Volume 29, Issue 1 p. 59-79
ARTICLE
Open Access

Do I care for you more when you really need help? An experimental test of the effect of clinical urgency on compassion in health care

Alina Pavlova

Corresponding Author

Alina Pavlova

Department of Psychological Medicine, University of Auckland, Auckland, New Zealand

Te Whatu Ora Counties Manukau, Auckland, New Zealand

Correspondence

Alina Pavlova, Department of Psychological Medicine, Faculty of Medical and Health Sciences, The University of Auckland, Building 507, F1018, Auckland, New Zealand.

Email: [email protected]

Contribution: Conceptualization, ​Investigation, Writing - original draft, Methodology, Visualization, Formal analysis, Project administration

Search for more papers by this author
Sarah-Jane Paine

Sarah-Jane Paine

Te Kupenga Hauora Maori, Faculty of Medical and Health Sciences, The University of Auckland, Auckland, New Zealand

Contribution: ​Investigation, Conceptualization, Validation, Writing - review & editing, Supervision, Data curation, Resources, Project administration

Search for more papers by this author
Alana Cavadino

Alana Cavadino

Epidemiology and Biostatistics, Faculty of Medical and Health Sciences, The University of Auckland, Auckland, New Zealand

Contribution: Software, Formal analysis, Writing - review & editing, Validation, Methodology, Resources

Search for more papers by this author
Anne O'Callaghan

Anne O'Callaghan

Department of Psychological Medicine, University of Auckland, Auckland, New Zealand

Auckland City Hospital, Auckland, New Zealand

Contribution: Conceptualization, ​Investigation, Writing - review & editing, Project administration, Resources

Search for more papers by this author
Nathan S. Consedine

Nathan S. Consedine

Department of Psychological Medicine, University of Auckland, Auckland, New Zealand

Contribution: Conceptualization, ​Investigation, Methodology, Writing - review & editing, Formal analysis, Project administration, Data curation, Supervision, Resources

Search for more papers by this author
First published: 30 August 2023

Abstract

Objectives

To experimentally investigate whether more urgent patient presentations elicit greater compassion from health care professionals than less urgent, facilitating future research and thinking to address systemic barriers to compassion in health care.

Design

This is a pre-registered online study with an experimental, within-subjects repeated-measure study design. Two clinical vignettes that systematically varied the urgency of patient presentation were utilized. Both vignettes depicted a patient with difficult behaviours typically associated with lower compassion.

Methods

Health care professionals (doctors, nurses and allied health practitioners) recruited from all 20 District Health Boards across Aotearoa/New Zealand completed two vignettes in a counterbalanced order. Paired-sample t-tests were used to test the effect of the presentation urgency on indices of compassion.

Results

A total of 939 participants completed the vignettes (20% doctors, 47%, nurses and 33% allied health professionals). As expected, participants reported greater care and motivation to help the more urgent patient. However, the more urgent patient was also perceived as less difficult, and exploratory analyses showed that perceived patient difficulty was associated with lower caring and motivation to help, particularly in the less urgent patient.

Conclusions

This is the first work to experimentally test the relationship between the urgency of patient presentation and compassion in health care. Although the association between urgency and difficulty is complex, our findings are consonant with evolutionary views in which urgent distress elicits greater compassion. A system-wide orientation towards efficiency and urgency may exacerbate this ‘bias’ which must be addressed to ensure more equitable compassion in health care.

BACKGROUND

Compassion—defined as ‘a sensitivity to suffering […] with a commitment to try to alleviate and prevent it’ (Gilbert, 2013)—is valued by patients (Bond et al., 2022), motivates and is expected from health care professionals (Baguley et al., 2020; Frampton et al., 2013; Goel et al., 2018), predicts better clinical outcomes (Fogarty et al., 1999; Kelley et al., 2014; Trzeciak et al., 2017) and forms an integral part of professional code of practice worldwide (American Medical Association, 2016; New Zealand Medical Association, 2020; NHS, 2022). Unfortunately, patients' experiences of compassion vary. Health care professionals may not be equally compassionate towards all patients (Zandbelt et al., 2007) and some appear to be more caring than others (Pavlova et al., 2021). Interestingly, although compassion has been conceptualized as a transactional phenomenon governed by the interactions among a range of influences (Fernando & Consedine, 2014) and is a strongly relational construct (Sinclair et al., 2016), most compassion research to date has only focused on how characteristics of health care professionals might be relevant (Pavlova et al., 2021; Wang et al., 2022)—an approach that has not proven to be particularly fruitful in producing effective interventions (Malenfant et al., 2022). To start addressing this research gap, our study was designed to experimentally examine the role of patient-related factors in facilitating (or hindering) compassion.

Evidence suggests that compassion in health care varies depending on how health care professionals respond to different types of patients (Bernardes & Lima, 2011; Carson et al., 2004; Escaffre et al., 2000; Hall et al., 2015; Hoffman et al., 2016; Singh et al., 2018). A recent systematic review identifying the predictors of physician compassion and related constructs (Pavlova et al., 2021) found that compassion was lower for less educated patients (Batley et al., 2016; Bertakis & Azari, 2011) and minoritized racial and ethnic groups (Foo et al., 2017; Park et al., 2018; Porthe et al., 2018), but more likely for patients reporting higher socio-economic status (Batley et al., 2016; Bertakis & Azari, 2011) as well as the elderly and children (Bayne et al., 2013; Zandbelt et al., 2007). Unsurprisingly, cross-sectional evidence also suggests that compassion is greater for patients who are cooperative or thankful (Picard et al., 2016; Porthe et al., 2018; Street Jr et al., 2007) but is challenged by patients seen as ‘difficult’—aggressive or angry, demanding or having crossed some moral boundary (e.g., drunk drivers, drug dealers or perpetrators of family violence; Derksen et al., 2016, 2018; Sandikci et al., 2017). In some views, ‘patients who are rude, demanding or difficult suck the oxygen from compassion’ (Fernando et al., 2016). Clearly then, compassion varies across patients, even when medical ethics requires equal care for all (Baker, 2001).

