Cognitive and behavioural bias in advance care planning

Abstract

Background:

We explore cognitive and behavioural biases that influence individual’s willingness to engage advance care planning (ACP). Because contexts for the initiation of ACP discussions can be so different, our objective in this study was to identify specific groups, particular preferences or uniform behaviours, that may be prone to cognitive bias in the ACP decision process.

Method:

We collected data from the Australian general public (n = 1253), as well as general practitioners (GPs) and nurses (n = 117) including demographics, stated preference for ACP decision-making; six cognitive bias tests commonly used in Behavioural Economics; and a framing experiment in the context of ACP.

Results:

Compared to GPs (M = 57.6 years, SD = 17.2) and the general public (58.1 years, SD = 14.56), nurses on average recommend ACP discussions with patients occur approximately 15 years earlier (M = 42.9 years, SD = 23.1; p < 0.0001 in both cases). There is a positive correlation between the age of the general population and the preferred age for the initial ACP discussion (ρ = 0.368, p < 0.001). Our shared decision-making analysis shows the mean share of doctor’s ACP input is viewed to be approximately 40% by the general public, significantly higher than health professionals (GPs and nurses), who believe doctors should only contribute approximately 20% input. The general public show varying relationships (all p < 0.05) for both first ACP discussion, and shared decision-making for five of six cognitive tests. However, for health professionals, only those who exhibit confirmation bias show differences (8.4% higher; p = 0.035) of patient’s input. Our framing experiment results show that positive versus negative framing can result in as much as 4.9–7.0% shift in preference for factors most relevant to ACP uptake.

Conclusion:

Understanding how GPs, nurses and patients perceive, engage and choose to communicate ACP and how specific groups, particular preferences or uniform behaviours, may be prone to cognitive bias in the decision process is of critical importance for increasing future uptake and efficient future healthcare provision.

Keywords

advance care planning cognitive bias end of life framing shared decision-making

Introduction

Advance care planning (ACP) and end-of-life (EOL) decision-making are topical issues given an ageing Australian population and its future impact on healthcare services.^1–3 ACP is a process that allows an individual to discuss, plan, and communicate their desires, wishes, and preferences about their future healthcare to family, friends, and health professionals.⁴ It is a complex process; however, it has clear benefits for the individual involved by improving quality of life throughout EOL care, and assuring patients’ wishes for care are explicitly met.⁵ ACP can also alleviate stress and anxiety for family and loved ones, as well as reduce the psychological, emotional, administrative and economic burden on the healthcare professionals and organisational systems involved.⁶ While the associated benefits of ACP appear clear in principle, patients’ understanding and uptake remain low. The available data indicate that only 14% of the Australian population had advance directives (a formal record of an individual’s directives for future healthcare), with numbers varying significantly between states and territories.⁷ This is also replicated in other parts of the world.⁸

A lack of patient knowledge about ACP has been shown to be one of the primary reasons for low ACP uptake.⁶ The literature suggests that interventions that increase communication about ACP naturally lead to increased directive completions.⁹ Older people express clear preferences for future EOL care; however, resulting healthcare communications continue to remain inadequate.¹⁰ ACP communication and decision-making research is globally topical, with some critics arguing that the current EOL model of shared decision-making is in effect ‘illusory’ (p. 114).¹¹ This is because in real-life situations, shared decision-making regarding EOL care choices will always be in some part ‘incomplete’ (p. 461)¹² as medical experts’ advice can effectively bias patient’s choices. The ability to make autonomous choices is even more compromised when complex care is required. A more comprehensive understanding of the factors influencing ACP decision-making warrants investigation. One way is by using behavioural economics (BE), which moves beyond the neo-classical and traditional health economics of unidimensional cost benefit analysis. BE instead incorporates the effects and impact of cognitive, emotional, psychological and socio-cultural factors in individual and organisational decision-making.¹³ BE research methods have previously been used to explore medical expert and patient communication and behaviour across a range of allied health settings, including pharmacy, reconstructive surgery and breast care nursing.^14–16 More specifically to ACP, BE research has shown that the way questions and information are framed to patients in EOL decision-making can impact their preferences and choices.^17,18 Studies have also explored other cognitive barriers to ACP uptake, and the potential for the use of behavioural theories in EOL care decision-making.^19,20 In fact, simply being aware of potential behavioural biases can assist patient’s ability to revise counterproductive beliefs in the ACP decision process.¹⁹

To further enhance the knowledge on this topic, the objective of this study was to explore cognitive biases and key differences in communication, preference and decision-making in the context of ACP for both the general public, as well as general practitioners (GPs) and nurses with an interest in primary care. The study also explored individuals’ perceptions of their role in choice and potential shared decision-making with medical experts and identified how framing effects might influence changes in preference for possible motivating factors to engage in ACP. Because contexts for the initiation of ACP discussions can be so different, studies such as this are useful in identifying specific groups, particular preferences or uniform behaviours that may be prone to cognitive bias in the decision process.

Methods

Data collection, sample size and response rate

Our study comprises of two samples: (Sample 1) an age-representative sample of the Australian population and (Sample 2) a sample of Australian healthcare professionals.

For the general public sample, participants were surveyed online using the Qualtrics survey software between 21 and 25 May 2021. Australian participants aged 18–80 (n = 1248) were recruited by Lucid (https://luc.id/marketplace/), a commercial research company with an online survey respondent community. All people 18 years of age and older at the time of the survey were eligible to participate. All responders received a token payment for the full survey completion. Data were collected from 21 to 25 May 2021.

