there are new diseases that were previously reported in residual groupings (see Box 2.1).

Conditions not includ​ed as specific diseases in the disease list

There were 3 key reasons for not including some conditions as specific diseases in the ABDS 2018 disease list:

• Scarcity of recent and/or robust data to reliably estimate prevalence in Australia in 2018 – these conditions could be incorporated into future burden of disease analyses should more recent or robust data become available. Examples include:

  • myalgic encephalomyelitis/chronic fatigue syndrome – although believed to be of significant impact, this condition is not monitored in Australia and recent robust data on incidence and/or prevalence are scarce. Although this was included in the ABDS 2003 as a separate disease, the data underpinning these estimates are now outdated. Myalgic encephalomyelitis/chronic fatigue syndrome was not separately estimated in global studies or the New Zealand Burden of Disease Study (NZBDS) 2006 (NZMOH 2013). In this study the burden of this condition is included in ‘other neurological conditions’.

  • fetal alcohol spectrum disorders (FASD) – although FASD is of policy interest, no national data source was identified. FASD was not separately estimated in GBD global studies but was separately estimated in the NZBDS based on hospitalisations (however, it was noted it would be an underestimate). In the ABDS 2018, the burden of FASD experienced by the child was grouped under the disease ‘brain malformations’ in infant & congenital conditions.

• The condition is the result of other underlying causes, or its burden is captured under other sequelae – these conditions do not fit within the mutually exclusive disease structure required for burden of disease analysis. Future analyses of these conditions might be possible by selecting corresponding diseases or sequelae. Examples include:

  • antimicrobial resistance – antimicrobial resistance includes many types of organisms (for example, staphylococcus) and types of resistance (for example, penicillin). Antimicrobial resistance was not included in previous burden of disease studies. Although it is of policy interest, and there are sufficient data for modelling, its outcomes were captured by other diseases already included in the study (for example, infectious diseases).

  • septicaemia – this is considered an intermediate, rather than underlying, cause of burden, and its impact was captured through the sequelae and the severity distributions for relevant diseases (for example, selected infectious, neonatal and maternal diseases).

  • heart failure – this is also considered an intermediate cause of burden, and its impact was captured through the sequelae and the severity distributions for relevant diseases (for example, cardiovascular disease, congenital heart disease).

• The condition was conceptualised as a risk factor – these conditions might not have been associated with health loss themselves, but place individuals at greater risk of other health conditions. Their impact is captured as burden attributable to various risk factors. Examples include:

  • osteoporosis – the health loss from osteoporosis is captured under falls in the injury disease group. The risk factor low bone mineral density was used in this study to estimate the proportion of falls attributable to osteoporosis (see low bone mineral density for more detail)

  • nutritional deficiencies – in the ABDS 2018, protein-energy deficiency and iron-deficiency anaemia are included as specific nutritional deficiencies in the disease list. Other nutritional deficiencies (such as diet low in calcium) are not included as diseases, but instead as risk factors for other diseases (see iron deficiency and dietary risk factors for more information).

Classification of diseases and injuries

To ensure that the disease list was both comprehensive and mutually exclusive, each included disease and injury had to be carefully defined. To ensure consistency between YLL and YLD estimation, the classification of each disease had to be suitable for both mortality and morbidity components.

As the internationally recognised and definitive set of codes to describe all health conditions, the International Classification of Diseases and Related Health Problems, 10th Revision (ICD‑10) (2010 version) (WHO 2016) was used to broadly define each disease in the disease list. To estimate YLL, ICD-10 classifications were used, but for YLD, classifications were adapted as necessary depending on the data that were available and appropriate for analysis (for example, the Australian modification ICD-10-AM was used for hospital separations data).

See Disease specific methods - morbidity for details of the specific classifications used for each disease group.

Mapping of ICD-10 codes to the disease list

The allocation of more than 12,000 ICD-10 codes to the 219 diseases in the ABDS 2018 disease list was based on the ABDS 2011 (AIHW 2016b) with expansion of some diseases. The ABDS 2011 disease list was informed by the code allocation used by the GBD 2010 study (hereafter referred to as the GBD 2010), the NZBDS 2006 and the ABDS 2003 (Begg et al. 2007).