While health professionals views of patients immorality (Maestri & Monforte, 2020) and personal responsibility (Fernando & Consedine, 2014; Reynolds et al., 2019; Wang et al., 2022) may impact care, compassion may also vary according to clinical presentation (Fernando & Consedine, 2014; Pavlova et al., 2021). Early qualitative evidence suggests that physicians are more likely to empathize with and express compassion towards patients whose diagnoses are more straightforward, without co-morbidities and fitting within the scope of particular specialties (Batley et al., 2016; Bayne et al., 2013; Bertakis & Azari, 2011; Butalid et al., 2014; Sandikci et al., 2017; Zandbelt et al., 2007). Equally, cross-sectional evidence indicates that compassion is greater for patients requiring urgent care versus those who are more stable or improving (Batley et al., 2016; Bayne et al., 2013; Bishop et al., 2014; Zandbelt et al., 2007), particularly where urgency is visible or apparent (Baker et al., 2018; Paul-Savoie et al., 2018).

In considering this unhelpful dynamic of variability in compassion towards patients with different clinical presentations, it is important to note that, although compassion is fundamentally interpersonal, it also occurs (and does not occur) in particular organizational and clinical contexts (Singh et al., 2018). Prior studies have often measured and conceptualized of compassion as a trait (Sprecher & Fehr, 2005), implying a level of stability of compassion as a personal characteristic with the underlying aim to make people more compassionate such that they can reduce the impact of external factors such as patient backgrounds or clinical presentation. However, in line with assertions within the Transactional Model of Physician Compassion (Fernando & Consedine, 2014), we suggest that compassion as a motivation to alleviate suffering can be ‘turned on’ or ‘turned off’ by contextual factors. For example, medical training and organizational protocols tend to focus on clinically categorizing patients, relying on triage processes, risk assessments, eligibility criteria and other forms of prioritization which impacts and may reduce compassion (Smith et al., 2008). Empirical evidence suggests that this system-wide orientation towards efficiency and clinical urgency negatively affects compassion (Anandarajah & Roseman, 2014; Baker et al., 2018; Kerasidou, 2019; Kerasidou & Kingori, 2019; Rawal et al., 2020; Roze des Ordons et al., 2020), possibly explaining the frequent labelling of more chronic (but still suffering) patients as ‘entitled’, ‘time wasters’ and ‘attention seekers’ and/or the tendency to see them as more ‘difficult’ and less worthy of compassion (Batley et al., 2016; Bayne et al., 2013; Sandikci et al., 2017; Stein, 1986).

Of course, prioritizing the provision of medical care towards presentations that require urgent medical attention (e.g., life-threatening situations, acute infections, broken limbs, etc.) appears logical, both in terms of how the compassion systems likely evolved [i.e., caring for another (e.g., infant) with the associated need to be attentive to their distress and needs to ensure survival and/or generating and supporting affiliative peer bonds necessary for survival; see Gilbert, 2019, 2020], as well as in terms of immediate risk management and limited resource contexts (Patel et al., 2008). However, prioritizing medical care for more urgent presentations should not translate into lower compassion towards patients who are less urgent (e.g., presenting with non-specific complaints, chronic pain, etc.), impairing clinical outcomes and exacerbating inequities in care; lower clinical urgency does not necessarily mean lower suffering. In studying and understanding the link between compassion and urgency in health care, our belief is that we can start to identify compassion-focused solutions.

The current report

While identifying the clinical factors that might predict variations in compassion is important, we cannot be certain that it is characteristics such as urgency per se that are responsible for reduced compassion. Prior work has predominantly been conducted using qualitative methods or cross-sectional designs in small and single-occupation-focused samples (Pavlova et al., 2021). While studies with such designs are important in nascent areas of research, they are mostly exploratory, not generalizable and are prone to issues related to confounding. Without good control over predictor variables, it is challenging to discern whether associations are causative or are confounded on the basis of some third (unaccounted for) variable such as socioeconomic status, gender or ethnicity/race. To evaluate potential causality, there is a need for studies that experimentally test candidate factors. Consequently, this pre-registered, large-scale report was designed to experimentally confirm and extend our knowledge regarding a possible link between clinical urgency and compassion in health care. In line with prior work, our primary expectation was that patient presentations requiring urgent attention would predict greater compassion than presentations that are less urgent.

METHODS

Design

This study was approved by the Auckland Health Research Ethics Committee on the 21st of October 2021 (Approval Number AH23221) and received independent locality approvals from all 20 of the District Health Boards (DHBs) in Aotearoa/New Zealand. The study used a within-person experimental design utilizing patient vignettes that were completed as part of a larger survey-based study on compassion in health care. This report adheres to The Consolidated Standards of Reporting Trials (CONSORT) guidelines for within-person randomized trials (Pandis et al., 2017). In line with best practice, the study design, operationalizations and hypotheses were pre-registered on AsPredicted (Registration number 75407: Hypothesis 3).

Participants and setting

Eligible participants were English-speaking health care professionals (doctors, nurses and allied health professionals), aged over 18 years who were currently practicing, were employed part-time or full-time in Aotearoa/New Zealand and had ‘regular contact with patients as a part of their job’. A total of 1376 people responded to the study advertisement with 1371 consenting. One hundred and twelve participants were excluded from analysis because they did not meet pre-registration eligibility criteria (e.g., did not answer screening eligibility questions, not currently practicing and no clinical patient contact). Of 1259 eligible participants, 950 participants completed the experimental vignette portion of the study and 939 were analysed after dropouts (Figure 1). The sample size was sufficient to predict small effects for main (d = .10 at 90% power for a non-directional, paired t-test with N = 939 and at .05 significance level) as well as exploratory analyses (f2 = .01 at 90% power for multiple regression with N = 939 conducted with three predictors at .05 significance level) calculated via pwr package (Champely et al., 2017).

Details are in the caption following the image
CONSORT 2010 flow diagram.