The healthcare professional sample comprised conference attendees of the General Practice Conference and Exhibition (GPCE), in May 2021 at Homebush, Sydney. Conference attendees were approached and invited to participate in person by the research team on the first two days of the conference. Our sample represents 48.1% (n = 25) of the 53 nurse attendees and 23.59% (n = 92) of the 390 GPs who attended the conference on those days. Healthcare professionals were incentivised to participate with a voluntary random prize draw of two amounts of AUD$500. A total 104 of the 117 healthcare professionals surveyed entered the random prize draw.

Survey design

We designed two surveys for Sample 1 and Sample 2. Both surveys (see Appendix 1 for survey questions in full) were designed to capture participants’ knowledge and preferences regarding ACP engagement and measure any cognitive bias. The questions used are validated survey measures repeatedly used in BE and applied psychology research.^13,15,16,21 For both samples, we asked participants (1) what the best age for initial ACP discussion between patient and healthcare professional is, and (2) their preference for the degree of shared decision-making between doctor and patient in deciding the content of any potential ACP. Cognitive bias of the participants was measured using six different bias tests, those being; conjunction fallacy, illusion of control, endowment effect, herd bias, confirmation bias and loss aversion. Each cognitive bias response is then treated as a binary variable, with the participant either exhibiting the bias, or not. All are commonly used in BE for scenarios of decision-making under constraint or risk.^13,21 Table 1 provides definitions for each bias, as well as practical examples.

Table 1.

Behavioural bias test and definition.

Bias	Definition	Practical example
1. Conjunction fallacy	When an agent’s decision-making is in error from the assumption that the conjunction of two possible events is more likely or probable than a single event	When considering an ACP, patients may join together the likelihood of multiple health outcomes, thus over-estimating, rather than see each individual outcome as independent
2. Illusion of control bias	When individuals overestimate their control over specific events that are patently not within their capacity or influence	Doctors and nurses may make decisions based on previous experiences in which outcomes were dictated by factors not relevant in the current setting
3. Endowment (effect) bias	When an agent’s maximum willingness to pay is typically lower than the least amount they are willing to accept. Loss aversion is associated with ownership	Patients without an ACP may overweigh the current value of not having an ACP, propitiate to talking the time to invest in creating one in the future
4. Herding bias	Refers to an agent demonstrating a tendency to follow or copy what others are doing. A misbelief that that is the right course of action purely because majority have chosen it	Patients, doctors and nurses may all gravitate to the behaviour the majority engage in, even if this is not necessarily the best outcome for themselves or others
5. Confirmation bias	Is the tendency for an agent to selectively interpret, favour or search for information that supports their own values or prior beliefs, all the while ignoring data or facts that are not supportive or their position	Patients, doctors and nurses may inadvertently exclusively seek out information that validates their own opinion (or diagnosis), rather than make an independent assessment
6. Loss aversion	Loss aversion refers to an agent’s tendency to favour avoiding losses to the acquisition of equivalent gains	Patients and doctors may make medical decisions based on a risk averse position, rather than an independent assessment based on the information available

ACP, advance care planning.

Moreover, we incorporated a randomised framing experiment into the survey design to assess how participants’ preference towards ACP uptake is affected by framing. Specifically, we asked participants to rank, from their most preferred to least preferred, five different reasons for ACP uptake presented in either positive or negative connotations. Half of the participants were randomly allocated to the treatment where reasons were framed as benefits (e.g. an ACP may reduce unwanted financial costs) and the other half to reasons framed as drawbacks (e.g. without an ACP you may experience unwanted financial costs). Furthermore, we included three additional questions on personal experience with ACP in the Australian general public survey, including (1) do participants know what ACP is, (2) if they have completed an ACP and (3) if they have assisted with or participated in an ACP for friends or relatives. Description of ACP was provided to the participants after indicating whether or not they previously knew about ACP (1). These personal ACP experience questions were asked before other ACP-related questions.

For both groups, we collected demographic information (age and sex), while for GPs and nurses, we also collected data on their job title and their years of experience in that role.

Statistical analyses

Data were analysed using Stata 16.1. We begin with some descriptive analysis of our outcome variables of interest, including Pearson’s correlation, pairwise comparisons, two-sample t-tests presented with 95% confidence intervals and level of significance with Bonferroni’s correction for multiple comparison. We then proceed to our ordinary least squares multivariate analysis.

Ethical approval

All participants provided informed consent, and all research was conducted in accordance with the QUT Human Research Ethics Committee protocol clearance (approval no. 2021000128).

Results

Participant characteristics of the Australian general public and healthcare professionals are summarised in Table 2. Average age of the general public sample is 41.3 years (SD = 17.4). Male participants (45.5%) are, on average, 10.7 years older than female participants. Approximately two-thirds of the GP participants are male (66.3%). The mean age of GPs is 54.1 years (SD = 13.6) with male GPs being 9.1 years older and with 6 years more on-job experience than female GPs, on average. Only two out of the 25 nurses surveyed are males. The average age of nurse participants is 55.7 (SD = 10.5). Pearson’s correlation between age and experience for health professionals is high (ρ = 0.842) and were run for both the combined sample 0.842 (n = 116), with GPs (ρ = 0.903) and nurses (ρ = 0.565), indicating that the sample of nurses has a larger variance in terms of age when career begin.

Table 2.

Summary statistics by group.

