Abstract
Quality of life (QoL) is an important outcome in aged care, but self-report is not always possible due to the high prevalence of cognitive impairment in older aged care residents. This study aims to assess the impact of family member proxy perspective (proxy-proxy or proxy-person) on interrater agreement with resident self-report by different cognition levels. The influence of proxy perspective and cognition level is a significant gap in the extant literature which this study seeks to address. A cross-sectional study was undertaken with residents classified into cognition subgroups according to the Mini Mental State Examination. Residents completed the self-report EQ-5D-5L, a well-established generic measure of health-related quality of life (HRQoL). Family member proxies completed EQ-5D-5L proxy version 1 (proxy-proxy perspective, where the proxy responds based on their own opinions) and proxy version 2 (proxy-person perspective, where the proxy responds as they believe the person would). Interrater agreement was assessed using the concordance correlation coefficient (CCC) for utility scores and the weighted kappa for dimension-level responses. Sixty-three residents (n = 22 no cognitive impairment, n = 27 mild impairment, and n = 14 moderate impairment) and proxies participated. EQ-5D-5L utility scores were lower for proxies compared with residents (self-report = 0.522, proxy-proxy = 0.299, and proxy-person = 0.408). Interrater agreement with self-report was higher for proxy-person (CCC = 0.691) than for proxy-proxy (CCC = 0.609). Agreement at the dimension level was higher for more easily observable dimensions, such as mobility, compared to less observable dimensions, such as anxiety/depression. Resident self-reported and proxy family member-reported HRQoL assessments, using the EQ-5D-5L, are different but may be more closely aligned when the proxy is specifically guided to respond from the person’s perspective. Further research is needed to address the impact of divergences in self-report and proxy-report ratings of HRQoL for quality assessment and economic evaluation in aged care.
1. Introduction
Quality of life (QoL) is an important outcome, and arguably, the most important quality indicator in aged care [1]. A recent Royal Commission into Aged Care Quality and Safety in Australia was critical of the quality of care received for older adults, particularly in residential care settings [2, 3], and asserted the need for routine measurement of QoL as an important person-centred indicator of care quality [4]. As well as being applicable for quality assessment in aged care, preference-based QoL measures are often used as an important measure of benefit to assess QoL impacts for economic evaluation [1]. To date, economic evaluation has been conducted more extensively in healthcare settings than aged care settings [5, 6]. However, economic evaluation has much to offer in aged care, providing a systematic framework for assisting policy makers charged with ensuring that the funding allocated to the aged care sector is allocated efficiently to maximise the QoL of older adults [7].
The EQ-5D-3L [8] and EQ-5D-5L [9] instruments are well established and the most widely used preference-based instruments for economic evaluation [10]. These instruments were developed to describe and value HRQoL for a wide range of populations. The instruments cover five dimensions of health (mobility, self-care, usual activities, pain/discomfort, and anxiety/depression), with a three-level response classification (no problems, some problems, and extreme problems/unable) or a five-level response classification (no problems, slight problems, moderate problems, severe problems, and extreme problems/unable). Both instruments also have a visual analogue scale (EQ VAS), which measures health on a scale from 0 (“worst imaginable health state”) to 100 (“best imaginable health state”). The use of the EQ-5D-3L is mandated in the UK for economic evaluations in healthcare settings [11]. The EQ-5D-5L was introduced by the EuroQol Group in 2009, with a view to reducing ceiling effects and improving sensitivity relative to the EQ-5D-3L; both instruments are available in over 150 languages [12]. Country-specific value sets are available for EQ-5D instruments, providing modelled representations of utility scores for the estimation of quality-adjusted life years (QALYs) for use in the economic evaluation. Utility scores are interpreted on a 0-1 scale, where 1 represents “full health” and 0 represents “dead” (negative scores reflect states worse than dead) [13].
Where possible, self-report rather than proxy-report of QoL is preferred [14]. However, this may not always be possible when participants have cognitive impairment or dementia, as is common in aged care settings. It is estimated that 54% of people living in Australian residential aged care facilities have dementia [15]. When older adults are unable to self-complete (e.g., due to severe cognitive impairment and dementia), proxies can be used to assess their QoL. In common with other HRQoL instruments, the EQ-5D-5L has proxy versions. Research into the use of proxy respondents for outcomes’ measurement comprises two different perspectives, the proxy can consider: (i) the “proxy-proxy” perspective, where the proxy is asked to think about their own perspective of the person’s HRQoL and (ii) the “proxy-person” perspective, where the proxy is asked to respond in the way they think the person would respond if they could self-complete. The EQ-5D-5L has two proxy versions available reflecting these two perspectives.
