Advance care directive prevalence among older Australians and associations with person‐level predictors and quality indicators

Abstract Background Advance care planning (ACP) conversations may result in preferences for medical care being documented. Objective To explore the uptake and quality of advance care directives (ACDs) among older Australians accessing health and aged care services, by overall ACP documentation prevalence, person‐level predictors and ACD quality indicators. Design and Setting National multi‐centre health record audit in general practices (GP), hospitals and residential aged care facilities (RACF). Participants A total of 4187 people aged ≥65 years attending their GP (n = 676), admitted to hospital (n = 1122) or residing in a RACF (n = 2389). Main Outcome Measures ACP documentation prevalence by setting and type including person‐completed ACDs and non‐ACD documents (completed by a health professional or someone else); person‐level predictors and quality indicators of ACDs. Results Overall ACP documentation prevalence was 46.5% (29.2% weighted). ACD prevalence was 25.3% (14.2% weighted). Unweighted ACD prevalence was higher in RACFs (37.7%) than in hospitals (11.1%) and GPs (5.5%). 35.8% of ACP documentation was completed by a health professional (9.7% weighted), and 18.1% was completed by someone else (10.6% weighted). Having an ACD was positively associated with being female, older, having two or more medical conditions, receiving palliative care, being divorced/separated and being in a RACF. Only 73% of ACDs included full name, signature, document date and witnessing. Conclusions and Contribution Low ACP documentation prevalence and a lack of accessible, person‐completed and quality ACDs represent an important ACP implementation issue. Low prevalence is complicated by poor document quality and a higher prevalence of documents being completed by someone other than the person.


| INTRODUC TI ON
In 2017, there were 56 million deaths globally, with 58% of these deaths recorded for people aged 65 years or older. 1 While some individuals will die suddenly and unexpectedly, many will experience a prolonged period of deteriorating health, chronic illness, comorbidity and diminishing decision-making capacity prior to death. 2 This period is associated with adverse health outcomes, including poorer quality of life, 3 increased hospitalizations, 4 unwanted and intensive medical treatments and greater health-care expenditure. 5 While almost half of those nearing the end of life will require treatment decisions to be made, the majority will lack the capacity to make their own decisions. 6,7 Advance care planning (ACP) and related medical treatment legislation enable individuals to maintain choice and control over their health-care decisions in the event they lose decision-making capacity in the future. ACP is an on-going process of reflection and discussion that supports a person to identify and discuss their goals, values and preferences for future care with health-care providers and loved ones. 8 The ultimate goal of ACP is that people receive medical care that is consistent with their preferences. 9 ACP is associated with a range of beneficial outcomes for individuals, their families, health professionals and the health system. 6,10 -14 However, a recent Australian study of people with cancer and support people found that ACP and assistance with the completion of ACDs were commonly not discussed as part of standard care, despite two-thirds of consumers having discussed ACP with loved ones and approximately half wanting ACP discussions with their doctor. 15 Ideally, outcomes of ACP discussions will be documented in an advance care directive (ACD). 9 Definitions, terminology and legislative frameworks surrounding ACDs vary internationally.
Broadly, ACDs are voluntary person-completed documentation of medical treatment preferences and may include future consent to or refusal of health care and/or the appointment of a substitute decision maker (SDM). 10,12 In Australia, an ACD has been defined as a written document recognized by specific legislation or common law that is completed and signed by a competent adult. 16,17 Three main types of ACDs are currently used in Australia: (1) state-specific statutory documents recording preferences for care, (2) state-specific statutory documents appointing an SDM and (3) non-statutory documents recognized under common law (these can record preferences for care but not appoint an SDM). [17][18][19] Sometimes ACP discussions lead to documentation being created on behalf of a person by health professionals, family members or SDMs to guide future medical treatment decision making in the absence of an ACD, but are not provided for within legislation in Australia. For this study, we use the term 'ACP documentation' as a catch-all term for all ACD types created by the person (including statutory and non-statutory ACDs) and any other non-ACD documents created on behalf of the person by a health professional or someone else. Non-ACD documents refer to any formal medically driven documents, usually completed by doctors, that outline treatment plans in the event of deterioration (eg medical orders or Goals of Care) or ACP discussion records produced by a health professional or someone other than the person. ACP documentation must be available where the person is receiving care to inform medical decision making. 20,21 Mechanisms exist in some health services to support people to complete ACDs or elicit existing documentation upon admission to hospital or residential aged care facilities (RACFs, ie long-term care facility for older adults who can no longer live at home). Health services in Australia use a range of systems to record, store and retrieve ACP documentation, including electronic medical records, hard copy files and the centralized national e-health system, 'My Health Record'. 22,23 When these documents are not available, the person may receive care that is inconsistent with their expressed preferences which may be a health-care safety and quality issue.
ACDs and the requirement for health professionals to act per a person's documented preferences are supported by legislation and national policy frameworks in Australia. [16][17][18][24][25][26] However, the availability and perceived validity of ACDs are reported as barriers to ACD adherence. 27,28 Nationally and internationally, ACD uptake remains low, with reported prevalence rates ranging from less than 1% to approximately 30%. [29][30][31][32][33][34][35] Where ACDs are found, barriers to health practitioner adherence to ACDs include quality concerns such as document currency (eg age of document) and confusion over the legal standing of ACP documentation. 27,28 The prevalence of ACDs versus non-ACD documents completed by a health professional or someone else may indicate how well Australians are being supported to complete and store persondriven documentation. To the best of the research group's knowledge, no Australian study has described ACP uptake across health sectors and jurisdictions by examining the uptake and quality of ACDs as a proportion of ACP documentation, person-level predictors associated with producing an ACD or the quality of ACDs present. A 2017 Australian multi-centre pilot prevalence feasibility study 36,37 found 30% (unweighted) of older people had at least one ACD in their health record, and 22% (unweighted) of people had ACP documentation other than ACDs in their health records, including documents written by health professionals or family members. It is critical to continue to build upon our current knowledge of ACP documentation, ACD prevalence [36,37] and predictors and quality of ACDs to identify ACP improvement priorities for the Australian health system. This study aimed to: 1. Describe the prevalence of all types of ACP documentation among older people accessing GPs, hospitals and RACFs, including the proportion and type categorized as ACD and non-ACD documents 2. Determine person-level predictors associated with ACD completion, and 3. Identify the quality of ACDs as determined by their alignment with quality indicators (eg personal information, signatories, witnessing requirements and date of completion).