Australian general public (n = 1248)	Mean	SD	Min	Max
Male (%)	45.5
Age	41.3	17.4	18	80
Female	36.51	15.4	18	78
Male	47.18	17.9	18	81
Knew about ACP (%)	33.3
Completed an ACP (%)	14.1
Assisted with or participated in an ACP of friends or relatives (%)	21.1
Optimal age for initial ACP discussion	58.1	14.6	16	80
Share of doctor’s input in ACP content	39	31.3	0	100
General practitioners (n = 92)	Mean	SD	Min	Max
Male (%)	66.3
Age	54.1	13.6	24	77
Female	48.0	13.4	26	70
Male	57.2	12.7	26	77
Years of experience	24.3	13.3	2	53
Optimal age for initial ACP discussion	57.6	17.2	16	80
Share of doctor’s input in ACP content	18	20.1	25	100
Nurses (n = 25)	Mean	SD	Min	Max
Male (%)	8
Age	55.7	10.5	31	69
Female	56.7	10.0	31	69
Male	44	11.3	36	52
Years of experience	27.3	13.1	4	50
Optimal age for initial ACP discussion	42.9	23.1	16	80
Share of doctor’s input in ACP content	19.2	22	30	100

ACP, advance care planning.

Approximately one-third of the general public participants were familiar with ACP and only 14.1% and 21.1% of the participants reported having completed an ACP (which is representative of broader Australian public)⁷ and have been involved with an ACP of their friends or relatives, respectively.

Preferred age to first discuss ACP

All participant groups were asked which age they believed was best to first open a discussion with a patient regarding ACP. To restrict outliers, responses were bounded between 16 and 80 years of age. The average ideal age of initial ACP discussion for the general public is 58.1 years (SD = 14.56), which is not statistically different (unadjusted p-value = 0.737) from the average of the GP sample (M = 57.6 years, SD = 17.2, see Figure 1). However, we find that the nurse participants prefer the first ACP discussions with patients to occur approximately 15 years earlier in patients who are in their early 40s (M = 42.9 years, SD = 23.1) compared to the general public and the GP sample. These differences are statistically significant (p < 0.001 in both cases). Furthermore, we find that the distribution of ideal age of initial ACP discussion to be bimodal in the nurse sample, whereas for GPs and the general public, the distribution appears to be left skewed.

Figure 1.

Ideal age of first ACP discussion by group.

We find some variation within each sample with respect to participant characteristics. For example, Australian males seem to prefer a slightly later initial ACP discussion in life (M = 59.4 years, SD = 14.8) compared to Australian females (M = 57.1 years, SD = 14.3; p = 0.0047); however, this is not apparent when age is controlled for, whereas for GPs, there were no statistically significant difference between male and females (p = 0.433). Sex difference comparisons for our nurse sample were not possible due to only two male participants. More interestingly, for the general public sample, we find that preferred age for first ACP discussion is positively correlated with participants’ age [ρ = 0.368, p < 0.001, see Figure 2(a)], for both male (0.334) and female (0.391). However, this age bias was not present in the GP and nurse samples [pooled ρ = 0.055, p = 0.563, Figure 2(b)], even when differentiated by occupation or sex. As such in Figure 2, GPs and nurses are grouped together for simplicity. Furthermore, ideal age of first ACP discussion is also not correlated with health professionals’ year of job experience.

Figure 2.

Correlation between participant age and preferred age of initial ACP discussion, by group.

Shared decision-making in ACP

In terms of participant preference for the share of contribution to an ACP between patients and doctors, the public hold a more mixed view compared to health professionals. As shown in Figure 3 (also see Table 2), the mean share of doctor’s ACP input is viewed to be approximately 40% for the general public, which is significantly higher compared to health professionals (GPs and nurses), who believe doctors should only contribute about 20% input in terms of designing the patient’s ACP. Moreover, the variance of the distribution for the general public is substantially larger than health professionals (p < 0.001, based on a two-tailed equality of standard deviations), indicating the former has more diverse opinions in this matter. Nonetheless, there was no statistically significant difference between the GP and nurse sample in level of shared ACP decision-making preference. Furthermore, there was a significant sex difference within both the general public (p < 0.001) and GP (p = 0.071) sample (see Figure 10), in which female participants deemed patients should have a higher share in deciding the content of an ACP (difference of 8.8% and 8% points, respectively). No apparent relationship between participants’ age and share of doctor–patient ACP decision-making was found (see Figure 11).

Figure 3.

Share of doctor–patient contribution in ACP decision-making, by group.

Cognitive bias and ACP

In Figure 4, we present our six cognitive bias test findings, differentiated by group. For five of our six [Figure 4(a)–(e)] tests (with the exception of loss aversion), there were statistically significant differences between the general public and GP populations. Specifically, the general public exhibit less conjunction fallacy and herd bias than GPs, but experience more illusion of control, endowment effect, and confirmation bias. For most biases, there was no statistically significant difference between nurses and the other two samples; however, this finding is likely due to the small sample size of nurses. The only significant difference was between GPs and nurses with less GPs exhibiting illusion of control bias [Figure 4(b)].

Figure 4.

Cognitive bias two-sample comparisons, by group.

Next, we examine whether these cognitive biases are correlated with the timing participants prefer one should initiate a discussion about ACP with healthcare professionals and their preferences for level of shared decision-making regarding ACP between doctor and patient. To do so, we first compare the averages of the outcome between participants who exhibit bias to a specific behavioural aspect to those who do not (Figure 5), then, using a multiple regression approach, we assert the effect of these behavioural bias by controlling for other potential confounding factors (Tables 3 and 4).

