## Years of life lost (YLL)

### Estimating the fatal burden

Expressed as years of life lost (YLL), fatal burden is a measure of years lost due to premature death. Analysis of fatal burden takes into account all deaths that occur in a population during a reference period. In the ABDS 2018, YLL estimates were based on deaths that occurred in the reference years: 2003, 2011, 2015 and 2018.

Deriving YLL requires both:

• mortality data – the actual number of deaths and the ages at which those deaths occurred; and
• a reference life table – a measure of life expectancy at each age to derive the years of life lost at each age.

#### Box 3.1: Key terms used in this chapter

redistribution: A method in a burden of disease study for reassigning deaths with an underlying cause of death that is not in the study’s disease list. Typically, the deaths reassigned include those with a cause that is implausible as an underlying cause of death, those with an intermediate cause in the chain of events leading to death, or those for which there is insufficient detail to ascertain a specific cause of death.

reference life table: A table that shows, for each age, the number of remaining years a person could potentially live – used to measure the years of life lost from dying at that age.

YLL (years of life lost): measures years of life lost due to premature death.

### Overview of methods

YLL measures the impact of dying prematurely; that is, the fatal component of burden of disease. YLD (discussed in Estimating the non-fatal burden) represents the non-fatal component.

The first step for estimating YLL is to compile all deaths by age and disease. Deaths are aligned to the study’s disease list using the cause of death.

YLL is then calculated for each disease using single year of age at death. Each death is weighted according to the remaining potential life expectancy at that age of death using the reference life table.

The weighted deaths are summed, and the result is the total number of years of life lost. For YLL from all causes, this is described mathematically as:

### Mortality data

Australian deaths data are collected through a vital registrations system. This is a system collecting and maintaining records of life events – such as births, deaths and marriages – by a government authority. In Australia, this is done by the Registrars of Births, Deaths and Marriages in each state and territory.

Information on causes of deaths nationally is sourced from the Registrars of Births, Deaths and Marriage in each state and territory and from the National Coronial Information System managed by the Victorian Department of Justice and coded to the International Classification of Disease (ICD) by the Australian Bureau of Statistics (ABS). The AIHW website About our data - Deaths Data provides detailed information on the registration of deaths and coding of causes of death in Australia (AIHW 2018a). The completeness, accuracy and coding of these data are described elsewhere (ABS 2018a). The deaths data are collated by the ABS into an administrative data set for statistical analysis. The AIHW houses a set of these data in the AIHW’s National Mortality Database (NMD). The data quality statements underpinning the AIHW NMD can be found in the ABS’s quality declaration summary for Deaths, Australia and Causes of death, Australia.

All deaths data used in the ABDS 2018 were extracted from the AIHW’s NMD. This is a register of all deaths in Australia since 1964, sourced from the cause of death unit record files as described above. The database comprises information about the causes of death and other characteristics about the person, such as sex, age at death, Indigenous status and area of usual residence.

Australian mortality data are believed to be virtually complete, so no adjustment needs to be done to account for missing death records. Despite completeness, causes of death that do not directly align to the study’s disease list need to be reassigned to a disease in the list (see ‘Redistribution of deaths’).

#### Mortality data in ABDS 2018

Cause of death data for deaths occurring in 2003, 2011, 2015 and 2018 were used for this analysis. Deaths for the four reference years were extracted from the NMD for deaths registered in 2003 up to and including deaths registered in 2019. As a result, the analysis set includes deaths that occurred in 2018 but were not registered until 2019; on average, between 4% and 7% of deaths that occur in a given year are not registered until a later year – most of these in the following 2 years (ABS 2019).

Deaths for the 2003, 2011 and 2015 reference years are almost all (at least 99%) based on a final version of cause of death data and most (95%) for 2018 are from a revised version of data. Since 2006, deaths certified by a coroner undergo revision and causes of death may be updated, pending the status of coroner investigation. As such, some cause of death information is subject to change. The ABS revisions process is described in detail elsewhere (ABS 2019).

