Classifying Place of Death in Australian Mortality Statistics

Place of death, the location or setting in which an individual dies, is an important indicator for end-of-life care. With the number of Australians aged 85 years and over expected to double to more than 1 million people by 2042 (ABS, 2018), the importance of information on place of death from an economic and health service provision perspective will continue to grow. End-of-life care is a focus for all levels of government, with the 2018 National Palliative Care Strategy focused on ensuring the highest possible level of palliative care is available to all people, and the Palliative Care Outcomes Collaboration focused on improving palliative care patient and carer outcomes.

Despite the value of place of death data, there are currently no nationally consistent systems in place to accurately monitor statistics relating to place of death in Australia (Productivity Commission, 2017). The Australian Productivity Commission highlighted this as a data gap and recommended the establishment of a “national minimum data set for end of life care (including collecting and publishing linked information on place of death, primary and secondary diagnoses and details of service provision at time of death).” The Australian Hospital Statistics report produced by the Australian Institute of Health and Welfare (AIHW) currently provides numbers of deaths in hospitals and the National Coronial Information System maintains place of death for coroner referred deaths. However, there remains a large proportion of deaths which are not represented in those collections.

Information on place of death is collected through the Civil Registration system on either a Death Registration Form or the Medical Certificate of Cause of Death. While this information is provided to the ABS by the Registries of Births, Deaths and Marriages (RBDMs), the data is mostly collected and reported in free text form, with significant variations in format and quality preventing its use in statistical outputs thus far.

The value of a place of death indicator on the national mortality dataset is clear. It could provide insights into end-of-life care transitions associated with particular diseases or conditions, and it could highlight end-of-life issues particular to certain population sub-groups or those living in particular regions. Given the increasing importance of this information, the ABS commenced a pilot project to examine ways of corralling source data into meaningful and usable information.

This paper provides information on how this project was conceptualised and how data was coded and analysed. Initial results are presented with an aim to assess the quality of the outputs and the feasibility of including place of death as an integrated part of the national mortality dataset into the future.

The objectives of this pilot study were to;

• Develop a place of death framework suitable for users in the Australian setting.
• Develop a method to assign free-text place of death data sourced from Death Registration Forms or Medical Certificates of Cause of Death to this framework.

A number of key factors were considered when deciding on an appropriate classification for place of death in the Australian context. These included:

1. Relevance of the framework for the user community.

• A process of stakeholder engagement was undertaken to understand user requirements in order to maximise the usefulness of the framework for public health planning and policy requirements particularly related to end-of-life care. A clear need to distinguish between hospital and residential aged care facilities was identified.
• Some interest was also highlighted for further specified categories including; Commonwealth Approved Aged Care Facilities, inpatient hospice facilities, public hospitals, private hospitals, residential mental health facilities, specialist disability accommodation and a number of other communal establishments and institutions.
• Interest was also shown for data relating to settings such as prisons and location of car accidents to assist with other areas of public health unrelated to end-of-life care.

2. Ability for information on place of death from death registrations to be assigned to key categories.

• Place of death data is supplied to the ABS in a free text format by each of the RBDMs in Australia. The questions asked and forms used to collect this information differ between each of the jurisdictional RBDMs. Some RBDMs supply data to the ABS exactly as it is captured on either the Death Registration Form or the Medical Certificate of Cause of Death, while others consolidate information from these sources and clarify any discrepancies before forwarding to the ABS. In some jurisdictions a single field for “place of death” or “exact place of death” is used, while in others separate fields capture components of the address. This leads to variations in how doctors and funeral directors complete the forms and contributes to variations in format and levels of specificity of place of death information provided to the ABS.

• The table below provides examples of the different formats and levels of detail in which the same place of death may be provided. Each of these different formats are accepted as part of the death registration process, but the differences pose a significant challenge when assigning them to a standardised framework.

• The varied formats in which place of death could be reported also influenced the level of specificity of the classification framework. For example, if the distinction between public and private hospitals was captured consistently then this distinction could be included in the classification, but given that many records contained only a broad place of death description (i.e. “Mater Hospital” or “The Mater”) a broader hospital category was found to be more suitable.

3. Ability for the framework to be included in future enhancements to death certification. 

• In Victoria and Tasmania, tick boxes are used to capture pre-classified place of death information on either the Medical Certificates of Cause of Death or Death Registration Form. Incorporation of similar pre-classified fields across other jurisdictions could enable high quality information on place of death to be captured in the future. As such, the feasibility of incorporating the classification into future forms needed to be considered, as well as alignment with existing forms, when deciding on included categories.