Figure 5.

Cognitive bias and ACP decision-making process.

Table 3.

Multivariate analysis on ideal age for initial ACP discussion.

	Australian general public				Health professionals
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Behavioural bias
Conjunction fallacy	0.1973 (0.8252)	0.2557 (0.7732)	0.2637 (0.7731)	0.0688 (0.8013)	2.642 (3.837)	2.135 (3.814)	2.535 (3.906)
Illusion of control bias	−2.632** (0.8553)	−0.0406 (0.8344)	−0.0683 (0.8355)	−0.3503 (0.8485)	−4.005 (3.674)	−3.436 (3.897)	−3.372 (3.905)
Endowment effect	0.8408 (0.8423)	1.006 (0.7908)	0.9371 (0.7938)	0.9356 (0.8249)	1.702 (3.43)	1.731 (3.498)	1.574 (3.53)
Herd bias	2.23** (0.8266)	1.24 (0.7907)	1.156 (0.7932)	0.4819 (0.8228)	−1.238 (4.028)	−0.9735 (4.104)	−1.238 (4.213)
Confirmation bias	1.544^† (0.8235)	1.123 (0.7767)	1.161 (0.7769)	1.073 (0.8081)	−5.546 (3.942)	−5.472 (3.949)	−5.613 (3.993)
Loss aversion	2.057* (0.8346)	0.7336 (0.7937)	0.7134 (0.7934)	0.6644 (0.8207)	1.249 (3.626)	1.193 (3.665)	0.9402 (3.76)
Participant’s age		0.3115*** (0.0235)	0.314*** (0.0237)	0.2554*** (0.0305)		0.0132 (0.1228)	0.1139 (0.2219)
Male		−1.151 (0.813)	−1.193 (0.8152)	−1.725* (0.8715)		3.753 (4.251)	3.523 (4.293)
Experience with ACP			1.13 (0.8038)	0.5721 (0.8348)
Nurse					−14.18* (5.415)	−12.24* (6.058)	−12.19* (6.043)
Years of job experience							−0.1174 (0.2399)
Constant	56.01*** (1.247)	43.42*** (1.53)	42.98*** (1.555)	39.16*** (3.196)	59.01*** (5.339)	55.73*** (9.566)	53.43*** (9.862)
Additional control	No	No	No	Yes	No	No	No
N	1248	1246	1246	1138	112	112	112
R ²	0.0201	0.141	0.143	0.191	0.141	0.148	0.15
Adjusted R²	0.015	0.136	0.137	0.165	0.083	0.073	0.066
AIC	10,213.7	10,038.3	10,038.2	9139.9	976.8	979.8	981.6
BIC	10,249.6	10,084.4	10,089.5	9326.3	998.6	1007.0	1011.5

ACP, advance care planning; AIC, Akaike information criterion; BIC, Bayesian information criterion.

Dependent variable: Ideal age for initial ACP discussion. Standard errors (robust) in parentheses.

†

p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Table 4.

Multivariate analysis on share of doctor–patient contribution in ACP decision.

	Australian general public				Health professionals
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Behavioural bias
Conjunction fallacy	−3.387^† (1.759)	−2.934^† (1.714)	−2.919^† (1.713)	−2.727 (1.75)	−0.5614 (4.259)	−0.344 (4.15)	−0.0472 (4.369)
Illusion of control bias	4.639* (1.852)	2.98 (1.881)	2.926 (1.882)	3.104 (1.937)	6.248 (3.971)	6.113 (3.907)	6.12 (3.937)
Endowment effect	−1.744 (1.807)	−1.791 (1.765)	−1.925 (1.762)	−2.716 (1.77)	−0.5707 (3.852)	0.3392 (3.71)	0.2643 (3.818)
Herd bias	5.6** (1.754)	5.96*** (1.717)	5.796*** (1.72)	4.949** (1.779)	2.517 (4.069)	2.803 (3.914)	2.589 (3.982)
Confirmation bias	2.861 (1.756)	2.574 (1.713)	2.647 (1.715)	2.474 (1.737)	−8.773* (3.953)	−9.087* (3.824)	−9.222* (4.006)
Loss aversion	−3.279^† (1.818)	−2.757 (1.786)	−2.796 (1.786)	−2.616 (1.831)	0.1331 (3.97)	0.9898 (3.881)	0.8475 (3.869)
Participant’s age		−0.3579*** (0.0518)	−0.3532*** (0.0518)	−0.3964*** (0.0697)		−0.2383 (0.1509)	−0.1611 (0.3346)
Male		12.63*** (1.791)	12.55*** (1.789)	12.64*** (1.893)		9.962* (4.078)	9.768* (4.233)
Experience with ACP			2.198 (1.734)	2.411 (1.768)
Nurse					−0.8542 (4.918)	5.375 (5.096)	5.413 (5.107)
Years of job experience							−0.0903 (0.3384)
Constant	36.44*** (2.627)	45.97*** (3.367)	45.12*** (3.431)	68.7*** (6.775)	16.51* (7.117)	21.63* (9.902)	19.87 (12.2)
Additional control	No	No	No	Yes	No	No	No
N	1248	1246	1246	1138	115	115	115
R ²	0.0226	0.0771	0.0783	0.167	0.0652	0.117	0.118
Adjusted R²	0.018	0.071	0.072	0.140	0.004	0.041	0.033
AIC	12,120.7	12,031.5	12,031.9	10,924.1	1026.7	1024.1	1026.0
BIC	12,156.6	12,077.6	12,083.2	11,110.5	1048.7	1051.6	1056.2

ACP, advance care planning; AIC, Akaike information criterion; BIC, Bayesian information criterion.

