Empirical studies show that healthcare costs increase exponentially at end of life [1–4], but increased spending may not improve quality of care [5]. In addition, aggressive medical treatment at the end of life may be inconsistent with the care preferences of the patient [6] or prove to be unnecessary [7]. In Australia, approximately 80% of adults will be admitted to hospital in the last year of their life, with more than 30% spending more than 30 days in hospital, 55% dying in hospital, 12% attending the intensive care unit and only 24% receiving palliative care [8].

Advance care planning (ACP) is a process that explores a person’s values, beliefs and preferences regarding future health and personal care to guide medical decision making at a future time when that person can no longer communicate his or her decisions. This process usually involves the patient, family members, other important persons and healthcare providers [9]. ACP has been shown to improve the likelihood that doctors and family members will know and comply with the patient’s wishes [9–11], increase hospice and palliative care use, decrease inappropriate life-sustaining treatments and hospital admissions, improve patient and family satisfaction with care, as well as reduce stress, anxiety and depression in surviving relatives [10, 12–14]. ACP is particularly important in the dementia context because the cognition and decision-making capacity of individuals with dementia becomes increasingly limited as the disease progresses, and ACP becomes the sole vehicle to foster autonomous decision making [15].

Because ACP increases the likelihood that the medical treatment wishes of the patient will be known and respected, it also has an impact on the cost of care at end–of-life. Two recent systematic reviews examining the effects of ACP on cost of end-of-life care were identified. Klingler et al. (2015) showed that facilitated ACP had the potential to reduce net costs of care [16], while Dixon et al. (2015) found associations between ACP and healthcare savings for people living with dementia in the community, and for people living in areas with high end-of-life care spending [17]. However, the authors did not systematically identify both costs and outcomes, or the potential trade-off between these factors.

This study aims to overcome the limitations of past economic evaluations of ACP. We developed an economic model to assess the cost-effectiveness of an ACP intervention in a cohort of older people (aged 65+ years) who were at risk of cognitive decline in the context of the Australian healthcare system. The Markov modelling technique was used to simulate a decision problem involving risk over time (e.g. progression of dementia), when the timing of events is important (e.g. introducing ACP before loss of ability to communicate in people with moderate to severe dementia) and when important events may repeat (e.g. treatment or care for episodes of illness). It can mimic the clinical pathway and allows the capture of costs and outcomes associated with important events over time. Markov models have been widely used in economic evaluations of healthcare interventions to inform funding decisions.

This is the first dementia-specific study to evaluate the cost-effectiveness of ACP in a simulated population using the Markov modelling technique. These findings demonstrate that providing a nationwide ACP intervention comprised of three to four consultations (30–90 min each) for people aged 65+ years, who are at risk of dementia, is cost effective, provided that the ACP uptake and compliance with end-of-life wishes rates are at least 50% and 75%, respectively. Whilst we acknowledge that cost savings should not be the main aim of introducing a wider ACP policy, our model showed that a secondary positive outcome was the reduction in healthcare expenditure, making those resources available to other patients and population health interventions.

In line with the concept that ACP should be initiated ‘earlier rather than later’ in people with dementia, before the loss of capacity and ability to communicate occurs, our model indicated cost-effectiveness diminished when ACP was introduced later in life. This finding is not surprising as ACP discussion may be less feasible with an individual with moderate or severe dementia. For example, a systematic review examining the effectiveness of conducting ACP in people with dementia found that the majority of participants did not have the capacity to complete an ACP discussion in the nursing home setting and thus had to defer the decision-making responsibility to surrogates [34].

Our model showed that the economic benefit of a nationwide ACP program was dependent on a moderate uptake rate and a high compliance rate in an environment where dying outside hospital was a preferred choice. We found that an uptake rate as low as 26% was still cost saving if the majority of individuals preferred end-of-life care outside hospitals and if their wishes were respected. It is noted that the ACD is legally binding under statute legislation throughout most of Australia and under the Common Law in New South Wales and Tasmania, thus a high compliance rate (given a moderate-to-high uptake rate) is warranted. Additionally, if a national ACP program were to be introduced, it should contain an education component for both health care providers and community members to ensure that they understand that ACP compliance is not optional. Therefore, a critical consideration here is whether or not the demand for high quality palliative care services, both at home and in the hospice setting, can be met and, therefore, whether or not people’s wishes are feasible at the end of life.

Dying in hospital facilities and the use of life-sustaining measures are individual preferences that should be fully respected. However, delayed end-of-life decision-making usually leads to unwanted and/or unwarranted aggressive life-sustaining measures being instigated for those approaching end of life, including those who are imminently dying. Unwanted heroic measures have been shown to negatively influence quality of life in the week before death [37, 39], which may explain why most people prefer to die at home rather than in admitted inpatient settings [33]. This may be especially relevant for people in the advanced stages of dementia due to the prospect of imminent death, where caring and comforting, rather than curing, are needed. People with a preference to die at their residence (e.g. home or aged care facility) will require their carers and homes to be sufficiently supported and equipped to provide end-of-life care. Assessment of the current and potential risks for emergency and inpatient admission, and formulating a plan of action, may help prevent unwanted hospitalisation when a crisis or other intense situation occurs.