| ME THODS
This article reports major findings from a national multi-centre crosssector audit study examining the prevalence of ACP documentation, and person-level characteristics and quality of ACDs in selected Australian general practices (GPs), hospitals and RACFs. The protocol has been published elsewhere. 38 Learnings from the pilot feasibility study 37

| Setting, sample, recruitment and data collection
Participating sites were GPs, hospitals and RACFs in Australia, recruited via an expression of interest process advertised through key stakeholder networks. Following initial recruitment, additional organizations were approached by the project team to promote representativeness across sectors and jurisdictions. 38 Organizations were eligible to participate if they had (1) access to a minimum of 30 health records likely to meet inclusion criteria for the audit and (2) sufficient staff and resources to conduct the study. Independent data collectors were provided as needed to undertake the audit at sites that were otherwise unable to participate due to limited resources.
Up to three data collectors for each site completed mandatory online training and had access to state-specific data collection manuals (reflecting jurisdictional differences in ACP legislation, documentation and terminology). Data collectors were provided with clear definitions and examples for document classification and flowcharts to assist in identifying and classifying documents. 38 Participating sites were required to nominate how many records they would audit in advance (minimum of 30, maximum of 50). No funding was provided to participating sites.
For health records to be eligible for the audit, patients/residents needed to be aged 65 years or older and admitted to a participating hospital or RACF for more than 48 hours before the audit, or visiting a participating GP on the study day(s). Records were randomly selected from a list of eligible people in hospitals and RACFs, while consecutive eligible records were audited in GPs. 38 Data collectors searched each record for a maximum of 15 minutes. This timeframe was applied to reflect the need for ACP documentation to be quickly and easily accessible in clinical settings.
Information about any ACP documentation identified in the record was extracted (eg type of document, time taken to find, characteristics), together with demographic and clinical information about the person. All data were entered and stored on a secure online purpose-built database, which was adapted from the pilot feasibility study based on key learnings. 37,38

| Health record audit
Demographic and clinical data extracted included age; gender; country of birth; English language status; relationship status; current/active medical condition(s); palliative care status; and functional status.  were misclassified by data collectors as statutory ACD preferences for care and were recoded before analysis as statutory ACD-SDM (n = 2), non-statutory ACDs (n = 11) and medical orders (n = 2).