Dependent variable: Share of input contributed by the doctor. Standard errors (robust) in parentheses.

†

p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

This simple mean (t-test) comparison analysis reveals that participants who exhibit the illusion of control bias are more likely to prefer the initial discussion of ACP to happen in earlier life stages [difference by 2.5 years (p = 0.005) for Australian general public participants and 7.6 years (p = 0.035) for health professionals, respectively]. Other behavioural biases appear to have no related effect on health professionals, while the general public who exhibit herding bias, confirmation bias, and loss aversion state an older ideal age for first ACP discussion.

Interestingly, general public participants who exhibited behavioural biases rated the share of doctor–patient contribution in ACP decisions differently. Specifically, participants rate the share of doctor’s input in deciding ACP content to be higher if they exhibit herding bias (4.77%, p = 0.01) or illusion of control bias (6.24%, p = 0.0004), but if conjunction fallacy or loss aversion is present, participants tend to rate patient’s input to be higher (3.49%, p = 0.049 and 3.77%, p = 0.041, respectively). In contrast, behavioural bias did not appear to affect health professionals’ view on ACP decision-making between doctor and patients, with the exception of confirmation bias. More specifically, health professionals with confirmation bias rated patient’s input to be 8.4% higher than those who do not (p = 0.035).

In our multivariate analysis, we controlled for basic demographics (i.e. age and sex) of participants as they were previously identified to be correlated with the two outcome variables. For the general public sample, we also controlled for participants’ experience with ACP, which is coded as a binary variable with value equals to one if the participants have answered ‘Yes’ to any of the three questions relating to personal experience with ACP. Furthermore, we included an extensive range of socio-demographic variables to the analysis of the general public sample, including education, type of schooling, ethnicity, household income, marital status, number of offspring, religion, political views, self-rated happiness and self-rated health. For healthcare professionals, we included years of job experience in addition to sex and age. Control variables were procedurally added in the regression analysis in a stepwise manner as a robustness check for coefficient estimates.

In Table 3, after the participants’ age and sex were controlled for, the effects of behavioural biases were not statistically significant. Those with a history of any form of ACP experienced no difference in their preference to those without.

In Table 4, participant age had a statistically significant negative correlation with the general public’s preference for amount of input by doctors into ACP content. Males compared to females preferred greater doctor’s input, and those who exhibited herding bias also preferred greater doctor’s contribution. Importantly, a history of any form of ACP experience appeared to have no impact on preference for contribution by doctor or patient.

As the majority of our general population sample have no previous experience with ACP, it is not surprising that cognitive short-cuts are employed in the decision-making process. As a robustness check, we explore the interaction of ACP experience and bias, on our two outcome variables in our general population sample.

For age of first ACP discussion, our multivariate results are presented in Table 5 with specification (7) visualised as Figure 6. All specifications include additional controls [those previously included in Table 3 specification (4)]. We find that those who exhibit confirmation bias or herding bias, and have prior knowledge of ACP, state a preference for later age for first ACP discussion.

Table 5.

ACP experience and bias interaction for age of first ACP discussion – General Pop.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Experience with ACP	0.9349 (1.177)	−0.6379 (1.383)	−0.6127 (1.334)	−1.735 (1.281)	−0.9236 (1.182)	1.328 (1.37)	−4.108 (2.504)
ACP experience × Conjunction fallacy	−0.7117 (1.64)						−0.4084 (1.659)
ACP experience × Illusion of control bias		1.878 (1.722)					1.37 (1.714)
ACP experience × Endowment effect			1.984 (1.69)				1.639 (1.712)
ACP experience × Herd bias				4.373** (1.634)			4.279** (1.653)
ACP experience × Confirmation bias					2.991^† (1.625)		2.807^† (1.653)
ACP experience × Loss aversion						−1.157 (1.69)	−0.9696 (1.685)
Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	1138	1138	1138	1138	1138	1138	1138
R ²	0.191	0.192	0.192	0.196	0.194	0.191	0.201
Adjusted R²	0.164	0.165	0.165	0.169	0.166	0.164	0.170
AIC	9141.7	9140.7	9140.4	9134.3	9138.4	9141.4	9138.5
BIC	9333.1	9332.1	9331.8	9325.7	9329.8	9332.9	9355.1

ACP, advance care planning; AIC, Akaike information criterion; BIC, Bayesian information criterion.

Dependent variable: Ideal age for initial ACP discussion. Standard errors (robust) in parentheses.

†

p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Figure 6.

ACP experience and bias interaction for first age of ACP discussion – General Pop.

For percentage share of ACP decision, our multivariate results are presented in Table 6 with specification (7) visualised as Figure 7. All specifications include additional controls [those previously included in Table 2 specification (4)]. In relation to contribution to an ACP. We find that those who exhibit confirmation bias in the general population, and have prior knowledge of ACP, prefer greater GP contribution in the decision process.

Table 6.