#### Missing age and sex

Age at death is missing from some records in the mortality database. As age at death is required to estimate YLL, death records missing this data item were coded according to the median age at death for all deaths in the same sex-cause group.

There were no deaths with missing sex information for the reference years used in YLL calculations.

#### Indigenous identification

Due to small numbers, analysis of indigenous mortality was an average of the three years around the reference year. A separate calculation of non-indigenous burden was calculated with the three years of data.

### Aligning causes of death to the ABDS disease list

Having first assembled the deaths that are to be counted when calculating YLL, the causes of those deaths are then ascribed to diseases in the ABDS disease list (as described in Overarching methods and choices for ABDS 2018).

Deaths data used in the ABDS 2018 are coded to the ICD-10 (ABS 2019; WHO 2016). The procedure for assigning ICD-10 coded death records to items in the ABDS disease list is set out in the next section.

Some ICD-10 codes could not be classified directly to a specific disease in the ABDS disease list. To include these deaths in the calculation of YLL, they were redistributed using methods described in the section ‘Redistribution of deaths’.

It is important to note that the alignment of ICD-10 codes to diseases in the ABDS disease list might not be the same as alignment to the disease lists used in other burden of disease studies. In particular, a disease in the ABDS disease list might have the same label but comprise different ICD-10 codes compared with other studies’ disease lists. Table 2.1 provides a list of ICD-10 codes for each disease used for the estimates of fatal burden in the ABDS 2018.

### Redistribution of deaths

#### Identifying deaths for redistribution

Some ICD-10 codes are not appropriate or valid causes of death for burden of disease analysis. Some examples are:

• causes considered implausible as the underlying cause of death (such as hypertension and paraplegia)
• intermediate causes that have a precipitating cause (such as septicaemia and pneumonitis)
• immediate causes that occur in the final stages of dying (such as cardiac arrest and respiratory failure)
• causes that are ill-defined or unspecified, such as ill-defined digestive diseases and unspecified diabetes

Despite their overall high quality, Australian deaths data are affected by these issues. To quantify their contribution to the fatal burden, deaths coded to these underlying causes must be reassigned to one or more of the diseases (target diseases) according to what could be a more probable underlying cause. This process, referred to as ‘redistribution’ ensures that all the deaths in the reference year, hence all years of life lost, are counted in calculating YLL and is undertaken using the methods described.

#### Redistribution groups

The ICD-10 codes identified for redistribution were firstly assigned to redistribution groups. Each group was redistributed as a whole to the same range of target diseases. For example, non-specific digestive cancers formed one redistribution group and were reassigned to digestive cancers only. All deaths assigned to a group were redistributed using the same algorithm.

The redistribution groups used in the ABDS 2018 largely align with those used in the ABDS 2015. The table below shows the ABDS redistribution groups, target diseases and method for redistribution. The method by which each group was redistributed depended upon the level of available evidence.

#### Methods for redistribution

Deaths identified for redistribution were reassigned to one or more diseases in the disease list using statistical algorithms. Each death identified for redistribution may be reassigned in portions to multiple diseases.

The redistribution methods used in burden of disease studies have been refined over time, and algorithms have been developed and improved to redistribute deaths coded with inappropriate or invalid codes, by exploiting available evidence of a plausible alternate cause of death. The ABDS 2018 has extended these methods using Australian-specific data and Australian-specific direct evidence.