Procedure

Participants were recruited to a study entitled ‘Institutional barriers to care for kaimahi haoura (healthcare workers)’ via convenience sampling after encountering study advertisements that were distributed via newsletters or organizational employee portals [i.e., DHBs, professional unions, Primary Health Organisation, or Kaupapa Māori or Hauora Māori Organisation (Indigenous health care organizations)]. Additionally, the call to participate was disseminated via health care alumni lists at the University of Auckland and the University of Otago—the two universities that train medical students in Aotearoa. Recruitment took place from February to May 2022. Participants were blind to study hypotheses and the nature of the experiment.

After informed consent was obtained, participants took approximately 20 min to complete an electronic survey. First, the participants completed demographic and occupational measures capturing clinical experience, specialization and the type of organization they worked for. Second, participants were presented with two gender- (She/He/They) and setting-matched (Inpatient/Outpatient) vignettes in a randomly allocated counterbalanced order (cross-over design). Both vignettes depicted a patient with difficult behaviours that have previously been associated with lower compassion (i.e., raising voice, being aggressive and/or demanding; Pavlova et al., 2021; Podrasky & Sexton, 1988; Steinmetz & Tabenkin, 2001; Wang et al., 2022) but vignettes were designed to systematically differ in terms of clinical presentation urgency, where urgency was defined as a situation that was acute and likely require urgent medical attention (clinical picture of a potential stroke or traumatic brain injury) and non-urgent situation as a non-acute chronic situation of a person with mild medically unexplained fatigue and other symptoms (Appendix 1). To maximize face validity, vignettes were constructed in conjunction with clinical experts, including Indigenous Māori clinicians.

To increase the validity of participants' ratings, ‘cognitive loading’ was applied prior to each vignette by asking the participants to memorize a random six-digit number/letter combination (either a National Health Index number or a telephone number extension). Engaging cognitive resources (‘cognitive loading’) is thought to make reasoning less deliberate and more in line with behaviour (Miles, 2015), an important consideration in compassion research which is prone to desirability bias (Baguley et al., 2020; Fernando et al., 2017). Being a part of a larger study, survey data that indexed trait compassion, compassion beliefs, self-efficacy and other psychometric measures were collected after the vignette portion of the study to avoid potential priming. All participants who completed the survey were given a chance to opt-in to a prize draw to win an iPad. Participants who opted in to participate in the draw were redirected to a different survey link to maintain participant anonymity.

Materials and measures

Participants and organizational characteristics

Participants' demographic and occupational data such as gender, age and ethnicity, occupation, years of experience and organizational characteristics such as organizational size (small/medium <250 employees, large >250 employees), funding source (private or public) and setting (community/primary vs. secondary/tertiary; urban vs. rural) were collected.

To collect social desirability data, the Marlowe Crowne Social Desirability Scale short-form version C (MC-C) that is a true or false 13-item measure was administered where higher scores indicate a greater tendency towards providing socially desirable responses (Reynolds, 1982).

Patient vignette manipulation checks

To confirm that patient vignettes systematically varied in the degree of perceived urgency, participants were asked to rate the extent to which each scenario ‘required urgent attention using a 0 (“Not at all”) to 100 (“Extremely”) Visual Analogue Scale (VAS)’. Because ceiling effects are common in compassion measurement and compassion is most challenged by ‘difficult’ patients, we also asked participants ‘how likely they were to consider the patient depicted in the scenario difficult’. The same VAS anchors for perceived urgency were used.

Measures of compassion

A face-valid proxy measure of compassion used in similar study designs (Fernando et al., 2017; Reynolds et al., 2019) was used to capture the two core elements of compassion in health [involving both feeling and motivational components, i.e., the awareness of suffering and a motivation to help (Gilbert, 2019; Jazaieri et al., 2014; Kirby, 2017; Strauss et al., 2016)]. Participants were asked how caring they felt towards the patient depicted in the vignette and how likely they would want to help. A 0–100 VAS scale was used (0 = ‘Not at all’ and 100 = ‘Extremely’). Although as is common, caring and motivation to help were moderately to strongly positively correlated (r = .69–.76, p < .05), both measures of compassion were analysed independently.

Data analysis

All analyses were performed in R. Paired-sample t-tests were used to conduct manipulation checks, as well as testing the main effect of the patient presentation urgency on indices of compassion. Chi-square and t-tests analyses were used to test for balance on sample characteristics by randomization order. The effect of order on manipulations (perceived urgency and difficulty) and primary outcomes (measures of compassion) was analysed by conducting additional t-tests. When order had an effect on manipulations and/or primary outcomes, manipulation checks and main effect analyses were analysed by each randomization subgroup. Non-parametric Wilcoxon signed-rank tests were used to confirm the t-test results when the dependent variables were not normally distributed.

Exploratory analyses were conducted to assess the relationship ratings of compassion had with patient difficulty and urgency, additionally testing for a potential interaction between difficulty and urgency. Linear regression analyses were performed separately for each vignette to avoid multicollinearity. The analyses were performed in two steps with main effects entered before interaction terms. Statistical significance was set at p < .05 for all tests, and 95% confidence intervals (CI) were reported.

As per our pre-registered analytic plan, missing data were either excluded (i.e., where missing variables represented more than 20% of the data) or imputed using group means. Outliers were assessed via Tukey fences (Tukey, 1977); if the data were skewed, medians were used instead of means. Imputed data represented less than 2%. Because data were negatively skewed, we deviated from the pre-registered data management protocol and did not use an ln-transformation. Instead, we ran the analyses on non-transformed data, confirming that test assumptions were met without transformation. When data were skewed, non-parametric tests were used alongside standard parametric tests to check whether results still held. No analyses were run prior to this decision.