ACP experience and bias interaction for share of ACP decision – General Pop.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Experience with ACP	3.339 (2.546)	−1.165 (2.95)	−2.224 (2.845)	0.9507 (2.458)	−1.831 (2.558)	2.795 (2.968)	−8.626 (5.27)
ACP experience × Conjunction fallacy	−1.819 (3.505)						−1.675 (3.519)
ACP experience × Illusion of control bias		5.551 (3.682)					4.697 (3.686)
ACP experience × Endowment effect			7.763* (3.614)				6.65^† (3.637)
ACP experience × Herd bias				2.769 (3.509)			2.5 (3.513)
ACP experience × Confirmation bias					8.484* (3.506)		7.638* (3.523)
ACP experience × Loss aversion						−0.5872 (3.668)	−0.3728 (3.656)
Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	1138	1138	1138	1138	1138	1138	1138
R ²	0.168	0.169	0.171	0.168	0.172	0.167	0.176
Adjusted R²	0.140	0.141	0.143	0.140	0.144	0.139	0.145
AIC	10,925.8	10,923.8	10,921.3	10,925.5	10,920.2	10,926.1	10,923.9
BIC	11,117.2	11,115.2	11,112.8	11,116.9	11,111.6	11,117.5	11,140.5

ACP, advance care planning; AIC, Akaike information criterion; BIC, Bayesian information criterion.

Dependent variable: Share of input contributed by the doctor. Standard errors (robust) in parentheses.

†

p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Figure 7.

ACP experience and bias interaction for share of ACP decision – General Pop.

Framing effects on preferences for factors motivating ACP uptake

Prior to any analysis, it is methodologically important to qualify that we find no statistical difference between participant’s sex (p = 0.397 in the general public sample and p = 0.932 health professional sample) and age (p = 0.122 in the general public sample and p = 0.454 health professional sample). Years of experience also do not differ between health professionals who were exposed to the different conditions (p = 0.84) and all other sample characteristics (e.g. education, income, political views) do not differ across the positive and negative framing general public subsample.

In Table 7 and Figure 8, we present our framing experiment results again differentiated by group. In Figure 12 (see Appendix 1), we also present the complete distribution of rank preferences by ACP alternative for both positive and negative conditions.

Table 7.

Framing effect on priority of preference for engaging advance care planning.

Sample	Framing	Positive		Negative		z-statistics	p-value
Sample	Option	Mean	SD	Mean	SD	z-statistics	p-value
General public	Exact medical care	1.86	1.14	2	1.17	−2.52*	0.012
	Optimal EOL care	2.49	1.15	2.55	1.2	−0.77	0.442
	Family impact	2.79	1.15	2.85	1.21	−1.03	0.302
	Hospital transfers	3.9	1.06	3.76	1.15	1.87^†	0.062
	Financial cost	3.97	1.26	3.84	1.37	1.11	0.268
GP	Exact medical care	2.39	1.54	3	1.58	−1.76^†	0.079
	Optimal EOL care	2.26	1.16	2.43	1.39	−0.4	0.692
	Family impact	2.76	1.21	2.5	1.3	1.12	0.264
	Hospital transfers	4.07	1.08	3.65	1.04	2.12*	0.034
	Financial cost	3.52	1.19	3.41	1.36	0.27	0.791
Nurse	Exact medical care	2.77	1.88	2	1.35	0.84	0.4
	Optimal EOL care	3	1.47	2	0.74	1.77^†	0.078
	Family impact	2.31	0.95	3.5	1.24	−2.44*	0.015
	Hospital transfers	3.46	1.39	4.17	1.11	−1.2	0.229
	Financial cost	3.46	1.13	3.33	1.3	0.17	0.865

EOL, end of life; GP, general practitioner.

Wilcoxon rank-sum test (two-tailed).

†

*, ** and *** represent 10%, 5%, 1% and 0.1% levels of significance, respectively.

Figure 8.

Framing effect on ranked order of reasons for ACP uptake, by group.

For our general public group, exact medical care is on average the most prioritised factor, but that when alternatives are framed positively, participants rank the exact medical care (p = 0.012) option higher than those in the negative frame. Conversely, hospital transfers are ranked lower in a positive frame, although only at a 10% significance level (p = 0.062).

For our GP group, we see similar results in that positive framing of the exact medical care (p = 0.079) option results in higher priority, while again positively framing hospital transfers (p = 0.034) results in lower order preferences.

Finally in our nurse group, we find novel results in comparison to our previous two groups, in that nurses (on average) in our positive frame condition place higher priority on family impact (p = 0.015), but in a negative frame, we see nurses place higher priority on optimal EOL care although again only at a 10% significance level (p = 0.078).

Furthermore, by comparing the order of preference in pairs of ACP uptake reasons (Table 8), we find that framing causes the order of preference to switch for certain pairs. In particular, in our general public sample, 79.4% of the participants rank family impact as more important factor for ACP uptake than hospital transfers, while this share drops by 7% point when reasons were negatively framed (p = 0.004). Similarly, we also observe a 4.9% point shift in preference rank order for optimal EOL care compared to financial costs (p = 0.044).

Table 8.

Pairwise comparison framing experiment.