A recent systematic review identified 50 peer-reviewed articles that compared self-reported and proxy-reported responses to QoL instruments in older adult populations, where preference-based scoring algorithms were available [16]. Dementia-specific QoL instruments, such as the QOL-AD (Quality of Life—Alzheimer’s Disease), were the most commonly used instruments; the EQ-5D-3L and EQ-5D-5L were the most widely used of the generic instruments. The review identified studies that included older adults with some degree of cognitive impairment as well as studies that compared dyad responses from adults with and without cognitive impairment with proxy assessments. Notably, the adopted proxy perspective was reported in only 56% of the studies [16]. The authors found a predominantly unidirectional discrepancy between self-reports and proxy-reports, such that proxies provided lower ratings of older adults’ QoL. There was also evidence that physical dimensions, such as mobility, showed higher levels of agreement than psychosocial dimensions, such as anxiety/depression [16].
The pattern of family member proxy ratings being lower than self-report has been demonstrated using a variety of QoL instruments, including generic adult HRQoL instruments such as the EQ-5D-5L [17, 18], older person-specific instruments such as the ICECAP-O [19], and dementia-specific QoL instruments such as the QOL-AD [20, 21]. The interrater gap between self-report and proxy-report has also been shown to widen as dementia advances [19, 20]. This has been argued to result from lower levels of self-awareness of the person as dementia advances [22].
Two of the 50 studies identified by Hutchinson and colleagues compared agreement statistics derived from analyses using proxy-proxy and proxy-person perspectives [23, 24]. The studies (both using EQ-5D-3L and where proxies were paid carers) found that agreement statistics for utility scores and the EQ VAS were higher when proxies adopted the proxy-person perspective (where the proxy responds as they believe the person would). It is important to assess the impact of proxy perspective using family members as proxies because the systematic review identified this as the most common proxy relationship for older adults [16]. Furthermore, if QoL is to be routinely assessed—as per the [4] recommendations—it is most likely that in instances where proxy assessments are sought, family members will be asked to assess QoL, as the intention is for these data to feed into quality ratings of facilities to support consumer choice, and assessments by paid carers could be considered biased.
To address the research gap on the impact of proxy perspectives between resident self-report and family member proxy-report, this study aims to measure the level of agreement in EQ-5D-5L responses (utility scores and dimension-level responses) between older adults in residential aged care and their family member proxies. To reflect the likely spread of cognition levels that are usually present in residential aged care settings, residents across three levels of cognition (no cognitive impairment, mild cognitive impairment, and moderate cognitive impairment) were included.
2. Materials and Methods
2.1. Study Design and Procedures
This is a cross-sectional study. HRQoL was assessed using the EQ-5D-5L self-report version for residents and both EQ-5D-5L proxy versions (proxy-proxy and proxy-person) for the respective family member. Data collection with residents was via an interviewer-administered survey comprising a cognitive assessment, the EQ-5D-5L, and sociodemographic questions (“Materials”). Family members could choose between completions of an online survey or answer questions over the telephone. Proxy surveys comprised the two EQ-5D-5 L proxy versions and sociodemographic questions (“Materials”). Ethics approval for the study was obtained from Flinders University Human Research Ethics Committee (Ref. number: 4229).
Aged care providers approached eligible residents to assess their initial interest in participating, provided participant information sheets to those expressing interest, and followed up with the resident within a period of one week. A list of residents was provided to the research team once verbal consent had been obtained. A member of the research team then contacted the facility to agree a suitable date to conduct data collection. On the day of the facility visit, potential participants had the opportunity to ask questions about the study, and if still willing to participate, we provided formal written consent before completing the survey with the interviewer.
Family members were initially provided an information sheet about the study by residential aged care staff. Family members who expressed an interest in participation were then contacted by a member of the research team to discuss the study and answer any questions they might have. If family members indicated their willingness to participate, they were provided with a participant information sheet via their preferred delivery method (e-mail or post). Data from proxies were collected within seven days of data being collected from residents.