RESULTS

Participants

Sample characteristics overall and according to the order of randomization are presented in Table 1. Almost half of the sample were nurses, with doctors and allied health professionals representing 20.1% and 33.0% of the sample respectively. The average years of clinical experience was 17.47 years (SD = 12.82) and the average age was 43 years (SD = 13). Females represented 84.5% of the sample, with 15.0% of participants identifying as male and five participants (.5%) were identified as gender diverse or non-binary. New Zealand Europeans (Pākehā) comprised 56% of the sample while Indigenous Māori health care professionals represented almost 15% of the sample. Most participants worked in large (75.4%), publicly funded (88.6%) organizations located in urban centres (88.5%). One-thirds of the sample worked in primary, and two-thirds in secondary or tertiary care. There were no significant differences in participants’ characteristics according to the order in which vignettes were presented (Table 1).

TABLE 1. Sample characteristics (N = 939).
All observations Randomization order T-test/χ2
Urgent vignette first (N = 474) Non-urgent vignette first (N = 465)
N (%)/M ± SD N (%)/M ± SD N (%)/M ± SD
Socio-demographic variables
Gender
Male 141 (15.0%) 69 (14.6%) 72 (15.5%)
Female 793 (84.5%) 403 (85.0%) 390 (83.9%) χ2 (2) = .39 p = .82
Non-binary 5 (.5%) 2 (.4%) 3 (.6%)
Ethnicity
Pākehā 525 (56.0%) 267 (56.3%) 258 (55.5%)
Māori 138 (14.8%) 73 (15.4%) 65 (14.0%)
Asian 118 (12.6%) 57 (12.0%) 61 (13.1%) χ2 (5) = 3.03, p = .70
Pacific people 23 (2.4%) 8 (1.7%) 15 (3.2%)
MELAA 18 (1.9%) 10 (2.1%) 8 (1.7%)
Other ethnicity 117 (12.5%) 59 (12.5%) 58 (12.5%)
Age 43 ± 13 43 ± 13 44 ± 13 t(936) = −1.93, p = .05
Occupational variables
Occupation
Doctors 189 (20.1%) 94 (19.8%) 95 (20.4%)
Nurses 440 (46.9%) 232 (48.9%) 208 (44.7%) χ2 (2) = 1.86, p = .39
Allied and midwives 310 (33.0%) 148 (31.2%) 162 (34.8%)
Years of experience 17.47 ± 12.82 16.82 ± 12.83 18.14 ± 12.79 t(936) = −1.58, p = .11
Dispositional variables
Social desirability 21.18 ± 2.66 21.13 ± 2.71 21.24 ± 2.61 t(869) = −.60, p = .55
Organizational variables
Organizational size (large) 708 (75.4%) 345 (72.8%) 363 (78.1%) χ2 (1) = 3.25, p = .07
Funding type (public) 831 (88.5%) 416 (87.8%) 415 (89.2%) χ2 (1) = .27, p = .60
Care type (primary) 313 (33.3%) 158 (33.3%) 155 (33.3%) χ2 (1) = .00, p = .99
Urban 830 (88.4%) 417 (88.0%) 413 (88.8%) χ2 (1) = .09, p = .76

Randomization order effects checks

Participants who were presented with the urgent vignette first reported more caring (t(914) = 2.53, p = .01, d = .17) and motivation to help (t(898) = 4.88, p < .001, d = .32) towards the urgent patient and perceived the non-urgent patient as significantly more difficult (t(931) = 4.91, p < .001, d = .32) than those who were presented with the urgent vignette second (Table 2). Non-parametric tests showed similar pattern of results, with the exception of the perceived urgency of the urgent vignette where non-parametric tests indicated a significant effect of the presentation order where higher urgency was perceived when urgent vignette was presented first (V(933) = 119,132, p = .03, r = .07); the effect size was negligible (Appendix 2).

TABLE 2. Randomization order effects checks (N = 939).
Randomization order T-test
Urgent vignette presented first (N = 474) Non-urgent vignette presented first (N = 465)
M ± SD M ± SD
Vignette 1 (urgent)
Perceived urgency 88.64 ± 14.82 86.91 ± 15.49 t(933) = 1.75, p = .08
Perceived difficulty 44.63 ± 28.08 45.13 ± 29.14 t(934) = −.27, p = .79
Caring 82.04 ± 17.32 78.96 ± 19.90 t(914) = 2.53, p = .01
Motivation to help 90.65 ± 13.65 85.78 ± 16.69 t(894) = 4.88, p < .001
Vignette 2 (Non-urgent)
Perceived urgency 63.24 ± 25.11 65.06 ± 24.14 t(937) = −1.13, p = .26
Perceived difficulty 57.07 ± 26.30 48.40 ± 27.83 t(931) = 4.91, p < .001
Caring 70.81 ± 21.91 70.78 ± 21.66 t(937) = .02, p = .98
Motivation to help 75.35 ± 22.54 77.60 ± 20.61 t(932) = −1.59, p = .11

Manipulation checks

Paired-samples t-tests compared urgency and perceived difficulty ratings across the two vignettes (Table 3; Figure 2a). Vignette 1 was rated as significantly more urgent than Vignette 2 with a mean difference of 23.64, 95% CI [21.79; 25.49], (t(938) = 29.53, p < .001, d = .96). However, while the two vignettes were constructed to represent comparably difficult patients, the patient described in Vignette 1 was also seen as less difficult than in Vignette 2. The mean difference of −7.90, 95% CI [−9.57; −6.23] was significant (t(938) = −9.26, p < .001, d = .30). Non-parametric tests (Appendix 2) and subgroup analyses by randomization order (Appendix 3) showed the same pattern of results.

TABLE 3. Mean self-reported care for patients and motivation to help urgent versus non-urgent patients (N = 939).
Vignettes Paired samples t-test
Vignette 1 (urgent) Vignette 2 (non-urgent)
M ± SD
Manipulation check
Perceived urgency 87.78 ± 15.17 64.18 ± 24.58 t(938) = 29.53, p < .001
Perceived difficulty 44.80 ± 28.58 52.71 ± 27.46 t(938) = −9.26, p < .001
Experimental analyses
Caring 80.51 ± 18.70 70.81 ± 21.73 t(938) = 16.03, p < .001
Motivation to help 88.24 ± 15.41 76.46 ± 21.62 t(938) = 18.09, p < .001
Details are in the caption following the image
(a, b) Manipulation checks and experimental analyses barplots.