Three methods were used for redistribution in the ABDS 2018:

• Direct evidence: This method uses direct evidence about particular deaths or causes of death—obtained through data linkage studies or extracted from sources other than the NMD—to ascertain probabilities of a more plausible cause of death.
• Indirect multiple causes of death (MCOD): This method uses tabulations of the underlying cause of death where the cause to be redistributed is reported as an associated cause of death. The frequency distribution of the corresponding underlying causes of death informs the redistribution algorithm. For example, the algorithm for pneumonitis redistribution was provided by the frequency distribution of the underlying cause of death for all deaths that included pneumonitis as an associated cause of death. This method was used for frequently occurring causes of death, and where supported by the mortality data (for example, septicaemia, pneumonitis and hypertension).
• Proportional redistribution: This method reassigns deaths across a specified range of target diseases according to patterns of causes of death observed in the mortality data set for the disease list. Target ranges can be prescribed (for example, by narrowing the range of target diseases to injuries only). This method has the advantage of being conceptually simple and easy to implement, but it is relatively blunt, as the patterns of causes observed in the mortality data set might not reflect which underlying causes of death are more or less probable for the particular redistribution cause under consideration.

Direct evidence was preferred where it was available, followed by indirect MCOD (or a combination of both). In the ABDS 2018, 87% of redistribution was based on one of these methods. Proportional allocation was used only when neither of these methods could generate sufficient information to develop an algorithm; only a small proportion of redistributed deaths (13%) were redistributed using this method (Table 3.1).

Changes made to ABDS 2018 YLL calculations since ABDS 2015 include:

1. C26.0 deaths were assigned to bowel cancer, instead of being redistributed as part of the ABDS algorithm for ill-defined digestive cancers. This aligns with cancer mortality reporting and practice by the ABS.
2. Updated algorithms were used for cancer of other and ill-defined digestive organs (C26–excluding C26.0) and cancers of ill-defined, secondary unknown primary sites (C76–C80). Their redistributed were based on direct evidence from the Western Australian Cancer Registry.
3. Updated algorithm for septicaemia. Changes to selection rules for coding causes of death in recent years have allowed more chronic conditions, such as cancers, coded to Part 2 of the death certificate (associated causes), to be selected as the underlying cause of death when septicaemia appears in Part 1 (underlying cause) of the death certificate. Following discussion with mortality data experts, deaths recorded with septicaemia as the underlying cause were not redistributed to selected chronic conditions, but instead to more acute conditions, such as urinary tract infections.
4. Redistribution of ICD-10 code X59 Exposure to unspecified factor. Previously these were redistributed proportionately across injuries. Using similar methods to AIHW injury reports, we used associated causes of death (fracture codes) to identify additional falls (2,766 deaths), thereby resulting in less X59 deaths being needed to be redistributed (727 deaths).

#### Impact of redistribution

Disease-specific YLL are influenced by the causes of death identified for redistribution, and by the methods used to reassign these to another disease. Redistribution can have an impact on the number of deaths classified to a disease, as well as the number of YLL from that disease. In the ABDS 2018, 15,675 deaths were identified for redistribution in the 2018 reference year, equating to 201,561 YLL. This amounted to 9.8% of deaths and 8.5% of YLL. The number and per cent of deaths redistributed and the associated YLL for each reference year are in the table below.

The number of deaths identified for redistribution varied with age (see table below). These generally followed the patterns of age at death for all causes of death tabulations for Australia. For example, most redistributed deaths occurred among older people.

Table 3.4 shows the number of deaths classified to disease groups before and after redistribution. The largest numbers of deaths gained by redistribution were for:

• cardiovascular (5,163 more deaths, an increase of 14%)
• cancer (3,893 more deaths, an increase of 8.7%)
• endocrine (1,762 more deaths, an increase of 76%).

Note the large apparent ‘gain’ in deaths for endocrine disorders was due to deaths coded to unspecified diabetes being reassigned to type 1, type 2 and other diabetes.

The largest proportional gains, other than described above, were for:

• skin (145 more deaths, an increase of 25%)
• gastrointestinal (1,038 more deaths, an increase of 19%)
• kidney and urinary diseases (449 more deaths, an increase of 14%).