Data were collected by three trained interviewers. All members of the research team who were involved in data collection had police checks, previous experience of interviewing older adults, and observed all COVID-19 regulations and entry requirements at the respective facility. Data were collected between June and December 2021. During this period, South Australian aged care facilities had strict entry protocols in place due to the COVID-19 pandemic but no significant restrictions.
2.2. Participants
Residents living at one of 10 aged care facilities in metropolitan and rural locations in South Australia were recruited with support from the South Australian Innovation Hub, a community of practice partnership of eight aged care providers that aims to support the creation and implementation of evidence-based innovation to improve the QoL of residents. To participate, residents needed to be 65 years or over, permanently resident in aged care, and able to communicate in English. Family member proxies were identified by residential aged care staff as someone who visited the resident regularly. The aim was to recruit at least 60 dyads (i.e., 60 residents, each with one proxy) for the study to achieve representation across cognition levels and to facilitate robust statistical analysis of interrater agreement [25].
2.3. Materials
The resident survey consisted of sociodemographic questions (gender, age, country of birth, level of education, and how long they had been resident at their facility), the Mini Mental State Examination (MMSE) [26], and the Australian self-report version of the EQ-5D-5L. The MMSE, which has a scoring range from 0 to 30, was used to assess cognition. In accordance with the published guidelines [27], the following classifications were used: 27 to 30, no cognitive impairment; 21 to 26, mild cognitive impairment; and 10 to 20, moderate cognitive impairment. Participants scoring less than 10—classified as having severe cognitive impairment—were excluded from the study. The five dimensions of the Australian EQ-5D-5L are mobility, personal care (as opposed to “self-care,” which is used in other English-language versions of the EQ-5D-5L), usual activities, pain/discomfort, and anxiety/depression, each with five response options (ranging from “no problems” to “extreme problems”/“unable to”). When answering the dimension-level questions and the EQ VAS, respondents are asked to select answers with regard to their health “today.” Utility scores for individual-level responses to the five dimensions were calculated using an Australian general population-specific value set [28].
The proxy survey included sociodemographic questions (gender, age, country of birth, level of education, and questions about the frequency of visits and phone calls with the resident) and the proxy-proxy and proxy-person versions of the EQ-5D-5L. The dimensions, response categories, and the EQ VAS are the same in the proxy versions as for the self-report version. The only difference is in the instructions regarding perspective. The proxy versions were administered in a fixed order (proxy-proxy, then proxy-person).
2.4. Analysis
Descriptive analyses were performed on sociodemographic data. EQ-5D-5L utility scores were calculated and descriptive analyses performed for EQ-5D-5L utility scores and EQ VAS scores, with the mean, standard deviation, median, and 25th and 75th percentiles calculated. Within each of the three rater groups (self-report, proxy-proxy, and proxy-person), the Kruskal–Wallis rank sum test was performed to test for statistically significant differences in utility scores and EQ VAS scores between cognition subgroups. Kruskal–Wallis rank sum tests were also used to test for statistically significant differences in utility scores and EQ VAS scores between the rater groups, within cognition subgroups. In circumstances where the null hypothesis was rejected for the Kruskal–Wallis rank sum test (i.e., there is a significant difference in measurements (utility scores or EQ VAS scores) between the groups (cognition subgroups or rater groups)), the Dunn test was used for pairwise comparisons.
Agreement between combinations of rater pairs (self-report and proxy-proxy, and self-report and proxy-person) for utility scores and EQ VAS scores was assessed using (i) Lin’s concordance correlation coefficient (CCC) [25] and (ii) Bland–Altman plots [29]. The CCC is an appropriate statistic to measure agreement when the distribution of the data is not normal. Bland–Altman plots were used to further explore the relationships between pairwise ratings (self-report and proxy-proxy, and self-report and proxy-person) for utility scores and EQ VAS scores, providing an illustration of the level of agreement between raters’ scores across the respective scoring ranges. Briefly, the Bland–Altman plot is a plot of the difference between two sets of scores (y-axis) and their mean (x-axis). To aid interpretation, the plot includes lines representing the mean difference (here, a blue line) and the “limits of agreement” (here, yellow lines), calculated as the mean difference ± 1.96 standard deviation of the difference. Good agreement between the scores would show a mean difference near to zero, with approximately 5% of scatter points lying outside the limits of agreement. An equal spacing weighted kappa was used to measure the level of agreement amongst raters for the dimension-level responses [30]. Kappa statistics were interpreted using the following guidelines: 0.01–0.20 (slight agreement), 0.21–0.40 (fair agreement), 0.41–0.60 (moderate agreement), 0.61–0.80 (substantial agreement), and 0.81–1.00 (almost perfect or perfect agreement) [31].