Experimental analyses

Health care professionals reported greater care (mean difference of 9.57, 95% CI [8.29; 10.84], t(949) = 14.78, p < .001, d = .52) and desire to help (mean difference of 11.58, 95% CI [11.11; 12.05]; t(949) = 48.13, p < .001, d = .59) for the more urgent patient compared to the less urgent patient (Table 3; Figure 2b). Non-parametric tests (Appendix 2) and subgroup analyses by randomization order (Appendix 3) showed the same pattern of results.

Exploratory analyses

Because both perceived urgency (the target variable) and perceived difficulty varied across the two vignettes, additional exploratory analyses were conducted to investigate whether perceived urgency, perceived difficulty or their interaction best-predicted variation in ratings of care and the desire to help.

Vignette 1 (urgent)

Correlation analyses for the urgent patient vignette showed that caring (r = −.31, p < .05) and reported motivation to help (r = −.27, p < .05) were negatively correlated with difficulty ratings. Conversely, caring (r = .47, p < .05) and motivation to help (r = .53, p < .05) were positively associated with ratings of urgency for this patient. Urgency and difficulty were weakly associated (r = −.10, p < .05). In regression analyses (Table 4) for the urgent patient, perceived difficulty predicted lower (f2 = .09; f2 = .06) and perceived urgency greater (f2 = .21; f2 = .28) care and motivation to help, respectively, explaining 28.14% of the variance in care (F(2,936) = 184.7, p < .001) and 33.06% in motivation to help ratings (F(2,936) = 232.6, p < .001). Interactions between difficulty and urgency included in Model 2 were not significant for neither care (p = .49) nor motivation to help outcomes (p = .05; Appendix 4A,B), although for motivation the interaction approached significance (p = .05).

TABLE 4. Model summaries of linear regression analyses assessing the effect of presentation urgency and perceived patient difficulty (Model 1) and urgency and difficulty interaction (Model 2) on compassion as indexed by caring and motivation to help (N = 939).
Urgent patient (vignette 1) Non-urgent patient (vignette 2)
Caring Motivation to help Caring Motivation to help
B (CI) p B (CI) p B (CI) p B (CI) p
Model 1
(Intercept) 40.84 47.66 57.43 58.98
Urgency .54 (.47 to .61) <.001 .52 (.47 to .58) <.001 .41 (.37 to .46) <.001 .45 (.41 to .50) <.001
Difficulty −.17 (−.21 to −.14) <.001 −.12 (−.14 to −.09) <.001 −.25 (−.29 to −.21) <.001 −.22 (−.26 to −.18) <.001
R 2 28.14% 33.06% 36.84% 39.19%
F[2936 df] 184.7 <.001 232.6 <.001 274.5 <.001 303.3 <.001
Model 2
(Intercept) 44.66 42.20 64.8 73.25
Urgency .50 (.36 to .63) <.001 .43 (.32 to .54) <.001 .31 (.21 to .40) <.001 .25 (.16 to .34) <.001
Difficulty −.25 (−.48 to −.02) .03 −.29 (−.48 to −.11) <.001 −.38 (−.11 to −.10) <.001 −.47 (−.57 to −.36) <.001
Difficulty*Urgency .001 (−.002 to .003) .49 .002 (−.000 to .004) .05 .002 (.000 to .003) .01 .004 (.002 to .005) <.001
R 2 28.10% 33.25% 37.22% 40.82%
F[2935 df] 123.2 <.001 156.8 <.001 186.4 <.001 216.6 <.001

Vignette 2 (non-urgent)

Correlation analyses for less urgent patient vignette showed that caring (r = −.40, p < .05) and motivation to help (r = −.37, p < .05) were weakly negatively correlated with difficulty ratings. Conversely, caring (r = .52, p < .05) and motivation to help (r = .56, p < .05) were moderately positively associated with greater urgency. Again, urgency and difficulty were weakly associated (r = −.17, p < .05).

As in regression analyses of the more urgent patient ratings, greater perceived difficulty predicted lower (f2 = .13; f2 = .11) and perceived urgency greater (f2 = .25; f2 = .30) care and motivation to help, respectively, explaining 36.84% of variance (F(2,936) = 274.5, p < .001) for caring and 39.19% for motivation to help (F(2,936) = 303.3, p < .001; Table 4). However, the interaction terms analyses of the less urgent patient were also significant in the prediction of both care and desire to help (p < .001; Appendix 4C,D), improving both models' fit [caring outcome: R2Δ = .4%, (1935) = 6.67, p < .01; motivation to help outcome: (R2Δ = 1.6%, (1935) = 26.67, p < .001)].

DISCUSSION

Findings and implications

This pre-registered study experimentally demonstrated a relationship between a greater urgency of patient presentation and reports of greater compassion from health care professionals, extending and confirming prior qualitative and cross-sectional work (Pavlova et al., 2021). However, while the two vignette scenarios were designed to depict comparably difficult patients, our findings suggest that more urgent patient presentations were also perceived as less difficult. Exploratory analyses showed that greater patient difficulty was associated with lower compassion, particularly for the less urgent patient, in line with prior studies (Derksen et al., 2016, 2018; Fernando et al., 2016; Sandikci et al., 2017; Wang et al., 2022).

In understanding the present findings, we must first consider why urgency appears to lead to greater compassion. Having evolved as a response to suffering, compassion incorporates elements of both noticing suffering together with the motivation (and action) to alleviate it (Gilbert, 2019; Jazaieri et al., 2014; Kirby, 2017; Sinclair et al., 2016; Strauss et al., 2016). According to Gilbert's (2020) evolutionary perspective, the compassion system operates according to a ‘stimulus – response’ algorithm where the first part of the process entails detecting a signal for distress/need before being motivated to alleviate it (e.g., rescue, feed, thermoregulate). Analogously, in clinical scenarios, urgent suffering or distress may be more immediately apparent creating the initial cue necessary for compassion to emerge. Conversely, it is possible that in situations where the distress/need is less overtly salient (and numerous competing demands for clinician attention exist), compassion systems are not as readily activated.