The ability to fully realise the benefits demonstrated in the modelling is dependent on accessible high quality community and palliative care services. Without access to support at home, it is likely that people at end of life will end up dying in hospital despite their expressed wishes. Our model suggested that an additional 15% of older people would die outside of hospital if this model of ACP was fully implemented. It is unlikely that current services would have capacity to care for these additional people at home and implementing this model of ACP in Australia would necessarily require an additional investment in community services by government and/or a realignment of funding from hospital care to community and residential aged care facilities. It is noted by the Productivity Commission [37] that approximately 70% Australians preferred to die at home but less than 10% were able to do so, and 20% of older people living in residential aged care facilities died in hospital due to a lack of palliative care expertise and qualified staff.

A complex multi-faceted ACP intervention is more likely to increase compliance with individuals’ end-of-life wishes [11]. The ACP program proposed in our paper is based on a series of consultations in the primary care setting, which is often considered an effective setting for ACP. First, general practitioners and other community-based health professionals are in an ideal position to facilitate ACP because they have an ongoing relationship with the patient and often understand his/her healthcare needs [20]. It is also recognised that people are more empowered in primary care than in other healthcare settings [21]. Second, community dwellers are still generally in good health and more likely to have decision-making capacity. As noted in Robinson et al. (2012) [34], residential aged care settings might be too late for starting the ACP journey as many residents have lost their decision-making capacity.

Furthermore, recent best-practice recommendations (Draft Recommendation 4.3 from the Productivity Commission’s Draft Report of June 2017 [37]) suggest the Australian Government should promote and aim to increase the uptake of ACP in primary care by a) requiring the general practitioner to introduce the concept of ACP and b) introducing a new Medicare item number to fund practice nurses to facilitate ACP. If such initiatives were implemented nationally we believe the feasibility of meeting the ACP uptake and compliance rates required in our model would be met and, thus, increase the likelihood of ACP being a cost-effective intervention in the primary care setting.

The lack of continuity of care and poor information sharing between primary and tertiary medical practitioners currently limits the effectiveness of ACP [40]. Technological innovations such as smart phone applications or digital wrist bracelets are potential vehicles to disseminate ACP information and documents during an emergency [38]. The recently-developed electronic health record in Australia also provides an ideal measure to carry ACP with the patients regardless of their location.

Despite using the best available evidence, a number of assumptions and simplification were necessary in our model. First, we estimated the age-dependent utility, mortality and end-of-life hospitalisation costs from the literature using multivariate regression analyses. Second, we assumed a constant rate of transition from mild to moderate to severe dementia, as well as stable preferences for end-of-life hospitalisation across disease state and age. We have performed extensive sensitivity analyses and highlighted influential factors (e.g. compliance rates and hospitalisation preferences). While these were strongly grounded on the existing literature, they should be revisited when new information becomes available. Lastly, we could only evaluate the ACP program using a health-system perspective, instead of a societal perspective. It is known that end-of-life care and medical decisions often impose a heavy burden, represented by potential productivity loss and a reduction in the mental and physical health of family and caregivers. While there are a large body of qualitative literature highlighting the burden, our literature search indicated that there were no reliable quantitative studies that could be used for this model. Additionally, if the (base case) ACP model was found to be cost-effective under the health-system perspective, which ignored the social costs avoided by family and caregivers if an individual has an ACP, a model with a social perspective would have a result that leads to the same cost-effective conclusion. The thresholds for ACP uptake and compliance rates and preferences for end-of-life hospitalisation, from a social perspective, would have been lower than those for the health system perspective.

Our model indicates that a nationwide ACP intervention in the Australian primary care setting is a cost-effective option compared to usual care for the older population at risk of developing dementia. The result is largely driven by providing treatment and care consistent with patient ACP preferences, leading to fewer hospitalisations and less-intensive care at end-of-life. In order to meet the ACP uptake and compliance rates required in the model for ACP to be cost-effective, we recommend the Australian Government supports a national and consistent approach to ACP in the primary care setting.

Australia is ready to take the next step toward respecting end-of-life choices. Taking the initiative and time, and a positive attitude towards anticipating future scenarios seems universal across different parties and studies. All jurisdictions have developed regulatory and legislative frameworks to support ACP. There is also strong support for normalising ACP and promoting it earlier in life. Raising community awareness of ACP and adjusting attitudes towards ACP in primary care settings will enable the existing healthcare infrastructure to accommodate ACP, without a significant paradigm shift in service models. The ACP, which allows an individual to expressly state their wishes for end-of-life care, helps families, caregivers and health professionals to act, as far as possible, in accordance with the person’s own expressed and considered wishes and expectations.