Reasons for ACP uptake	General public (n = 1248)			Health professionals (n = 117)
Reasons for ACP uptake	Negative framing (%)	Positive Framing (%)	Difference	Negative framing (%)	Positive framing (%)	Difference
Exact medical care > Optimal EOL care	32.90	28.50	4.4^†	56.90	50.80	6.0
Exact medical care > Family impact	30.30	27.20	3.1	46.60	39.00	7.6
Exact medical care > Hospital transfers	16.40	12.60	3.7^†	36.20	27.10	9.1
Exact medical care > Financial cost	20.10	17.60	2.5	39.70	30.50	9.1
Optimal EOL care > Family impact	38.80	37.40	1.4	37.90	44.10	−6.1
Optimal EOL care > Hospital transfers	22.00	18.20	3.8^†	25.90	20.30	5.5
Optimal EOL care > Financial cost	26.80	21.90	4.9*	27.60	28.80	−1.2
Family impact > Hospital transfers	27.60	20.60	7.0**	24.10	18.60	5.5
Family impact > Financial cost	26.60	22.60	4.1^†	31.00	30.50	0.5
Hospital transfers > Financial cost	42.20	41.30	0.9	62.10	59.30	2.7

ACP, advance care planning; EOL, end of life.

†

p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Discussion

Previous research exploring factors impacting EOL decision-making have primarily focussed on sample populations of the seriously ill, as well as the elderly,^9,22,23 and did not explore the role of bias in decision-making. Our study instead provides new and novel empirical findings from both frontline healthcare professionals and potential future patients relating to ACP communication and preference.

Triggers for engaging an ACP discussion are most often related to a significant new or ongoing health issue. That said, our study shows that the mean age where people consider starting discussion about ACP is 57, 58 and 42 years among general population, GPs and nurses, respectively. Nurses state a distinctly younger priority for the age of first ACP discussion with a patient (by 15.26 to the public and 14.71 to GPs, on average (p < 0.0001 in both cases), which is not surprising given the extensive involvement of nurses in day to day provision of EOL care. While the public’s preference exhibits a positive correlation with age, GPs and nurses show no such related age bias. The fact that the public prioritise EOL care decision-making primarily dependent on their own age [0.3677 (p < 0.001)] sheds possible light on why ACP uptake appears conditional on diagnosis, as well as significant health deterioration, rather than a conscious awareness of its future priority.

Our ACP shared decision-making analysis shows distinct differences between preferences of healthcare professionals and the public, with GPs (mean age = 54.1 years; mean years of experience = 24.3) and nurses (mean age = 55.7 years; mean years of experience = 27.3) stating (on average) approximately 20% greater patient contribution compared to what the public states they prefer. This finding that patients prefer substantially less input in such an important EOL health decision demonstrates the challenges associated with shared decision-making in practice and lends weight to critics of a shared decision-making model.^10,11 It also speaks to patient preference for paternalism²⁴ in credence markets^14,15 where frontline healthcare workers are the far more experienced medical experts. In such a large-scale health context (EOL decision-making), these empirical findings are novel and confirmatory, leading to conclusions that can have clinical meaning and inform future practices as they highlight the potential for conflict in decision-making and poor patient and carer experiences if expectations are not met. These findings raise concerns relating to patient expectation and guidance, and particularly relating to informed consent, and the practicalities of achieving shared decision-making when perspectives, knowledge and power differentials exist.

Key group differences in the way ACP stakeholders (patients, GPs and nurses) process and communicate information present challenges for efficient healthcare provision. Our cognitive bias analysis shows significant differences between GPs and patients for five of the six behavioural tests administered. For the general public, we find varying relationships (all p < 0.05) between both preferred age for first ACP discussion, and level of shared decision-making in ACP for five of six cognitive tests. However, for health professionals, only those who exhibit confirmation bias show differences in preference for patient’s input (8.4% higher; p = 0.035). That said, when we controlled for all factors in our multivariate regression analysis, we find only age, gender and herding bias as statistically significant factors in the general public preference for both age of first discussion and shared decision-making.

From a practical standpoint, our study provides evidence to support alternative ways to increase awareness of ACP through targeted communications based on the identification of key group differences in preference. For example, our framing experiment demonstrates that when ACP outcomes are presented in different ways, the general public and GPs show malleable preferences for more acute health-related issues like exact medical care (p = 0.012) and hospital transfers (p = 0.034), while nurses instead show changes in priority for more palliative or interpersonal related issues such as family impact (p = 0.015). Our pairwise comparisons take this a step further demonstrating the possibility of preference reversal in some cases, where differences in positive versus negative framing can result in as much as 4.9–7% change in general public preference for particular factors most relevant to ACP uptake. Practically, these findings demonstrate the importance of providing the appropriate examples in educational development and teaching aids for nurses, decision-making tools and counselling support services for patients.

This study is not without limitations, first, although our study collected a large sample of GP and nurse cognitive bias data from the GPCE conference, it was a convenience sample it may lack generalisability to all healthcare professionals. The nurse sample was also very small. Voluntary participation is another limitation of this study, as is the sample source, which includes people who are registered with a professional survey company. In addition, participant responses for content relating to ACP are stated preference, not revealed preference and reflect a point in time. Preferences for ACP and EOL decisions may, of course, change over time, depending on a range of social, clinical and environmental factors. Sex ratios and age for our GP and nurse samples are also highly skewed, although the demographic profile is broadly reflective of the current age and gender profiles for the related occupations. It is also important to note that other health professionals working in the ACP space may exhibit different cognitive processes and behaviours, for example, palliative care physicians. Furthermore, the study does not account for potential patient cognitive impairment, which is often the catalyst for initiating ACP discussions and processes. Finally, because the broader contextual complexity of ACP is so intricate (e.g. the role of culture, social norms, disease patterns, sex differences, etc), our study is exploratory in nature and seeks to offer a primer for the study of cognitive bias in ACP decision-making

Understanding how GPs, nurses and potential patients understand and communicate their preferences regarding ACP is of critical importance for efficient healthcare provision and future uptake. Overall, our study provides novel empirical evidence that cognitive bias plays a significant role both within and between (general public, GP and nurse) groups behaviour in the context of ACP. For the general public, age appears to be a robust and re-occurring factor associated with ACP preference and shared decision-making. This study can be a primer for future applied behavioural research in this important healthcare decision-making space.