Analyses were conducted in R version 4.2.0 [31]; tables were produced using the packages ggplot2 [33], DescTools [34], and gmodels [35].
3. Results
3.1. Participant Characteristics
Sixty-three dyads participated in the study. Of the residents, 22 (34.9%) were assessed as having no cognitive impairment, 27 (42.9%) mild cognitive impairment, and 14 (22.2%) as moderate cognitive impairment. Forty-one residents (65.1%) were female, 16 (25.4%) had tertiary qualifications, and the mean age was 87.6 years. The majority were born in Australia (76.2%) or the United Kingdom (UK) (15.9%), and 68.3% had been resident at their aged care facility for at least one year (Table 1).
The proxy family member samples were predominantly female (79.4%). The most common relationship to the resident was daughter/son (46.0%) or daughter-/son-in-law (22.2%). Proxies had a mean age of 66.5 years, and 54.0% were retired. As with residents, proxies were mostly Australian or from the UK (92.1%). Proxies reported a higher level of education than residents, with over half having tertiary qualifications. Almost a third reported that they spoke to their relative most days of the week, if not daily; 74.6% reported that they spoke to their relative at least once per week (Table 1).
3.2. EQ-5D-5L Dimension-Level Responses and Utility Scores
Figure 1 shows the distribution of EQ-5D-5L dimension-level responses by the rater group (self-report and the two proxies). Irrespective of the perspective adopted, proxies reported higher levels of “any impairment” (i.e., responses across levels 2 to 5) across all dimensions compared with residents’ self-reported ratings.
Descriptive statistics are provided in Table 2. The mean utility score was lowest for the proxy-proxy instrument (0.299), followed by proxy-person (0.408) and self-report (0.522). EQ VAS mean scores were lower for the proxy-reported instruments (60.9 for both) compared with self-report (74.6). The highest self-reported utility score was in the moderate cognitive impairment subgroup (0.583); the lowest was for the subgroup with no cognitive impairment (0.473). For the EQ VAS, residents with moderate cognitive impairment provided the lowest mean score (69.8); the highest mean self-reported EQ VAS score was in the mild cognitive impairment subgroup (77.7). Within each rater group, there were no statistically significant differences in utility scores or EQ VAS scores across cognition subgroups (Table 2). The only statistically significant difference between raters’ utility scores and EQ VAS scores was for residents with mild cognitive impairment (EQ-5D-5L H (2) = 7.22, ; EQ VAS H (2) = 15.14, ) (Table 2). For the utility scores, Dunn’s test revealed a statistically significant difference between the self-report and proxy-proxy pairwise comparison (Z = −2.66, ). For the EQ VAS scores, statistically significant differences were found for the self-report and proxy-proxy comparison (Z = −3.62, ) and the self-report and proxy-person comparison (Z = −3.05, ).
3.3. Interrater Agreement
Bland–Altman plots are presented in Figure 2, depicting agreement between the difference between scores and the average of scores (for utility scores and EQ VAS scores) when comparing self-reported responses to proxy responses. Regarding utility scores (panel a and panel b), the limits of agreement are of similar width. While the mean difference between scores was smaller for the proxy-person perspective compared with the proxy-proxy perspective, there was a higher proportion of markers beyond the limits of agreement (6% compared with 3%). Bland–Altman plots for the EQ VAS scores were similar for the two pairwise ratings (see panels c and d).
Table 3 reports the CCCs for the total sample and by cognition subgroups. For the total sample, agreement between self-reported and proxy-reported utility scores was higher for the proxy-person perspective (0.69) relative to the proxy-proxy perspective (0.61). The same observation was found for agreement between EQ VAS scores, albeit with notably lower correlations (0.24 compared with 0.13). By the cognition subgroup, the highest agreement in utility scores was observed for those with no cognitive impairment (0.70 for self-report/proxy-proxy, 0.78 for self-report/proxy-person), followed by mild impairment, and then moderate impairment. For the EQ VAS, CCC statistics did not exceed 0.33 in any analyses.