The impact of redistribution on YLL is also shown in Table 3.4. The largest number of YLL gained was for:

• cancers (54,912 more YLL, a 7.3% increase)
• cardiovascular (47,763 more YLL, an 11% increase)
• injuries (29,023 more YLL, a 9.2% increase).

Other large percentage gains in YLL were for:

• endocrine disorders (24,187 more YLL, a 78% increase)
• skin (1,282 more YLL, a 23% increase)
• mental and substance use disorders (1,366 more YLL, a 14% increase).

Note that the majority of these increases were based on targeted redistribution using direct evidence or indirect MCOD. To illustrate the method underlying the redistribution of deaths and its impact, Box 3.2 steps through the number and type of deaths that were redistributed into the cancer disease group for 2018 YLL estimates.

#### Box 3.2: How redistribution works

This box explains the redistribution process, showing, as an example, where additional cancer deaths came from as a result of redistribution.

Table 3.4 shows 44,777 deaths were coded to a cancer in the ABDS disease list. After redistribution, there were 48,670 cancer deaths, reflecting a gain of 3,893 deaths, or an additional 8.7%.

Table 3.1 shows that non-specific cancer deaths were reassigned to specific cancers using the direct evidence method, and that the target diseases were all in the cancer disease group. In 2018, 2,917 deaths were coded to a non-specific type of cancer, and 159 deaths were coded to a non-specific digestive cancer. So, in total, 3,076 non‑specific cancer deaths were identified for redistribution into a cancer cause.

So far, 79% of the overall gain in cancer deaths (3,076 out of the overall 3,893) has come from deaths initially coded to (non-specific) cancer-related causes, which have been redistributed into (specific) cancers in the ABDS disease list.

Table 3.1 also shows a further 1,506 deaths (initially coded to ‘all other non‑specific, intermediate and immediate causes’) were identified for redistribution that would be reassigned using the proportional allocation method across the whole range of ABDS diseases. A proportion of those deaths consistent with the proportion of cancer deaths (identified pre-redistribution) were reassigned to cancers in the ABDS disease list. As can be seen from Table 3.4, pre-redistribution, 28% of deaths were cancers, so about 28% of the 1,506 deaths (equivalent to around 423 deaths) were also redistributed to a specific cancer.

The foregoing redistribution steps account for around 90% of the overall gain in cancer deaths (3,076 plus 423 deaths).

The remaining 10% of the gain (393 cancer deaths) came from other redistribution causes where cancer was in scope as a target disease. For example, a proportion of septicaemia and pneumonitis deaths could be reassigned to a specific cancer in the ABDS disease list, provided there was evidence in the multiple-causes-of-death data of a combination of septicaemia or pneumonitis with a specific cancer cause. The redistribution groups and methods that have cancer in scope of target diseases are shown in Table 3.1.

### Reference life table

#### Life expectancy and life tables

The measure of life expectancy shows how long, on average, a person is expected to live, based on current age- and sex-specific death rates in the population. It is a summary measure commonly used to describe the health of a population. It specifies the remaining life expectancy at each age, with life expectancy at birth (the number of years of life that a person born today can expect to live) being the most commonly used. For a given country, estimates of life expectancy are derived from its actual life tables, which summarise the observed pattern of mortality and survival in the population.

YLL is an estimate of years of life lost due to premature death, and so has the character of a ‘health gap’ measure. As such, it requires an aspirational or potential life span to be able to quantify the gap between the current observed mortality and the counterfactual scenario where all mortality is averted until very old age.

Burden of disease studies use a reference life table, which corresponds to the aspirational or maximum life span for an individual in good health. It is typically more favourable than the actual life table of the population being studied, because it can be used across population groups and over time. It is used to produce estimates of life expectancy at each age, so that the number of years of life that are lost from dying at a specific age can be derived. For example, if the remaining potential life expectancy of a person aged 55 is 30 years (that is, at 55 a person could potentially, based on the reference life table, live to 85), then a death at 55 represents a loss of 30 years of life.