Following an iterative process of investigation, engagement and analysis of the current and potential place of death data available from administrative sources, a simple five category framework for coding place of death was selected. 

A comprehensive overview of the inclusions and exclusions for each of these categories can be found the Appendix. Further detail on the benefits and limitations of the framework are outlined in the discussion section of this publication.

The pilot study sought to assign place of death to the five-category framework for all deaths in scope of the 2019 reference year. Considerable thought was given to how to overcome differences in the format and specificity of input data. Given no single method would work for all formats of input data, a hierarchical coding model was instead developed to progressively assign records to particular place of death categories using methods targeted at particular responses. This process is described in the table below.

The hierarchical coding process effectively prioritises the criteria used to assign a place of death code to each record. For instance, in step 1 of the process, all records were assessed against metadata relating to services subsidised under the Aged Care Act 1997. Records that matched to this metadata list were assigned the relevant code and then taken out of scope for subsequent steps in the coding process.

Where input data was limited and could not be directly matched to a particular facility, decisions needed to be made about what category responses would default to. For instance, some companies provide Commonwealth subsidised aged care or nursing home facilities, but also provide independent retirement living options. Where only a company name was provided as place of death without further indication of the type of facility, these were assigned to the residential aged care facility category.

The RBDMs in Victoria and Tasmania currently collect pre-classified place of death data as part of the death registration process. In Victoria, this information is collected using the Medical Certificate of Cause of Death, while in Tasmania pre-classified data is collected on the Death Registration Form.

For the purposes of this pilot, the Victorian RBDM provided the ABS with pre-classified place of death information (July to December 2019 only) to help assess the quality of data obtained from the hierarchical coding method. Neither jurisdiction had previously provided pre-classified place of death data as part of routine reporting to the ABS.

The Victorian pre-classified data has four tick box categories:

• Home/residence
• Nursing home
• Hospital
• Other  

The table below shows how these tick box categories were concorded to the ABS place of death framework. Results of this analysis are provided in the section below.

• The proportion of records assigned through each of the different steps in the hierarchical coding methodology varied across each of the states and territories. For instance, across Australia 8.7% of records were coded based on a match to the list of services subsidised under the Aged Care Act 1997, while at a jurisdictional level the proportion varied from 20.2% in Western Australia to 0.9% in the Northern Territory.

The place of death framework and methodology was applied to all deaths in scope of the 2019 reference year. The results are outlined below: 

• The majority (51.0%) of deaths in Australia occurred in a hospital/medical service area.
• Residential aged care facilities were the second most common place of death accounting for 29.5% of all deaths.
• Other specified locations were the least common place of death (1.4%).
• Of the 169,301 records in scope of the 2019 reference period, 5,792 records (3.4%) did not contain sufficient free text place of death information to be assigned to a specified place of death category.

• The study highlighted state variations in patterns of place of death.
• The highest proportion of deaths in the home/residence category was seen in the Northern Territory (29.3%). The Northern Territory also recorded the highest proportion of deaths in hospital/medical service areas (54.9%) but the lowest proportion of deaths in residential aged care facilities (9.2%).
• South Australia recorded the greatest proportion (35.0%) of deaths in residential aged care alongside the lowest proportion (46.3%) of deaths in hospital/medical service areas.
• The proportion of unspecified records was highest in Queensland. 16.3% of Queensland records (5,337) did not contain sufficient place of death information for categorisation. See the ‘Discussion’ section of this report for further information.
• The proportion of records with insufficient place of death information was also high in the Australian Capital Territory with 224 unspecified records (9.9%). 212 of these 224 unspecified records were certified by a coroner.
• The proportion of deaths assigned to home/residence were relatively low in Queensland (2.0%) and the Australian Capital Territory (9.4%). These low results compared to other states and territories may indicate specific issues with data inputs or the suitability of the coding methods for those jurisdictions.

• Place of death was analysed for selected cause of death groupings, with large differences in the proportions assigned to particular categories for some causes. 
• Amongst deaths due to neoplasms, the most common place of death was hospital/medical service area, accounting for 67.1% of neoplasm deaths. This was the highest proportion of hospital/medical service area deaths recorded across the selected cause groupings analysed.
• Deaths from dementia, including Alzheimer’s disease had the highest proportion that occurred in residential aged care facilities (76.8%) and the lowest proportion that occurred in hospital/medical service areas (19.8%) and home/residence (2.5%).

• The likelihood of a death occurring in a residential aged care facility increased with age.
• 50.1% of deaths of people aged 85 and over were in a residential aged care facility.
• The likelihood of death occurring in a home/residence was higher among younger age cohorts.
• Among people aged 85 and over, residential aged care facilities were the most common place of death for each of the selected cause groupings except for external causes of morbidity and mortality (V01-Y98), neoplasms (C00-D48) and other (all other codes). Hospitals were the most common place of death for those cause groupings.