At the dimension level, agreement between self-report and proxy assessment was generally higher for residents with no cognitive impairment, with at least “moderate” agreement for mobility, personal care, and usual activities (kappas from 0.44 to 0.69) irrespective of the perspective adopted (Table 3). For the mild impairment subgroup, agreement was “substantial” for mobility (0.65) and personal care (0.61) when proxies adopted the proxy-person perspective. Adopting the proxy-proxy perspective for the mild impairment subgroup resulted in “fair” agreement across all dimensions except for mobility (where agreement was “moderate”). For the moderate impairment subgroup, the highest levels of agreement were for personal care (0.45) and usual activities (0.42), when adopting the proxy-person perspective. Agreement was “slight” in four of the five dimensions (for mobility, agreement was “fair”) for the proxy-proxy perspective.
4. Discussion
This paper adds to the limited evidence base on the impact of proxy perspectives on interrater agreement between the HRQoL ratings of older adults in residential aged care and their proxies. To our knowledge, this is the first time the impact of family members differing proxy perspectives has been reported for residents using the EQ-5D-5L. As identified in a recent systematic review, only two papers have reported on the impact of proxies adopting proxy-proxy and proxy-person perspectives when assessing older adults’ QoL [16]. Both studies used the EQ-5D-3L as the instrument and paid carers, rather than family members, as the proxies (one study took place in aged care facilities [23] and one in a rehabilitation unit [24]). Our findings are in alignment with Leontjevas et al. [23]; in that when proxies adopted the proxy-person perspective rather than the proxy-proxy perspective, higher levels of agreement with self-reported HRQoL resulted. Similar to Leontjevas et al. [23], this study also found higher levels of agreement for EQ-5D-5L utility scores relative to the EQ VAS.
We also observed that agreement was higher for potentially more easily observable dimensions such as mobility, personal care, and usual activities compared to pain/discomfort and anxiety/depression. This trend held irrespective of the proxy perspective adopted and the cognition level of the resident, except for the mild cognition subgroup where adopting the proxy-person perspective for anxiety/depression resulted in moderate agreement. Stronger agreement for physical HRQoL dimensions has been observed in studies using the EQ-5D-3L, particularly in relation to mobility (e.g., references [36, 37]). The potential for observability of dimensions may therefore be a relevant consideration when choosing a QoL instrument that is to be completed by proxies as well as residents.
As with previous studies that have compared self and proxy completion with older adults, our findings show a unidirectional discrepancy between self and proxy ratings, such that residents rate their own QoL more highly than their family members do [16]. This may be accounted for by older adults’ QoL values shifting with age [38, 39]. Studies investigating the gap between older adults’ QoL ratings and those of family members have identified that carer variables such as carer burden [22, 40], depression [41], and their perceptions of their own quality of life [37] can result in lower proxy ratings of QoL. Furthermore, the cognition level of family proxies is usually not assessed, even though proxies can be spouses and adult children who are often older adults themselves [16]. It may be that the cognition level of the proxies influenced the assessments of their relative’s quality of life.
The lowest mean self-reported EQ-5D-5L score was observed in the subgroup without cognitive impairment, while the highest self-reported EQ-5D-5L and EQ VAS scores were in the moderate impairment group (Table 2). Further research in relation to the impact of comorbidities on HRQoL would need to collect clinical data to investigate other differentiating factors between the groups with different levels of cognitive decline.
We also observed that interrater agreement at the dimension level was higher when residents had no cognitive impairment, compared with those with mild or moderate cognitive impairment. This pattern was observed in a previous study [42] when using the EQ-5D-3L with a sample of 390 older adults and 357 family members. Longitudinal studies have observed that differences between self- and proxy-reported QoL increased over time [16], which has been attributed to a lack of awareness on the part of the person with dementia [43]. However, it has also been noted that awareness is complex and influenced by a range of factors that include not only cognitive functioning but also individual psychological responses and traits and other clinical factors such as neuropsychological functioning, sociodemographic factors, and social context [44, 45]. Though self-report is preferable to proxy-report, where possible [14], the use of proxy-report in aged care settings is inevitable for a proportion of the population, e.g., older residents with severe cognitive impairment and dementia.