To promote internal consistency and objectivity, the following principles were applied:

• Attribute the burden to the condition where the health loss was experienced (‘prevalence principle’). This principle was used mostly when mapping diseases or conditions that can be a long-term result of an earlier condition; diseases that are risk factors or sequelae for other diseases; or diseases that can be counted in more than one disease group. Examples include:

  • the burden from liver cancer or chronic liver disease due to hepatitis was counted where the condition manifested or was experienced (that is, in cancer or gastrointestinal conditions), not as a long-term sequelae of hepatitis. This is consistent with global studies and with the mapping practice for other conditions that are now known to be the result of previous infectious diseases.

  • the overlap in cardiovascular disease, chronic kidney disease and diabetes was dealt with by attributing the health loss to the condition experienced, rather than the underlying cause (for example, renal complications due to diabetes mellitus was counted under chronic kidney disease). The AIHW explored the overlap between these diseases to quantify their indirect impacts and collective burden. Results from these studies were published in the report Diabetes and chronic kidney disease as risks for other diseases (AIHW 2016c).

• Classify diseases according to Australian disease monitoring activities. Australian disease monitoring classifications were given priority over the GBD to provide better information for Australian health priority setting. For example, the GBD classified all neoplasms together, regardless of malignancy. In Australia, monitoring of neoplasms is restricted to malignant neoplasms, so they were classified separately to other neoplasms.

The proposed mappings of ICD-10 codes to diseases in the ABDS disease list were reviewed by disease specific expert groups before being finalised.

Assigning diseases to disease groups

Under the ABDS disease hierarchy, each disease is allocated to a single disease group. The allocation of particular diseases to a disease group affects the estimates of burden and ranking by disease group that are reported in the published analyses. Alternative disease group presentations of the ABDS 2018 results can be readily developed from the existing disease list. For example, gastrointestinal disorders do not include gastrointestinal infections, or gastrointestinal cancers, but the estimates for these diseases could be added to the gastrointestinal disorders group to obtain a broader picture of the burden for this area of interest.

For the most part, assigning diseases to disease groups relied heavily on the chapter structure of ICD-10. However, for a small number of diseases it was less straightforward, as they appeared potentially to bear some characteristics of more than one group. These diseases were allocated after discussion with experts from both potential disease groups, and, as with the prevalence principle, assigned according to where the health loss is actually experienced.

Major decisions referred to experts for advice included:

• suicide and self-inflicted injuries – the burden was included under injuries, consistent with ICD-10 coding and previous national and GBD studies.

• accidental poisonings involving drugs and alcohol (ICD-10 codes X41, 42 and 45) – the burden was included under injuries rather than substance use disorders, consistent with coronial assessment, on the basis that where the coroner found evidence of an underlying dependence, the cause of death would reflect this and be assigned to substance use disorders. The drug and alcohol experts expressed concerns about the reliability of distinctions between opioid overdose fatalities that are due to accidental overdose or those due to opioid dependence. There is evidence in Australian studies that most overdose deaths occur among people with a history of dependence, and very few deaths are deliberate. However, as the coding for X42 (Accidental poisoning by and exposure to narcotics and psychodysleptics [hallucinogens], not elsewhere classified) includes several drugs, not just opioids, this assumption would have to be made for those other drugs as well.

• gestational diabetes – the burden was counted in the reproductive & maternal disease group, rather than endocrine disorders, due to this condition only arising during pregnancy, and is consistent with previous national and GBD studies.

• cerebral palsy – the burden was allocated to the infant & congenital conditions disease group, rather than neurological conditions, as, in most cases, cerebral palsy is acquired in the prenatal and perinatal period and emerges as a leading cause of death for children aged under 5. As a sequela, cerebral palsy is acquired through several other infant & congenital conditions, such as birth trauma and birth asphyxia

• fetal alcohol spectrum disorders (FASD) – although counted under mental health and substance use disorders in the GBD 2010, the burden was assigned to infant & congenital conditions in the ABDS as the main sequelae are learning difficulties and disfigurement, and the burden is experienced by the child (not the mother).

• postnatal depression – the burden was not included as a separate disease in the ABDS due to data limitations. As available data did not distinguish whether the depressive disorder was associated with childbirth, postnatal depression was included in estimates for depressive disorders, within the mental & substance use disorders disease group. This is consistent with previous national and GBD studies.

Selection and assessment of data sources

All potential data sources to estimate disease burden (whether published or unpublished) were assessed for comparability, relevance, representativeness, currency, accuracy, validation, credibility and accessibility/timeliness (see Additional material for the guidelines used to direct data selection). Only data sources that met the guidelines were included in the study.