Considering this possible explanation more fully, similar evolutionary reasoning might help to explain our finding about an association between urgency and perceived patient difficulty. Although we cannot be sure, it is possible that urgent patients might be seen as less ‘difficult’ because urgency offers very clear targets regarding clinical management and professional functioning, simplifying the decisional space. As such, when urgency requires a health care professional to act quickly, there might be less ‘mental space’ for incorporation of secondary information and, potentially, fewer explicit judgements that could result in patients being experienced as less ‘difficult’ and, hence, facilitate compassion (Marewski et al., 2010).

Conversely, where patient presentations are not urgent, difficulty might act as a proxy for clinical complexity where there are more factors that need to be taken into account for clinical reasoning which may also entail judgement that might or might not be facilitative of compassion (Bayne et al., 2013; Sandikci et al., 2017; Zandbelt et al., 2007). Relatedly, having less obvious signals of suffering but still having a duty of care (American Medical Association, 2016; New Zealand Medical Association, 2020; NHS, 2022) could result in health care professionals perceiving patients as more ‘difficult’ and engaging in a competing motivational system (i.e., either ‘threat – fight or flight’ response; Gilbert, 2020). Studies show that health care professionals often experience anxiety when managing complex and possibly ineffective treatments, or having a risk of misdiagnosing someone or appearing incompetent in front of colleagues (Sutherland & Cooper, 1992; Tallentire et al., 2017). Consequently, this might trigger ‘fight or flight’ response (Gilbert, 2020) and maladaptive coping (i.e., avoidance, detachment and other blaming; Davidsen & Fosgerau, 2014; Peng et al., 2018), explaining why both patients who are perceived as non-urgent (Batley et al., 2016; Bayne et al., 2013; Bishop et al., 2014; Zandbelt et al., 2007) and those perceived as more ‘difficult’ often elicit lower compassion response (Derksen et al., 2016, 2018; Fernando et al., 2016; Sandikci et al., 2017), and why non-urgent patients are often pejoratively labelled (e.g., ‘attention-seekers’ and ‘time-wasters’; Batley et al., 2016; Bayne et al., 2013; Sandikci et al., 2017; Stein, 1986). Depending on a particular context, such a dynamic may then be suggestive of compassion operating via different behavioural mechanisms, a possibility we give greater attention to below.

Specifically, drawing on cognitive neuroscience reinforcement learning and decision-making framework (Gęsiarz & Crockett, 2015), it has been suggested that the stimulus–response–outcome contingency may operate via three types of decision-making systems: reflexive, habitual and goal directed or deliberative. In this light, it may then mean that feeling caring and being motivated to help patients who are more urgent is more reflexive or automatic but that compassion for less urgent patients is more impacted by habitual and goal-directed systems. This tendency would not have been problematic in the absence of systemic constraints (e.g., time pressures, strict clinical protocols and organizational focus on efficiency), especially considering health care professionals shared motivation and, hence, choice and commitment to be of benefit and provide compassion (Ratanawongsa et al., 2006; Wu et al., 2015). However, as health care organizations increasingly put emphasis on efficiency and speed (Ahrweiler et al., 2014; Anandarajah & Roseman, 2014; Baker et al., 2018; Kerasidou, 2019; Kerasidou & Kingori, 2019; Peng et al., 2018; Rawal et al., 2020; Roze des Ordons et al., 2020) and when compassion as an outcome is not organizationally recognized or acknowledged (Molinsky et al., 2012; Singh et al., 2018), health care professionals might deprioritize compassion for those patients perceived as less urgent. In addition, personal versus organizational goals conflict (compassion vs. efficiency; Pavlova et al., 2023) might then be projected back onto patients, where being less urgent is also perceived as being more difficult.

As noted, variations in compassion based on variation in clinical presentation contradict the very notion of equitable care. This is highly problematic considering that less urgent presentations are potentially more likely to be a result of chronic and preventable illnesses that are also more common among disadvantaged groups (Bachmann et al., 2003; Muenchberger & Kendall, 2010). Not less problematic is that seemingly ‘less urgent’ presentations can be resultant of a more stoic, secretive or shy patient demeanour. As evidenced by studies on evaluations of seriousness of abdominal pain (Kamin et al., 2003), predictors for adequacy of pain analgesia (Rupp & Delaney, 2004) and even in diagnosing cancer (Licqurish et al., 2017; Shahid et al., 2016), patient behaviours vary and low perceived urgency might not always mean lower acuity and need for care. Again, this dynamic is especially profound in minoritized patients. Chronic disease in general (Vogeli et al., 2007) as well as health inequities in particular (Paine et al., 2023; Reid et al., 2022) are incredibly costly to health care and patients' experiences of compassion matter at all parts of the patient journey, being linked to improved psychological coping, better treatment adherence, faster recovery times and fewer complications, among others (Fogarty et al., 1999; Kelley et al., 2014; Trzeciak et al., 2017). Consequently, compassion for presentations that are less urgent is important and just as necessary to ensure better outcomes and minimize the extent to which health care services become an ‘ambulance at the bottom of the cliff’ (Muenchberger & Kendall, 2010).

Strengths and limitations

To the best of our knowledge, this is the first study to experimentally test possible causal links between urgency and compassion. Other strengths include: our consistency/alignment with best reporting practice (i.e., pre-registration of study hypotheses, operationalizations and analytic plans); the large and professionally/demographically diverse health care sample which ensured sufficient power to identify small statistical effects; and use of a ‘cognitive load’ strategy—a technique that reduces the bias associated with more deliberate thinking pathways (Miles, 2015)—to reduce desirability and encourage habitual/reflexive responses.