• A higher proportion of records with an unspecified place of death were recorded in coroner certified records (8.3% unspecified) compared to doctor certified records (2.7% unspecified). The majority of these records were registered in Queensland and the Australian Capital Territory.

Further results and tables containing breakdowns of place of death by variables including causes of death, age, sex, indigenous status, certifier type, index of relative socioeconomic disadvantage and remoteness area can be found in the data downloads of this publication.

• The Victorian Registry of Births, Deaths and Marriages provided additional information on 21,988 deaths that were registered between July and December 2019, including pre-classified tick box data from the Medical Certificate of Cause of Death.
• Place of death according to the pre-classified tick box data was analysed alongside that derived from the free text responses. This analysis highlighted relatively strong alignment between places of death assigned using the two methods.
• Particularly strong alignment between data from the two sources was recorded for the residential aged care facilities and hospital/medical service area categories.
• Numbers of deaths assigned to the ‘other’ category were much higher among pre-classified tick box records (1,463 deaths) compared to the free-text records (229 deaths).
• Among those records coded to ‘other’ using pre-classified tick box data, 814 were coded to the home/residence category when coded using free-text data.
• Of those 814 records, 485 were assigned to the home/residence category in step 8 of the hierarchical coding process. That is, they could not be categorised based on keywords or matches to metadata so were allocated based on the assumption that records remaining at this stage which contain a numerical value were most likely a home/residence.

• A further 300 of those 814 tick box responses in the other category were coded to the home/residence category based on the free text place of death data matching the usual residence.

• Minor discrepancies between how records were assigned to residential aged care facilities and hospital/medical service areas using the free text and pre-classified data were investigated. These investigations highlighted how multi-care facilities (i.e. hospitals with on-site aged care facilities) could lead to different results when using different methods.

The COVID-19 pandemic has presented an unprecedented challenge for health systems around the world, with aged care facilities facing very particular challenges because of the vulnerability of elderly residents to this disease and the need to prevent further transmission when infections occurred. During the pandemic, a number of strategies were used to manage COVID-19 outbreaks in aged care facilities including transferring residents to hospitals, cohorting residents on-site, and providing hospital in the home services. These strategies were aimed at balancing the need to reduce the spread of COVID-19 alongside best meeting the needs and wishes of individuals diagnosed with the disease (Royal Commission into Aged Care Quality and Safety, 2020).

Disease surveillance data available from the Department of Health currently provides information on the number of deaths due to COVID-19, for people living in Australian Government–subsidised residential aged care facilities. However, this information reflects only whether an individual was receiving Australian Government–subsidised residential care at the time of their death. This may not necessarily reflect the location of the death, as many aged care residents who contracted COVID-19 were transferred and died in hospital (Royal Commission into Aged Care Quality and Safety, 2020).

Given the high importance of understanding where deaths have occurred during the COVID-19 pandemic, an extension of the place of death pilot project was undertaken using the ABS’s Provisional Mortality dataset. Place of death was coded for all COVID-19 deaths that occurred before 24 November 2020 and were received by the ABS by 31 December 2020. These deaths were coded in accordance with the place of death framework and coding processes outlined in this research paper. For further information relating to the scope and coverage of the Provisional Mortality Dataset please see the methodology section of that report.

A summary of the results is provided below. For further data tables please see the data downloads section of this publication.

• There were 864 deaths due to COVID-19 in scope for this case study.
• All deaths due to COVID-19 were reported in either a home/residence, residential aged care facility or hospital/medical service area.
• Place of death was most commonly a hospital/medical service area (63.5%).
• The proportion of deaths in residential aged care facilities increased with age. 

This pilot study had two key aims. The first was to develop a framework suitable for categorising place of death information available through the death registration process in Australia. The secondary aim was to develop and test a method for assigning free text place of death information from existing data sources to this framework.

The proposed framework and method for coding place of death was able to be applied to free text information collected through the existing death registration system in Australia. The data produced provides insights into where people die in Australia, though subject to some limitations. It was expected that approximately 30% of deaths would occur in residential aged care facilities, 50% in hospital/medical service areas and 20-25% in all other areas. Results of the study aligned closely with expectations from other sample studies, hospital separation data and engagement with experts working in the field (AIHW, 2020; Broad et al., 2013; McNamara & Rosenwax, 2007).