Finally, the findings of higher correlations between EQ-5D-5L utilities compared to those between EQ VAS scores are an interesting observation. The explanation for these differences is unclear. The team is currently undertaking work using a qualitative think aloud approach with older residents completing the EQ-5D-5L, which may shed further light on this issue. The low correlations between proxy and self-report EQ VAS scores suggest that caution is necessary when reflecting on future data collection initiatives in proxy/self-report situations.
The Australian Government Department of Health, in direct response to the recommendations of the Royal Commission on Aged Care Quality and Safety, is now testing a range of new quality indicators including clinical indicators of care quality and QoL. Given that some older adults receiving aged care services are likely to be able to self-complete QoL assessments, and some may require proxy completion, the question remains as to whether these data can be pooled or whether they need to be considered as two separate sources of data. A recent study attempted to address this issue for the DEMQOL, an instrument developed to assess the health-related QoL of people with dementia. Smith and colleagues [46] identified that older adults with dementia and their family proxies reported on the same underlying QoL construct for the DEMQOL, suggesting data from both groups can be pooled. However, some differences in responses were observed, leading the authors to develop an adjusted scoring algorithm. Statistical adjustments of this nature with other QoL instruments such as the EQ-5D-5L may facilitate the pooling of data collected from older adults and proxies.
Given that self-reporting QoL is preferable, more efforts need to be made by developers to make QoL instruments accessible for respondents with different types of impairment, such as through the use of easy read formats and pictographs to support understanding [47, 48]. Such attempts to improve accessibility would also potentially benefit other populations, such as people from culturally and linguistically diverse (CALD) backgrounds and younger adults with intellectual impairment.
4.1. Limitations
Though the current study has strengths in addressing an important gap in the extant literature on self and proxy completion of QoL instruments in aged care settings, it is important to acknowledge limitations. The study was conducted in (a single state) in Australia and in a limited number of residential aged care facilities. Hence, the residents and their family members included in this study may not be representative of residents and family members across Australia. Our analysis was unable to consider the impact of other sociodemographic variables and/or clinical factors on the EQ-5D-5L and EQ VAS scores and how these may affect comparisons between rater groups. Whilst every effort was made to encourage proxies to complete their assessment on the same day as residents, some took up to seven days to complete their assessments and it is possible that residents’ HRQoL may have changed over this time period. The study only focused on family members as proxies as the aim was to understand the potential impacts of self-report and proxy-report assessment being pooled when the quality indicators’ program rolls out in 2023, and these data will be used to inform “quality ratings” for aged care facilities. However, it would be valuable for future research to investigate the impact of both proxy perspective and proxy type. The study sample was relatively small, particularly in relation to residents with moderate impairment. Finally, unfortunately our research budget did not support the inclusion of translation and interpreting services to engage with older adults and family members from CALD backgrounds, who are estimated to make up 19% of Australia’s residential aged care population [49].
5. Conclusion
The findings from this study indicate that proxy assessments of HRQoL, when using the EQ-5D-5L, are more aligned with residents’ self-reported responses when adopting a proxy-person perspective (i.e., when proxies are asked to respond as they believe the person/resident would). More research is needed in larger and more diverse samples of older people to verify these findings. Addressing the self-report/proxy-report gap in QoL ratings for older people is an important area for further research to support the meaningful inclusion of QoL data in quality indicator programs and for economic evaluations in aged care.
Data Availability
The data used to support the findings of this study are available upon request to the corresponding author.
Conflicts of Interest
Lidia Engel and David GT Whitehurst are members of the EuroQol Group Association. The other authors declare that they have no conflicts of interest.
Authors’ Contributions
Conceptualisation of the study was led by JR with contributions from CH, DW, and LE; JR and CH provided funding; MC, KL, and CH performed data collection; analysis was performed by MC under the supervision of DW; CH drafted the manuscript; CH, DW, MC, KL, LE, and JR revised the manuscript.
Acknowledgments
The authors thank Simon Charlton of the SA Innovation Hub and the Hub members for supporting recruitment for this study and Dr. Jyoti Khadka for guidance on the statistical analysis. The authors also extend their thanks to their Project Advisory Group, project partners, and all older residents and family members who generously gave up their time to participate in this study. This work was supported by the EuroQol Research Foundation (grant no.: 194-2020RA). Open access publishing was facilitated by Flinders University, as part of the Wiley–Flinders University agreement via the Council of Australian University Librarians.