| Outcome measures
The primary outcome was ACP documentation prevalence by setting and type, including the prevalence of ACDs by document type, non-ACD documents completed by a health professional and non-ACD documents completed by someone else (eg family, SDM). Secondary outcomes were person-level predictors of ACDs and document quality of ACDs.
Prevalence rates of different types of ACP documentation were calculated by health sector and overall.
Given disproportionately high numbers of people in the 'old-old' age ranges (ie those aged 75 years and above), and people from particular jurisdictions, a weighting score was derived for each record, in order to calculate a weighted estimate of ACP documentation prevalence among the underlying population of Australians aged 65 years and over. Weighting scores were derived using Australian Bureau of Statistics population data for age, gender and state. 41 Due to lower response rates in three states/territories (Australian Capital Territory, Tasmania and Western Australia), participants from these states/territories were combined in the weighting process. Population values are not available for gender other than male and female. Where gender was coded as 'other', the average weights for males and females of the same age and jurisdiction were applied. The formula for generating weights and the table of derived weights is provided in Supplementary File, Table 1. The relatively over-represented sub-groups (by age, gender and state) were used as baseline values in weighted analyses.
Multivariate logistic regression was used to assess the relationship between person-level predictors and the presence of an ACD completed by the person. A pseudo intra-class correlation coefficient was calculated to estimate the influence of clustering of observations by site, and a random intercept regression model with independent covariance structure was fitted, to account for this clustering. For each predictor, adjusted and unadjusted results are provided. These model results also include adjustment based on the weighting scores discussed above. In the adjusted model, 876 participants (20.9%) were excluded from the regression analysis due to missing data. For some variables, there were little missing data, but notable were country of birth (n = 347), palliative care status (n = 299), relationship status (n = 322) and functional disability (n = 200). A sensitivity analysis was performed, including missing data as a separate level of each variable to assess the influence of excluding missing data on the results. The analysis indicated that the estimated odds ratios were highly consistent between both models. Therefore, missing data were excluded from the final model. Odds ratios for having an ACD are reported, together with 95% confidence intervals (CI). The statistical significance level was set at 0.05.
Multivariable multinomial mixed-effects logistic regression was used to assess the relationship between demographic characteristics of the sample with the ACP documentation completion outcome. Three groups composed the outcome: ACDs completed by the person, non-ACD documents completed by another and no ACP documentation. A random intercept with independent covariance structure was used to account for clustering of observations by site.
For each predictor, crude or unadjusted and adjusted results have been provided. These model results also include adjustment based on the weighting discussed above.
Data reported include unweighted prevalence rates by document type and sector and overall weighted prevalence rates for document types. Sub-category prevalence rate percentages are presented as a percentage of health records containing some form of ACP documentation, rather than as a percentage of all audited health records.

| Document quality analysis
Quality data extracted from ACDs were evaluated against criteria for quality sourced from ACP legislative formalities, the National Framework for ACDs and an Australian study. 17,18,42 Yes/no responses were recorded for each ACD to identify the presence (yes) or absence (no) of quality indicators. Yes responses were used to calculate the total percentage of documents containing information aligning with each quality indicator. Data were then organized by ACD type to describe the overall quality of ACDs by document type.
Quality indicators assessed included the full name, date of birth, address and signature of the person, witness signature and their relationship to the person or professional role, and the document date.
Document age was calculated from document date. RACFs representing all eight Australian jurisdictions. Of 4188 audited records, one person fell outside of specified age criteria and was excluded, leaving 4187 participants. The median age of participants was 82 years, and 60.3% were female (Table 1). Most were born in Australia (64.2%) and spoke English (89.6%). Participants had a median of three medical conditions: the most common being musculoskeletal/ connective tissue (53.7%), heart condition (52.4%) and dementia (33.2%). More than half had severe to very severe disability.

| Prevalence of all types of advance care planning documentation
Less than half the sample (46.5%, n = 1946) had at least one type of ACP documentation present in their health record. After weighting, the prevalence of ACP documentation was 29.2%. Of the entire sample, 808 records (19.3% unweighted, 10.7% weighted) contained only an ACD, 245 (5.9% unweighted, 6.5% weighted) contained only non-ACD documents completed by a health professional only, and 556 (13.3% unweighted, 7.2% weighted) contained only non-ACD documents completed by someone else. Just over a fifth of all records (n = 885, 21.1% unweighted, 15.0% weighted) contained only non-ACD documents completed by a health professional and/or someone else (ie did not also have a person-completed document).