However, the study is not without its limitations. First, it was conducted using clinical vignettes, rather than by observing behaviours in naturalistic settings or via patient ratings (Baguley et al., 2022; Sinclair et al., 2022). While evidence suggests that participants respond to hypothetical scenarios in a manner similar to how they respond in ‘real life’ and has advantages in terms of standardization (Evans et al., 2015), this limitation remains. Second, the predictors of prosocial experience versus behaviour are also likely different (e.g., Fernando et al., 2017). While behavioural measures are design expensive, analogues can be created in laboratories and with virtual reality being a promising tool. As in other recent studies of medical compassion (Baguley et al., 2020; Fernando et al., 2017), ratings of care and desire to help were consistently high, perhaps suggesting that desirability bias remains an issue. Although this possibility cannot be ignored, such effects are perhaps less relevant here given the experimental and within-person aspects of the design in which desirability effects should be distributed across groups and stimuli. Nonetheless, the predictors of compassionate experiences in physicians may not be the same as the predictors of compassionate behaviour, and incorporation of behavioural measurement is an obvious next step for studies in the health care context.

Perhaps more importantly, our interpretations are necessarily limited by the fact that less urgent patient presentations were unexpectedly considered more difficult in this study. Although exploratory analyses seeking to clarify whether difficulty of presentation predicted lower care were conducted, only urgency was experimentally manipulated meaning that patient difficulty may well be conflated with other factors (e.g., complexity). Additionally, while order effects were present, we believe they were unlikely to affect our results directly, and that this unexpected finding can help inform subsequent vignette studies in compassion research.

Finally, the study vignettes presented scenarios that were socio-demographically neutral. Therefore, in interpreting the findings, it is important to note that these results might not hold for patients from minoritized groups. While urgency might facilitate compassion generally, it may not account for differences in the quality of care received by diverse groups of patients or fully explain ethnic/racial, gender and other inequities in care as, in accordance with prior literature, more automatic/reflexive decision-making processes might also exacerbate implicit bias (Dehon et al., 2017; Guedj et al., 2021). Follow-up studies should be conducted to understand whether urgency–compassion dynamics might be affected by various patients' characteristics that have been previously associated with inequities in care.

Future research

Based on the findings of this study, there are a number of potentially fruitful avenues for future research. For instance, future research could explore the mechanisms behind the urgency effect on compassion (e.g., absence or presence of distress cues and conflicting motivational systems), the factors that result in the switching in motivational systems (e.g., being under time pressure, organizational contingencies and bias) and/or from our unexpected findings that order of patient presentation matters—whether certain motivational systems (e.g., motivation to help vs. ‘fight or flight’) are more ‘sticky’ than others.

Interventionally, future research could test whether making health care professionals aware of urgency/difficulty relationships discovered in this study could lead to changes in compassion as an outcome. Furthermore, in relation to the finding of lower compassion for patients who are less urgent, the facilitative effect of intention and action-planning interventions that would engage more goal-directed decision-making systems could be explored. Similarly, priming might be used to activate compassion for less urgent patients. In one recent study, for example, images that would appear to prime compassion facilitated ratings of care and desire to help challenging patients (Reynolds et al., 2019).

Conclusions

Prior studies have suggested that less urgent patients may receive less compassion (Batley et al., 2016; Bayne et al., 2013; Bishop et al., 2014; Zandbelt et al., 2007). This large-scale, pre-registered experimental study has shown that the problem is real—urgency of presentation does seem to causally impact compassion. However, the origins of the process and the possible association between urgency and perceptions of patient difficulty remain unclear. It must be acknowledged that compassion is an incredibly important but complex multi-system phenomenon requiring a more nuanced thinking in generating ideas of how best to facilitate it. Clinicians may well report less care for less urgent patients, but they do so in the context of complex situational and organizational contingencies that likely exert a profound impact on behaviour. While prior research in medical compassion tended to focus on individuals (Pavlova et al., 2021; Wang et al., 2022), to create the knowledge needed to facilitate multi-level compassion interventions designs, we must study compassion as a complex system (Fernando & Consedine, 2014)—looking at deeper psychological mechanisms of compassionate response while also taking into consideration broader environment.

AUTHOR CONTRIBUTIONS

Alina Pavlova: Conceptualization; investigation; writing – original draft; methodology; visualization; formal analysis; project administration. Sarah-Jane Paine: Investigation; conceptualization; validation; writing – review and editing; supervision; data curation; resources; project administration. Alana Cavadino: Software; formal analysis; writing – review and editing; validation; methodology; resources. Anne O'Callaghan: Conceptualization; investigation; writing – review and editing; project administration; resources. Nathan Consedine S: Conceptualization; investigation; methodology; writing – review and editing; formal analysis; project administration; data curation; supervision; resources.

ACKNOWLEDGEMENTS

We thank Greg West (Kai Tahu, Kati Mamoe) and Ariana Sutton (Kai Tahu, Kati Mamoe and Waitaha oku Iwi) who have provided their clinical expertise in designing this study; Greg West and Ariana Sutton were also our cultural advisors who helped to align this study to reflect Māori worldview and values. In addition, we would like to thank FMHS Responsiveness to Māori team at Waipapa Taumata Rau (The University of Auckland)—Dr Kimiora Henare (Te Rarawa, Te Aupōuri) and Dr Karen Wright (Ngāi Tahu)—who provided guidance with regards to the upholding to the Te Tiriti O Waitangi (The Treaty of Waitangi). We additionally thank Kiralee Schache and Amelie Tuato’o, who have helped with the study recruitment. Finally, we thank all clinicians who took part in this study as well as organizational managers who have helped to distribute this study. This study is a part of a PhD funded by the Margaret Burland Scholarship at the University of Auckland. The funder was not involved in the design and conduct of the study; collection, analysis and interpretation of data, writing of the report or decision to submit the article for publication. Open access publishing facilitated by The University of Auckland, as part of the Wiley - The University of Auckland agreement via the Council of Australian University Librarians.