#### Choice of reference life table

The choice of reference life table will affect burden of disease estimates. Other things being equal, a reference life table with longer potential life expectancies at all or most ages will result in greater YLL. Applying the same reference life table across multiple settings enables comparison between population groups and across time.

The ABDS 2018 uses the standard reference life table used in the GBD 2010 and 2013 (Murray et al. 2012) when calculating YLL for the Australian and sub-national populations. The standard reference life table has a life expectancy at birth of 86.0 years.

More recent global estimates of YLL are based on a newer life table – the Theoretical Minimum Risk Life Table (TMRLT) (GBD 2017 Causes of Death Collaborators 2018). This life table is based on the lowest observed age‐specific mortality rates from locations with total populations greater than 5 million in 2016. From this life table, life expectancy at birth is 87.9 years and 1.6 years at age 105 (the limit of the standard reference life table) and 1.4 years at age 110.

When preparing this report, the TMRLT was only available in an abridged format; that is, where life expectancy is reported for five-year age groups. YLL estimates are best made using a life table that describes life expectancy at each single year of age. Using an abridged version results in less accurate YLL (unpublished AIHW analysis of the NMD), therefore the standard reference life table was used for calculating YLL in the ABDS 2018. This is consistent with the previous ABDS.

#### GBD standard reference life table

The GBD 2010 standard reference life table was derived from worldwide experience of mortality rates (Murray et al. 2012). For each age, the GBD selected the lowest age-specific death rate observed in any of the countries the study covered, excluding those with very small populations. The result is a hypothetical life table based on the most favourable age‑specific mortality experienced anywhere. It shows potential life expectancy at any age; in particular, it shows potential life expectancy at birth to be 86.0 years for both males and females. Table 3.5 shows the GBD standard life expectancies for each age at death.

Important features of this reference life table are that it:

• is aspirational – that is, it reflects the lowest observed death rates to construct a measure of potential maximum life span
• applies to all population groups – that is, it assumes the same aspirational life expectancy for any population group. It is the same for males and females, and for residents of major cities and very remote areas, assuming no difference in the survival potential of any of those groups.

The estimates of potential life expectancy in the GBD standard reference life table are different to that for the Australian population derived by the ABS from actual Australian mortality rates.

The GBD life table represents a longer life span than the Australian life tables. The life expectancy for Australian males and females at birth in 2017–2019 was 80.9 and 85.0 years, respectively – lower than the aspirational life expectancy of 86.0 years used in both the GBD and the ABDS. Life expectancies for Australian males and females were also lower than the GBD standard in 2010–2012 (79.9 and 84.3 years, respectively) and in 2002–2004 (78.1 and 83.0 years, respectively). For comparison, life expectancies in the GBD 2010 standard life table and for the Australian population for 2002–2004, 2010–2012 and 2014–2016 are shown for selected ages.

### Indigenous mortality data

Indigenous mortality data were sourced from the NMD in the same way using records identified as Aboriginal, Torres Strait Islander or both.

#### Dealing with small numbers

The number of deaths due to any particular cause varies from year to year. Fluctuations are more noticeable for diseases that are less common, and the instability is yet more severe for Indigenous deaths.

To reduce the impact of random fluctuations, Indigenous YLL estimates were based on the annual average of 3 years of deaths data. For the 2003 reference year, deaths were averaged from deaths occurring in 2002, 2003 and 2004. For the 2011 reference year, deaths were averaged from 2010, 2011 and 2012. For the 2018 reference year, deaths were averaged from 2016, 2017 and 2018.

Every year, a number of deaths of Aboriginal and Torres Strait Islander people are not identified as such when registered (ABS 2013a). This might arise from the non-reporting of a deceased person’s Indigenous status on the death registration form, or from incorrect identification of a deceased person’s Indigenous status (recording a person as non-Indigenous when they are Indigenous, and vice versa). The net effect is an under-identification of Aboriginal and Torres Strait Islander people in the deaths data.