| Prevalence of non-advance care directive documents completed by a health professional
The overall (unweighted) prevalence of non-ACD documents completed by a health professional was 35.9% (weighted prevalence 9.7%, n = 1504), see Table 2. Roughly one fifth (21.3% unweighted, 6.5% weighted, n = 891) of the audited records only contained non-ACD documents completed by a health professional (ie did not also have a person-completed ACD or ACP documentation by and GPs (0.1%, n = 1), while other types of ACP discussion records were more common in RACFs (14.8%, n = 354) than hospitals (8.5%, n = 95) and GPs (6.2%, n = 42). Other types of ACP discussion records completed by a health professional included ACP clinical notes and letters.

| Prevalence of non-advance care directive documents completed by someone else
The prevalence of non-ACD documents completed by someone else was 18.1% (weighted prevalence 10.6%, n = 757), see Table 2.

| Prevalence of advance care directives as a proportion of all ACP documentation
ACDs comprised 25.3% (n = 1061, 14.2% weighted) of all health records audited (n = 4187, see Table 2). Statutory ACD preferences for care accounted for 5.9% (n = 249, 3.6% weighted), statutory ACD-SDMs accounted for an additional 12.2% (n = 511, 6.1% weighted), and non-statutory ACDs accounted for 11.5% (n = 480, 6.9% weighted) of all ACP documentation. Non-statutory ACDs were most prevalent in RACFs (n = 429, 18.0% unweighted) as compared to GPs and hospitals.  (compared to those who were widowed), had two or more medical conditions (compared with one medical condition), receiving palliative care (compared with not receiving palliative care) and/or residing in a RACF (compared with being in a hospital or general practice).

| Quality of advance care directives
All three types of ACDs included in the quality assessment contained some records that were missing one or more quality indicators (Tables 4 and 5     The 25% unweighted ACD prevalence rate and the higher prevalence of ACDs in RACFs reflect the prevalence rates reported in the 2017 pilot feasibility study. 36 The weighted ACD prevalence rate found in the current study is similar to other Australian research reporting a general community ACD prevalence rate of 12 31 to 14%. 35 Internationally, higher ACD prevalence rates have been reported, with one systematic review and meta-analysis reporting a prevalence of any type of ACD for US adults to be 37%. 43 A separate