    CONFLICT OF INTEREST STATEMENT

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

    APPENDIX 1

    CLINICAL VIGNETTES (FEMALE VERSION)

    Difficult patient with more urgent presentation

    Inpatient care

    Whilst you are in a corridor of a hospital ward, you come across a middle-aged woman in a hospital gown who appears unsteady on her feet, is holding her head and muttering. You ask her if she is ok. She says enough to convey she has a sudden severe headache but then starts rambling about seeing her deceased mother and appears confused. You ask her to repeat what she has said, and she starts to shout and swear. She calls you incompetent and kicks the trolley nearby, knocking things off. By now, a couple of other visitors have gathered around.

    Outpatient care

    While walking through the waiting room, you come across a middle-aged woman, who appears unsteady on her feet, is holding her head and muttering. You ask her if she is ok. She says enough to convey she has a sudden severe headache but then starts rambling about seeing her deceased mother and appears confused. You ask her to repeat what she has said, and she starts to shout and swear. She calls you incompetent and kicks the table nearby, knocking things off. By now a couple of other visitors have gathered around.

    Difficult patient with less urgent presentation

    Inpatient care

    You are on the way to a coffee bar in the outpatient clinic to take a few minutes out of your busy day. A woman walks up to you and starts talking rapidly, with hardly a break between words. She tells you that she has come to the clinic because she is always tired and easily gets short of breath. Despite your attempts to convey that you are about to leave and that someone else will come shortly to help her, she persists telling you that she is constantly fatigued, has difficulties concentrating and staying awake and that it gets worse when reading or watching a film. She says she sees stars and floaters especially when she is under stress and a number of tests have already been done to rule out something serious, but nobody is really trying to help her. She says that it is your job to take care of her right now and that you cannot leave her here like this. She grows more agitated and is talking very loudly; people are starting to look.

    Outpatient care

    You are on the way to a coffee bar across the road to have a short break from your busy day. A woman walks up to you as you pass through the waiting room and starts talking rapidly, with hardly a break between words. She tells you that she is always tired and easily gets short of breath. Despite your attempts to convey you are about to leave and that someone else will come shortly to help her, she persists telling you that she is constantly fatigued, has difficulties concentrating and staying awake, and it gets worse when reading or watching a film. She says she sees stars and floaters especially when she is under stress and a number of tests have already been done to rule out something serious, but nobody is really trying to help her. She says that it is your job to take care of her right now and that you cannot leave her here like this. She grows more agitated and is talking very loudly; people are starting to look.

    APPENDIX 2

    Randomization order non-parametric tests

    Randomisation order Wilcoxon rank sum test
    Urgent vignette presented first (N = 474) Non-urgent vignette presented first (N = 465)
    M ± SD M ± SD
    Vignette 1 (urgent)
    Perceived urgency 88.64 ± 14.82 86.91 ± 15.49 V(933) = 119,132, p = .03, r = .07
    Perceived difficulty 44.63 ± 28.08 45.13 ± 29.14 V(934) = 109,526, p = .87
    Caring 82.04 ± 17.32 78.96 ± 19.90 V(914) = 118,876, p = .04, r = .07
    Motivation to help 90.65 ± 13.65 85.78 ± 16.69 V(894) = 130,728, p < .001, r = .17
    Vignette 2 (non-urgent)
    Perceived urgency 63.24 ± 25.11 65.06 ± 24.14 V(937) = 105,591, p = .27
    Perceived difficulty 57.07 ± 26.30 48.40 ± 27.83 V(931) = 129,269, p < .001, r = .15
    Caring 70.81 ± 21.91 70.78 ± 21.66 V(937) = 109,221, p = .81
    Motivation to help 75.35 ± 22.54 77.60 ± 20.61 V(932) = 104,857, p = .20

    Manipulation and experimental non-parametric tests

    Vignettes Wilcoxon rank sum test
    Vignette 1 (urgent) Vignette 2 (non-urgent)
    M ± SD
    Manipulation check
    Perceived urgency 87.78 ± 15.17 64.18 ± 24.58 V(938) = 333,507, p < .001, r = .75
    Perceived difficulty 44.80 ± 28.58 52.71 ± 27.46 V(938) = 109,482, p < .001, r = .30
    Experimental analyses
    Caring 80.51 ± 18.70 70.81 ± 21.73 V(938) = 253,943, p < .001, r = .50
    Motivation to help 88.24 ± 15.41 76.46 ± 21.62 t(938) = 241,134, p < .001, r = .57

    APPENDIX 3

    Subgroup analyses of manipulation checks and experimental tests by randomization order

    Urgent vignette presented first

    Vignettes Paired samples t-test
    Vignette 1 (urgent) Vignette 2 (non-urgent)
    M ± SD
    Manipulation check
    Perceived urgency 88.63 ± 14.82 63.24 ± 25.11 t(473) = 21.67, p < .001
    Perceived difficulty 44.63 ± 28.08 57.07 ± 26.29 t(473) = −11.76, p < .001
    Experimental analyses
    Caring 82.04 ± 17.32 70.81 ± 21.91 t(473) = 13.24, p < .001
    Motivation to help 90.65 ± 13.65 75.35 ± 22.54 t(473) = 16.07, p < .001

    Non-urgent vignette presented first

    Vignettes Paired samples t-test
    Vignette 1 (urgent) Vignette 2 (non-urgent)
    M ± SD
    Manipulation check
    Perceived urgency 86.91 ± 15.49 65.06 ± 24.14 t(464) = 20.15, p < .001
    Perceived difficulty 45.13 ± 29.14 48.40 ± 27.83 t(464) = −2.49, p = .01
    Experimental analyses
    Caring 78.96 ± 19.01 70.78 ± 21.66 t(464) = 9.49, p < .001
    Motivation to help 85.78 ± 16.69 77.60 ± 20.61 t(464) = 9.56, p < .001

    APPENDIX 4

    Linear regression slopes representing interaction effects of perceived patient difficulty and urgency on compassion (as indexed by caring and motivation to help; N = 939).

    DATA AVAILABILITY STATEMENT

    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.