Footnotes

Appendix 1 Author contribution(s)

Stephen Whyte: Conceptualisation; Data curation; Formal analysis; Investigation; Methodology; Project administration; Writing – original draft; Writing – review & editing.

Joanna Rego: Conceptualisation; Data curation; Methodology; Project administration; Writing – original draft; Writing – review & editing.

Ho Fai Chan: Data curation; Formal analysis; Methodology; Writing – original draft; Writing – review & editing.

Raymond J. Chan: Conceptualisation; Methodology; Writing – original draft; Writing – review & editing.

Patsy Yates: Conceptualisation; Methodology; Writing – original draft; Writing – review & editing.

Uwe Dulleck: Formal analysis; Funding acquisition; Methodology; Writing – original draft; Writing – review & editing.

Conflict of interest statement

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Ethical approval

All research was conducted in accordance with QUT human research ethics committee approval no. 2021000128.

ORCID iD

Stephen Whyte

Data availability statement

Data and code are available from the corresponding author on request.

References

Scott

Mitchell

GKJ

Reymond

, et al. Difficult but necessary conversations – the case for advance care planning. Med J Australia 2013; 199: 662–666.

Mitchell

. Facilitating better end-of-life care. Med J Australia 2012; 197: 68–69.

Mitchell

Willmott

White

, et al. A perfect storm: fear of litigation for end of life care. Med J Australia 2019; 210: 441.

Schuldt

Royal Commission into Aged Care Quality and Safety, 2019. https://www.aph.gov.au/About_Parliament/Parliamentary_Departments/Parliamentary_Library/pubs/rp/rp1920/Quick_Guides/RoyalCommissionAgedCare

Brinkman-Stoppelenburg

Rietjens

van der Heide

. The effects of advance care planning on end-of-life care: a systematic review. Palliat Med 2014; 28: 1000–1025.

Kermel-Schiffman

Werner

. Knowledge regarding advance care planning: a systematic review. Arch Gerontol Geriatr 2017; 173: 133–142.

White

Tilse

Wilson

, et al. Prevalence and predictors of advance directives in Australia. Intern Med J 2014; 44: 975–980.

Mast

. Against autonomy: how proposed solutions to the problems of living wills forgot its underlying principle. Bioethics 2020; 34: 264–271.

Houben

CHM

Spruit

Groenen

MTJ

, et al. Efficacy of advance care planning: a systematic review and meta-analysis. J Am Med Dir Assoc 2014; 15: 477–489.

10.

Heyland

Barwich

Pichora

, et al. Failure to engage hospitalized elderly patients and their families in advance care planning. JAMA Intern Med 2013; 173: 778–787.

11.

Drought

Koenig

. ‘Choice’ in end-of-life decision making: researching fact or fiction? Gerontologist 2002; 42(suppl_3): 114–128.

12.

White

Braddock

Bereknyei

, et al. Toward shared decision making at the end of life in intensive care units: opportunities for improvement. Arch Intern Med 2007; 167: 461–467.

13.

Kahneman

Thinking, fast and slow. London: Macmillan, 25 October 2011.

14.

Smith

Whyte

Chan

, et al. Pharmacist compliance with therapeutic guidelines on diagnosis and treatment provision. JAMA Netw Open 2019; 2: e197168.

15.

Whyte

Bray

Chan

, et al. Cognitive bias and therapy choice in breast reconstruction surgery decision-making. Plast Reconstr Surg 2022; 149: 629e–637e.

16.

Whyte

Bray

Chan

, et al. Knowledge, consultation time, and choice in breast reconstruction. Br J Surg 2021; 108: e168–e169.

17.

Dumont

Kissane

. Techniques for framing questions in conducting family meetings in palliative care. Palliat Support Care 2009; 7: 163–170.

18.

Vélez Ortiz

Martinez

Espino

. Framing effects on end-of-life preferences among Latino elders. Soc Work Health Care 2015; 54: 708–724.

19.

Weiner

Cole

. Three principles to improve clinician communication for advance care planning: overcoming emotional, cognitive, and skill barriers. J Palliat Med 2004; 7: 817–829.

20.

Winter

Parker

. Current health and preferences for life-prolonging treatments: an application of prospect theory to end-of-life decision making. Soc Sci Med 2007; 65: 1695–1707.

21.

Pompian

. Behavioral finance and wealth management: how to build investment strategies that account for investor biases, vol. 667. Hoboken, NJ: John Wiley & Sons, 2011.

22.

Steinhauser

Christakis

Clipp

, et al. Factors considered important at the end of life by patients, family, physicians, and other care providers. JAMA 2000; 284: 2476–2482.

23.

Lum

Sudore

Bekelman

. Advance care planning in the elderly. Med Clin 2015; 99: 391–403.

24.

Kassirer

Levine

Gaertig

. Decisional autonomy undermines advisees’ judgments of experts in medicine and in life. Proc Nat Acad Sci 2020; 117: 11368–11378.