| D ISCUSS I ON
Canadian study also showed a higher ACD prevalence rate, with 20% of participants reporting having completed an ACD and 47% reporting having appointed an SDM. 44 Variation in prevalence rates may reflect differences in recruitment processes, prevalence rate calculation approaches (including population weighting), demographic profiles or differences in the storage of ACDs. However, further exploration of this variation is warranted.
When looking at ACD prevalence by sector, ACDs were more prevalent in RACFs, with GP health records least likely to contain any ACDs. Higher rates of ACP documentation in RACFs may reflect the need to plan for future medical treatment decisions due to cognitive decline, palliative or end-of-life care in aged care. 45 Differences in sector processes for identifying ACP documentation on admission and the increased likelihood that an individual's health information will be stored at their RACF as their primary residence may also contribute to this finding. National and international ACD prevalence rates have produced similar results to those in the current study, with hospital prevalence rates reported between 0% and 35% 11,46-50 and RACF prevalence rates in Canada reported between 11% and 44%. 45 For GPs, ACD rates have been reported from 20% to 33% 51,52 with much lower rates reported in Australian GPs. 53 Possible explanations for the lower prevalence of ACDs in GPs may include a lack of systems, processes, time and/or funding in place to support ACP discussions with patients or issues with the approach used by general practitioners to identify patients they believe would benefit from ACP discussions. 13,[54][55][56] The low prevalence rates for ACP documentation and especially ACDs across health-care sectors in this study are concerning given the importance of person-centred care and needs of a study population with a median age of 82 years, a median of three medical conditions, with more than half showing severe or very severe functional decline, half having heart disease and one-third having dementia.
Many of these people might be expected to have a limited prognosis, be at risk of deteriorating health and require significant medical treatment decisions to be made.
Non-ACD documents, including medical orders and ACP discussion records, completed by a health professional were more common in hospitals than RACFs and GPs. This result is likely to reflect jurisdictional and hospital policy promoting the regular use of medical orders during patient admissions to hospital. It may also reflect medical practitioner preference to use doctor completed rather than personcompleted documentation, or as a result of sudden deterioration or delirium after admission. 15 Research suggests health professionals prefer using clinician-led documents, as these are often easier to translate into clinical plans than general instructions contained in ACDs that may or may not relate to the illness being treated. 27 Alternatively, the higher prevalence of medical orders in hospitals than ACDs may reflect system-specific factors and/or processes that limit the amount of time health professionals can spend discussing ACP with patients or a lack of processes to assist with locating an ACD. Given the purpose of ACP is to promote person-driven planning for future medical treatment decision making, further research is needed to examine whether or not health professionals are encouraging the person receiving care to create an ACD rather than creating non-ACD documents on behalf of a person.
After weighting to account for the larger proportion of ACP documentation found in RACF health records, the prevalence of non-ACD documents completed by someone else was 10.6%, with the majority found in RACFs. Ideally, future health-care planning should result in the earlier completion of one or more person-driven ACDs.
However, some individuals may be unwilling or unable to complete an ACD because of cognitive decline, cultural factors, poor literacy or distrust. 57,58 In these scenarios, others may decide to create non-ACD documents on their behalf to inform future care decisions.
Several person-level predictors were associated with higher odds of having an ACD on record. Women and those who were older were more likely to have an ACD in their health record. These results are consistent with existing research. 30 More than two-thirds of those with an ACD on file had two or more comorbidities, and just over half had severe or very severe functional impairment. Other research has also identified an association between the presence of an ACD with health status and/or multimorbidity. 45,46,60 Consistent with previous research, people receiving palliative care were also more likely to have an ACD. 46,59,60 These associations may reflect the way that health professionals identify individuals they deem appropriate for conversations about ACP, or the greater proportion of people living in RACFs in this study.
The results of the quality assessment suggest ACD prevalence rates may overestimate the presence of documents that are likely to be enacted by a health practitioner at a person's point of care.
National policy guidance in Australia recommends an ACD should use the prescribed form when available and contain quality indicators such as name, document date, person signature and witnessing. 17,18,64,65 The presence of these quality indicators allows health practitioners to verify the identity of the patient the document applies to and provides evidence the document was created voluntarily by the person at a time they had decision-making capacity. 17,18 However, only 73% of ACDs included all four of these patient identifiers, and 58% of ACDs were found to be three or more years old. National and international studies have identified a lack of congruence between patient preferences and medical treatment decision making. [66][67][68][69] In a study of people with end-stage kidney disease, 15 of 57 participants received medical treatment that was inconsistent with their known preferences at end of life. 70 An international study of emergency physicians found 86% of respondents were willing to honour legal ACDs. When informal ACP documentation was used, only 7% were willing to adhere to the specified preferences. 66 Similarly, a study of 649 medical specialists found only 32% of specialists complied with the law when presented with an ACD scenario, with doctors more likely to make clinically indicated medical decisions than make medical decisions that follow the law. 68 When examining ACD quality indicators by document type, con-

| Strengths and limitations
The strengths of this study include the cross-sector involvement, participation across all Australian jurisdictions, comprehensive training of auditors, reliability testing and national research governance. However, this research has several limitations. Recruiting via expression of interest likely resulted in a sample containing primarily services with an interest in the study topic. Weighted analyses contributed to a more representative description of ACP in Australia; however, disproportional results across organization types and jurisdictions still limit the generalizability of findings across Australia.
Similarly, weighting was performed for some variables but could not be weighted for others, such as site. Because not all Australian jurisdictions had representation from each sector, caution should be used when interpreting results related to jurisdictions, and the presence of missing data may have also influenced results. Where more than one document was present in a health record, only the date of the most recent document was recorded. As such, documents with no date identified were removed from the quality analysis and listed as missing data. Data collected included assessments of functional status, but not cognitive status. Cognitive status may be a stronger predictor of whether an ACD exists than a person's functional status if this assessment is made at the time an ACD is created. The quality analysis did not consider whether ACD content could be used in a clinical setting to inform decision making. Lastly, it was not possible to report quality data by jurisdiction due to a lack of representation. and individuals, to ensure access to and support for ACP. ACP implementation strategies will need to address ways to improve the prevalence and quality of ACDs. Further research exploring the impact of different ACP systems, program characteristics, policy, funding, clinician priorities and/or training on ACD prevalence may be useful.

ACK N OWLED G EM ENTS
This study was funded by the Australian Government, Department of Health (Grant / Award Number: 'ID 4-5833ZYN').

CO N FLI C T O F I NTE R E S T
All authors declare that they have no conflict of interest related to this work.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.