Potential data sources were required to: have case definitions appropriate to the disease or risk factor being analysed; be relevant to the Australian population; and be timely, accurate, reliable and credible. Where possible, national data sources, rather than sources relating to particular regions or subpopulations, were used.

Administrative data sources (for example, disease registers, hospitalisations) were evaluated for their level of ascertainment (how well the data correspond to the disease or sequela in question) and coverage (the proportion of the population included in the data).

Surveys were evaluated for their representativeness, potential selection bias, and measurement bias (validity and reliability of measurement).

Epidemiological studies were evaluated for the quality of their study design, their timeliness, credibility, representativeness, and sources of bias or error.

There are new data sources for many diseases in the ABDS 2018, notably greater use of linked hospital/deaths data.

The key data source used in estimating mortality is described in Estimating fatal burden, and key data sources used in estimating morbidity are listed in Estimating non-fatal burden.

Methodological choices specific to Indigenous estimates

Additional factors needed to be considered when calculating burden of disease estimates for Aboriginal and Torres Strait Islander people. As a general principle in the ABDS, the methods used to produce Indigenous burden of disease estimates were consistent with those used to produce national estimates. For example, the same reference life table, disability weights and disease list were used. However, it was not always possible to adopt completely consistent methods due to differences in data availability, data quality and population size and characteristics.

Indigenous under-identification

While in recent decades major improvements have been made to the quality and availability of information about Indigenous Australians, existing data are subject to several limitations regarding data quality and availability. These include under-identification of Indigenous Australians in administrative data sets (and changes in people’s inclination to identify as Indigenous over time), and lack of available data on the prevalence of certain diseases in the Indigenous population. Methods employed to address these issues in the ABDS are discussed in the relevant sections of this report on fatal and non-fatal burden.

Dealing with small numbers

An important consideration for Indigenous burden of disease is the robustness and reliability of estimates produced, and the level of disaggregation supported by the data, given the small size of the Indigenous population compared with the much larger non-Indigenous population.

To ensure validity of the results, the AIHW combined several years of data and/or age groups as necessary to produce Indigenous estimates. Additionally, the level of disaggregation used to report Indigenous estimates was broader than that reported for the total Australian population. This included collapsed age groups for those aged 0–4 and 85 and over.

Measuring the gap between Indigenous and non-Indigenous Australians

Direct age-standardisation was used to compare rates between Indigenous and non-Indigenous Australians, and to measure the gap in burden between the 2 populations. The direct method was chosen, following a series of sensitivity analyses undertaken by the AIHW, which looked at the impact and robustness of using the direct method compared with the indirect method on resulting Indigenous YLL estimates (see AIHW 2015 for more information). The direct method enables multiple comparisons (for example, disease by sex) and can be used for comparisons over time. A limitation of the direct method is that less reliable estimates can be produced when it is applied to a small number of deaths and prevalent cases; this should be kept in mind when interpreting gap results for less common diseases and conditions.

Age-standardised rate differences and rate ratios were reported as measures of the gap. Rate differences provide a measure of the absolute gap between 2 populations, while rate ratios are a measure of the relative gap between 2 populations.

For the most accurate estimate of the gap in disease burden between Indigenous and non-Indigenous Australians, comparisons have been made to estimates calculated for the non-Indigenous population. Estimates for the total Australian population should not be compared with those for Indigenous population.

Choice of population denominator for Indigenous estimates

In estimating the Aboriginal and Torres Strait Islander population for the years prior to each Census, the Australian Bureau of Statistics makes a number of assumptions regarding past mortality rates, migration and improvements in life expectancy. As such, several population backcast and projection series are produced in addition to the Estimated Resident Population for each Census year.

Following sensitivity analyses by the AIHW to look at the impact of using different Indigenous population denominators in burden of disease rate calculations, it was agreed to use the backcast population series based on the 2016 Census, which applies the Indigenous identification level in 2016 to earlier years. Using this backcast population for all reference years provides consistency between the denominators used for the Indigenous burden of disease estimates in the ABDS 2018.

For more information on these choices, see Impact and causes of illness and death in Aboriginal and Torres Strait Islander people 2018 (AIHW forthcoming 2022).