Adjustment factors to account for Indigenous under-identification in death registration records have been produced from national and state/territory data linkage studies. These studies include the ABS Census Data Enhancement Study (CDE) (ABS 2013c; 2018c) and the AIHW Enhanced Mortality Database (EMD) study (AIHW 2012; 2017).

Based on the results from a series of sensitivity analyses on the impact of using the different mortality adjustment factors, and discussion with the study reference group, the approach taken for the 2018 reference year was to use adjustment factors from the ABS 2016 CDE Study (ABS 2018c) to adjust Indigenous deaths for YLL estimates and gap measures. The ABS adjustment factors take into account under-identification in both mortality and population data and therefore, in theory, provide consistency in the numerator and denominator used in Indigenous YLL calculations. It should be noted, however, that while the ABS Australia-level adjustment factors are provided for 3 age groups, the state and remoteness factors are not provided by age. Given that both the ABS and AIHW studies have shown that under-identification of deaths does vary by age (and to a lesser extent, by sex), the lack of age- and sex-specific adjustment factors is a limitation of this approach.

The adjustment factors used for the 2003 and 2011 reference years in the ABDS 2018 were the same as those used in the ABDS 2011, that is, factors from the ABS 2011 CDE Study for the national, state/territory and socioeconomic levels, and factors from the AIHW EMD study for remoteness levels.

### Sub-national estimates

#### State and territory

YLL estimates by state and territory were derived directly from the NMD. Deaths were classified to state and territory according to the state of usual residence of the deceased. YLL were calculated accordingly.

The state and territory analyses used the national redistribution algorithms.

#### Remoteness

Analysis for remoteness was based on the remoteness area in each death record in the NMD. Remoteness area refers to the level of remoteness of each deceased person’s usual residence, and is derived using the Australian Statistical Geography Standard (ASGS): Volume 5—Remoteness Areas July 2016 (ABS 2018a). In this study, remoteness areas were aligned to the ABS 2016 geography standard, for deaths that occurred in 2015 or 2018. Deaths that occurred in 2011 were aligned to the ABS 2011 geography standard.

Deaths where there was insufficient information to ascribe a remoteness area were excluded from the sub-national analysis. These amounted to less than 0.8% of deaths in any one reference year.

#### Socioeconomic group

As discussed in Overarching methods and choices for ABDS 2018, the ABDS did not have information on socioeconomic status at the individual level. Instead, the ABDS 2018 derived population-based socioeconomic quintiles from the 2016 Index of Relative Socio-Economic Disadvantage of the SEIFA index (ABS 2018b), which is based on the socioeconomic characteristics of the deceased person’s area of usual residence. These were applied for deaths that occurred in 2015 and 2018. For deaths in 2011, the 2011 Index of Relative Socio-Economic Disadvantage was used.

Death records with an unknown or non-specific geographical location were excluded from the analysis. These amounted to less than 0.8% of deaths in any one reference year.

### Indigenous subnational estimates

Indigenous YLL estimates by selected state and territory, remoteness and socioeconomic group were derived directly from the NMD according to the deceased’s place of usual residence.

For Indigenous YLL estimates by state and territory (reported for New South Wales, Queensland, Western Australia and the Northern Territory), and for estimates by remoteness in 2018, deaths were adjusted for Indigenous under-identification using state/territory specific adjustment factors from the ABS Census Data Enhancement study.

For Indigenous YLL estimates by remoteness in 2011 and 2003, remoteness specific adjustment factors from the AIHW’s Enhanced Mortality Database project were used to adjust Indigenous deaths.

For Indigenous YLL estimates by level of socioeconomic disadvantage, deaths were adjusted using the national age-specific adjustment factors from the Census Data Enhancement study. The Indigenous Relative Socioeconomic Outcomes Index was used to classify Indigenous deaths into socioeconomic groups (Biddle & Markham 2017).