The majority (86,276, 51.0%) of deaths were coded to the hospital/medical service area category of the framework. The Australian Institute of Health and Welfare reported 82,712 deaths occurring in hospitals (combined admitted patient and emergency department deaths) during the 2018-19 reference period (AIHW, 2020). This variance between the deaths recorded within the ABS framework and the AIHW hospital separation data may be explained by scope differences. The ABS data is collated by calendar year while the AIHW data is collated by financial year. The ABS framework also includes a number of additional medical services areas (e.g. medical centers, outpatient clinics and day procedure centres) which would account for some additional deaths being assigned to the hospital category.

The proportion of deaths categorised as having occurred in residential aged care facilities increased with age. Across age groups, a higher proportion of female deaths compared to male deaths occurred in residential aged care facilities. This fits with the reported profile of aged care consumers in Australia, where the average age of entry to permanent residential aged care is 82 for men and 85 for women, with more women using aged care services (Grove A, Parliament of Australia website, 2019). These findings are also consistent with a prior study by Broad et al. (2013) which looked at place of death in Australia using customised reports of discharges from hospitals and residential aged care facilities. These results show that the framework and coding model are able to accurately assign place of death for residential age care facilities in Australia from the death registration system.

While hospital and medical service area was the most commonly assigned place of death in each state and territory there were variations in the proportion of deaths assigned to other categories. These variations reflect a combination of administrative effects, demographic differences and end-of-life care policy and service delivery in jurisdictions. For example, in the Northern Territory approximately 30% of deaths were categorised as having occurred at home or in a private residence and only 10% coded to a residential aged facility. This is likely to reflect the markedly different population profile of the Northern Territory, including a younger population, a higher proportion of males than females and a larger proportion of people living in remote and very remote areas compared to other states. This leads to differences in causes of death and the way services are accessed and delivered.

Both Queensland and the Australian Capital Territory recorded a higher proportion of deaths with no specified location compared with other states and territories. While this is due to a range of factors a key reason is that a higher number of coroner referred deaths have no place of death filled out in the death registration system for these jurisdictions. As deaths that occur at home account for the highest proportion of coroner referred deaths, this increased proportion of not stated records also leads to a decrease in the proportion of deaths occurring in a home or private residence. The ABS is working closely with the RBDMs to understand how certification by a doctor or referral of a death to the coroner may affect the way in which information on place of death is recorded. This may lead to improvements in data provided via the RBDMs, but it is important to note that place of death information for coroner certified deaths is available from other sources in Australia. Currently the National Coronial Information System codes place of death for coroner certified records in Australia. The classification is more detailed and designed to capture the range of unique factors relating to place of death that may be present for coroner certified deaths, for example roadways.   

A key strength of this pilot study was the use of existing administrative data already collected through the death registration process in Australia. Previous studies on place of death in Australia have relied on sample populations or have focussed on one jurisdiction only (McNamara & Rosenwax, 2007). The use of death registration data enables the collection and dissemination of a complete national dataset with a consistent approach to assignment of place of death category. Furthermore, attaching place of death to the national mortality dataset will mean that it can be analysed with cause of death and other demographic variables produced in official mortality statistics.

The classification that was applied to assign place of death in Australia is simplistic. This allows for deaths to be categorised more easily and avoids some of the complexities associated with a more detailed framework. As noted in the methods section, many facilities can have multiple purposes, so additional categories might increase the level of detail but reduce the accuracy of resulting data.

The classification produced robust outputs for deaths occurring in hospitals and residential aged care facilities. With an ageing population and the majority of deaths occurring in hospitals and aged care in Australia, this classification can provide insights into end-of-life care and fill an existing data gap in mortality statistics. This framework also aligns with information and outputs used internationally allowing meaningful comparisons to be made.

There was close alignment of results produced by the automated model and the pre-classified tick box from the Victorian Medical Certificate of Cause of Death. As electronic certification is developed and progressively adopted by RBDMs in Australia, the applied framework presents an opportunity to implement a standardised process to collect pre-classified place of death information. The pre-classified tick boxes ensure that the certifying doctor provides their view on the type of facility in which the person died. This information is reported directly and requires no further coding which might bias or reduce the quality of resulting data.

The close alignment between the outputs of the two methods supports a staged approach to implementation, i.e. information based on pre-classified data can be used alongside that coded using the hierarchical coding methodology and as collection methods change these can be integrated into the national mortality dataset. The ABS will continue to engage with RBDMs to support the implementation of a standardised place of death tick box that is consistent across jurisdictions as part of the death registration process.

Limitations of the pilot study must be considered before using the data for research purposes. Free-text information collected as part of the death registration process was the key data source for this pilot study. The variations in question wording on official registration forms did lead to some differences in how the place of death field is filled out. For example, a residential aged care facility could be received as an address, a facility name or just the word “nursing home”. The final classification could be affected by the input text.   