Methodological choices specific to sub-national estimates

Sub-national estimates include state/territory, remoteness categories and socioeconomic groups. These are defined as:

• state and territory classifications – the 8 Australian jurisdictions: New South Wales, Victoria, Queensland, Western Australia, South Australia, Tasmania, Australian Capital Territory and Northern Territory. Disaggregation by state/territory is well supported by the data, with the majority of data sources (except for epidemiological studies and small surveys) defining and reporting state or territory in a standard way.
• remoteness categories – based on the 2016 Australian Statistical Geographic Standard (ASGS) for 2018 and 2015 estimates, or the 2011 ASGS for 2011 estimates. The ASGS is divided into 5 remoteness areas: Major cities, Inner regional, Outer regional, Remote and Very remote. Remoteness areas aggregate to states and territories and cover the whole of Australia. Most major data sources, except for epidemiological studies and small surveys, were able to be broken down by remoteness area. This study reported estimates for 4 remotes areas: Remote and Very remote were combined.
• socioeconomic groups – presented as quintiles of lowest to highest socioeconomic position. Ideally, it would be better if detailed individual-level measures of socioeconomic characteristics were available in key data sources. But the most consistently available approach across the national data sources was the geographically-based proxy of socioeconomic group based on the relative socioeconomic characteristics of the area of residence, known as SEIFA (Socio-Economic Indexes for Areas). SEIFA is a measure of socioeconomic disadvantage developed by the ABS that ranks geographic areas in Australia according to relative socioeconomic advantage and disadvantage. The ABS broadly defines relative socioeconomic advantage and disadvantage in terms of ‘people’s access to material and social resources and their ability to participate in society’. The AIHW generally reports analyses of socioeconomic differences using SEIFA divided into population-based quintiles. It is also the standard for the majority of national agreement indicators. This approach ensures that, regardless of the underlying geographical unit, about 20% of the population is allocated to each quintile. SEIFA contains 4 indexes, with the Index of Relative Socioeconomic Disadvantage (IRSD) historically being the most commonly used at the AIHW for health-related analyses. For more information on SEIFA. SEIFA was only used for disaggregation of national estimates.

Sub-national methodology

Sub-national estimates were based on breaking down national estimates at a level of disaggregation (disease, sex and broad age group) supported by the underlying data, rather than being derived using separate data sources. This ensured that comparisons across each disaggregation were based on common data definitions, which is often not the case when sub‑national data sources are combined.

The preferred approach for sub-national estimates was to derive sub-national disaggregation directly from the primary data source using geographical identifiers. When this was not available, secondary data sources were used to identify health loss gradients between the sub-national regions that could then be applied to the national data. Lastly, when neither of these approaches were possible, the national sex/age prevalence rates were applied to the population structure of the sub-national unit. This assumed no difference in disease prevalence rates between sub-national and national populations.

Specific details on the methods used for sub-national estimates for mortality and morbidity are included in Disease specific methods – mortality and Disease specific methods – morbidity.

Key considerations

The validity of sub-national results is influenced by the availability and quality of data at the level of disaggregation, and by the population size in the various groups.

For state and territory estimates, analyses used the same age groups as the national analysis. For remoteness and socioeconomic group analyses, age groups were restricted to 5-year age groups 0, 1–4, 5–9, …, 85+ to overcome limitations with data.

Indigenous sub-national estimates

Indigenous sub-national estimates were considered reliable to calculate and report at the disease group level, but not at the specific disease level. This was due to:

• limited availability of Indigenous data for individual diseases at the geographical levels of interest
• limited availability of Indigenous identification adjustment factors at sub-national levels for relevant administrative data collections
• small numbers if Indigenous estimates were broken down at sub-national levels.

Indigenous sub-national estimates were considered adequate to report for 4 states and territories (New South Wales, Queensland, Western Australia, and the Northern Territory). Estimates were not calculated for Victoria, South Australia, Tasmania or the Australian Capital Territory due to small numbers of Indigenous deaths in these jurisdictions, and lack of suitable mortality adjustment factors. However, these jurisdictions account for the majority of burden in most cases (see Table 2.3).

Estimates for all 5 categories of remoteness were reported (Major cities, Inner regional, Outer regional, Remote and Very remote).

For Indigenous burden estimates by level of socioeconomic disadvantage, an Indigenous-specific index (the Indigenous Relative Socioeconomic Outcomes Index) (Biddle & Markham 2017) was used. This was considered to more accurately reflect levels of disadvantage in the Indigenous population than the SEIFA index used for the national component. As such, the Indigenous estimates by socioeconomic disadvantage were not compared with national estimates by socioeconomic disadvantage.

Indigenous sub-national estimates of YLL were calculated directly from mortality data (adjusted for Indigenous under-identification) using state/territory and remoteness specific adjustment factors.

Hospitalisation data (adjusted for under-identification), ABS health survey data (2018–19 Australian Aboriginal and Torres Strait Islander Health Survey; 2012–2013 AATSIHS biomedical data), or population proportions (depending on the disease group) were used to break down the national-level Indigenous YLD into subnational categories. Hospitalisation data were used for 10 disease groups, and health survey data were used for 5 disease groups for state/territory and remoteness estimates. A combination of Indigenous health survey data and data from the Australia and New Zealand Dialysis and Transplant Registry (ANZDATA) was used for 1 disease group (kidney & urinary diseases), and the subnational Indigenous population structure distribution data was used for the final disease group (skin disorders). For estimates by socioeconomic group, hospitalisation data were used for all disease groups, as Statistical Area Level 2 data (required to calculate the Indigenous Relative Socioeconomic Outcomes Index) were available from this data collection.

The data sources used to break down Indigenous YLD into subnational categories can be found in Table 2.2. The proportions used to break down Indigenous YLD estimates for each disease group can be found in tables 2.3 to 2.5.

State-level data were not generally used to build the national burden of disease estimates for the Indigenous population (that is, fatal burden estimates were calculated using national mortality adjustment factors, and non-fatal burden estimates were largely calculated using national prevalence estimates sourced from national data collections). As a result, Indigenous estimates reported at the national level are not subject to the same data quality issues as the state and territory estimates.

For more information on the methods used for Indigenous subnational estimates see Australian Burden of Disease Study: impact and causes of illness and death in Aboriginal and Torres Strait Islander people 2018 (AIHW forthcoming 2022).

Note: AATSIHS = Australian Aboriginal and Torres Strait Islander Health Survey, ANZDATA = Australia and New Zealand Dialysis and Transplant Registry, NATSIHS = National Aboriginal and Torres Strait Islander Health Survey.

Note: See Table 2.2 for data sources used for proportional splits.

Note: See Table 2.2 for data sources used for proportional splits.

Note: All proportions calculated from the NHMD.

Methodological choices specific to 2003, 2011, 2015 and 2018 estimates

Comparable YLL, YLD, DALY and attributable burden estimates were produced for each disease for the national population. Sub-national estimates for 2003 were not within the scope of this study.

As the 2003, 2011, 2015 and 2018 estimates are point-in-time estimates, their comparison with each other does not constitute a time-series analysis. Several issues must be considered before analysing and interpreting time trend data. A key issue is that 4 points in time can provide misleading information about changes over time – assuming that there is a straight-line trend between these 4 points might mask variation that exists but is not measured in this analysis, and results must be interpreted with this in mind. In addition, interpretation of changes over time also needs to take into account other aspects, such as the impact of confounders over time related to the estimates, and changes in metadata between reference periods. Any major changes between the previous studies and 2018 data that have an impact on the interpretation are highlighted in the relevant chapters in this report.

2015, 2011 and 2003 estimates

Where there were no changes in methods or data sources, the 2015 estimates from the ABDS 2015 were adjusted because the underlying population estimates were updated following the release of the 2016 Census. In contrast, the 2011 and 2003 estimates were kept the same because the underlying populations are based on 2011 census. If there were changes in methods or data source for 2018, the estimates for previous years were re-estimated using the new methods to keep comparability across all four years.

Specific details on methods for previous years estimates for mortality, morbidity and risk factors are included in disease specific methods – mortality, disease specific methods – morbidity and risk factor specific methods.

Indigenous 2011 and 2003 estimates

Issues relating to changing Indigenous identification over time and potential inconsistencies in identification in numerator data and population denominators have an impact on the comparability of Indigenous burden of disease rates over time. These issues also have implications on the choice of population denominator used for 2003 and 2011 Indigenous burden of disease estimates.

Where possible, adjustments have been made to account for changes in Indigenous identification over time in the numerator data used for rate calculations of disease burden. For example, Indigenous deaths and hospitalisations for 2003 and 2011 estimates were adjusted using factors based on identification levels relevant to these reference years.

The population denominator used for 2003 and 2011 Indigenous burden of disease estimates were consistent in terms of Indigenous identification with that used for 2018 estimates, which is important for assessing rate changes over time. Indigenous population estimates based on the 2016 Census were used, which applies the Indigenous identification level in 2016 to earlier years in the series, including for 2011 and 2003.

Additional material