There were some jurisdictional differences, with the most notable being the high proportion of unspecified records in Queensland and the Australian Capital Territory. These appear to affect comparisons in the home, other and unspecified categories. Considering certifier type (i.e. doctor or coroner) and cause of death are important when interpreting this data.

A simplistic framework was applied to the data in order to best capture and categorise place of death information collected via the death registration system and one that addressed data gaps in relation to end-of-life care. The framework was not designed to capture greater specificity of other locations which may be more pertinent to other areas of public health, including deaths in prisons, psychiatric wards and locations of car crashes. Many of these deaths are referred to a coroner for investigation. The National Coronial Information System currently code a place of death data for these deaths, providing an alternative data source which may address some of these limitations. 

The hierarchical framework for assigning place of death was designed to be applied to large administrative datasets in an automated environment. This could lead to some errors in coding. Any records in scope at step 3 of the hierarchical coding process as outlined in the methods section (i.e. without a code already assigned through steps 1 or 2) that contained the word “hospital” in the place of death text were assigned to the hospital/medical service area category. This step was implemented after extensive testing which indicated records containing the word “hospital” were in the majority of cases a hospital or medical services area. However, there may be rare instances where this assumption is not correct. For example a death may occur in a home/residence (or any other place of death category) where the location is “Hospital Road”. In this instance the record would have been inaccurately assigned to the hospital category based on the keyword “hospital”. While it is expected these occurrences are rare it is an important consideration of the coding process. Integration of tick boxes on the death registration form would eliminate this issue in future.

The hierarchical coding framework relied on a substantial metadata base, including current aged care registers and lists of hospitals. The maintenance of this metadata would be resource intensive and presents issues of sustainability of the model into the future. While pre-classified tick box data presents an opportunity for moving to a more sustainable model, this move will take time across all jurisdictions. The ABS will continue to work with data providers and users into the future on this issue.

The framework used in this study provides a descriptive variable of the location of death for an individual. It does not capture movement between facilities during the end-of-life process. It also does not capture details regarding an individuals’ preferred place of death or other qualitative factors such as access to services, levels of pain, comfort, family support, psychological care and spiritual aspects that are also important in end-of-life experiences. If the place of death classification is included on the national mortality dataset, data linkage may provide additional insights into some of these other important factors contributing to the end-of-life experience that cannot be fully understood through place of death data alone.  

The pilot study indicates that a robust representation of place of death in Australia can be achieved using the proposed classification and coding model. Data for hospitals and residential aged care facilities was closely aligned with other data sources and expectations. Pending feedback from data users on the utility of this information for policy and research there is an opportunity to integrate this information into the national mortality dataset. The ABS will continue to work with stakeholders on this topic to look for sustainable solutions for implementation. 

Pre-classified place of death data completed as part of the death registration process has been identified as a potential solution which could address some of the current limitations of this study. Tick box data, as opposed to the free-text data collection method provides a more accurate assessment of the type of location in which a death occurred. Certifying practitioners make a decision on place of death based on their immediate understanding of contextual circumstances, which removes biases that could unintentionally be introduced as part of the free-text hierarchal coding process. Pre-classified tick boxes would also remove the requirement to maintain detailed metadata used to code place of death information from free text data. 

Comparability between jurisdictions could also be enhanced by use of pre-classified place of death tick boxes. Victoria and Tasmania are the two jurisdictions which currently collect a detailed pre-classified place of death variable during the death registration process. Standardisation of collection methods across jurisdictions using the proposed classification would result in nationally consistent collection and reporting of place of death data. The ABS is working with jurisdictions to look for opportunities to adopt collection of standardised pre-classified place of death data, especially as RBDMs progressively move to electronic platforms. For instance, the Queensland RBDM is currently in the process of implementing a new electronic Medical Certificate of Cause of Death, and this will include a pre-classified place of death question that can be completed by the certifier. 

The data presented in this publication reflect the results of a pilot study and should be interpreted with consideration to how the framework has been applied. Please see the methods section of this research paper for further information.

The principle authors of this report were Megan Christison, Sue Webster, James Eynstone-Hinkins and Lauren Moran. They would like to gratefully acknowledge the contributions and support received from Professor Kathy Eagar and Dr Tony Ireland of the Palliative Care and End-of-Life Care Data Development Working Group. 

The authors wish to thank the Registries of Births, Deaths and Marriages, the Coroners and the National Coronial Information System for enabling COD URF data to be used for this publication.

This publication was previously released under catalogue number 3303.0.55.005