Article Text
PDF
Abstract
Objective The aim of this exploratory study was to investigate the association between health anxiety and self-triage decisions among emergency department non-urgent patients.
Design Cross-sectional single-centre study
Setting Emergency department in the Princess Alexandra Hospital in Brisbane, Australia
Participants Between 13 December 2022 and 30 August 2023, an exhaustive recruitment strategy was deployed to recruit 400 patients. Eligible participants were patients aged 18 years or above who belonged to the Australasian Triage Scale category four or five (non-urgent), were physically and mentally capable of participating in the study and presented to the emergency department between 6:00 a.m. and 23:00 p.m. during the study period.
Main outcome measures The primary outcome was accurately self-triaged decisions, while the secondary outcome was inaccurately self-triaged decisions, including both overtriaged and undertriaged decisions. Self-triage decisions were assessed using six hypothetical medical scenarios.
Results Regression results revealed that health anxiety was not associated with accurately self-triaged decisions. However, compared with non-urgent patients exhibiting no health anxiety, those in the third and fourth quartiles (the upper two quartiles) of the Whiteley Index-6 were expected to make 0·29 (95% CI 0·09 to 0·50) and 0·25 (95% CI 0·07 to 0·44) more overtriaged decisions (mean=0·42; SD=0·71), respectively. In contrast, negative associations between health anxiety and undertriaged decisions were observed. Subgroup analyses by age showed statistically significant associations between health anxiety and inaccurately self-triaged decisions among the older adult patient group (aged 35–79 years). Moreover, analyses stratified by sex revealed that female and male patients in the fourth quartile of the Whiteley Index-6 were expected to make 0·26 (95% CI 0·02 to 0·49) and 0·27 (95% CI 0·05 to 0·48) more overtriaged decisions, respectively, compared with those with no health anxiety.
Conclusions Our results suggest no significant association between health anxiety and accurately self-triaged decisions. In contrast, health anxiety was associated with inaccurately self-triaged decisions. This implies that patients with health anxiety overestimate the need for healthcare and therefore could substantially impact the misuse of health services, particularly emergency departments.
- Health Services
- Health policy
- Economics
Data availability statement
No data are available. The study protocol, survey instrument and STATA do files can be made available upon request. However, the deidentified individual patient data used in this study are the property of the Queensland University of Technology and cannot be made available. Please contact the corresponding author for further information.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
STRENGTHS AND LIMITATIONS OF THIS STUDY
We collected information on patients’ emergency department (ED) visits and the utilisation of other health services. This allowed us to further validate our hypotheses using actual self-reported behavioural proxies.
Our study may be influenced by social desirability bias, where patients might have provided responses that they perceived as socially acceptable.
We excluded patients who did not speak English from our study, which limits the sample to English-speaking non-urgent patients only.
Our study has the influence of seasonality on ED visits, where the inflow of non-urgent patients could exhibit distinct patterns due to weather conditions, influenza seasons, holidays and other seasonal trends. However, we have attempted to address this by adjusting for the time, date, week and month of survey submission.
This study is exploratory, and findings should be interpreted with caution.
Introduction
Self-triage, the initial process by which individuals evaluate their symptoms and determine whether and where to seek care, is crucial for effective healthcare utilisation and improved health outcomes. This process, however, involves risk, uncertainty and ambiguity, since it depends on subjective self-evaluation, intuition and heuristics.1 2 Consequently, these challenges often result in erroneous self-triage decisions, contributing to the inefficient allocation of healthcare resources. For example, the use of emergency departments (EDs) for non-urgent health issues, typically handled by general practitioners (GPs), exemplifies this inefficiency. In Australia, nearly half (48%) of the 8 million ED visits were deemed less urgent,3 with the management of these cases incurring significant costs. Moreover, inappropriately frequenting EDs is a persistent global public health concern.
Numerous factors, including sociodemographic characteristics and behavioural and psychological aspects, influence self-triage decisions. While existing research primarily focuses on developing efficient self-triage tools and evaluating their performance accuracy in directing patients to appropriate care,4–9 these studies often have limited scopes. Specifically, symptom-based, algorithm-assisted tools do not fully integrate behavioural and psychological dimensions and are generally risk-averse. However, a few studies have emerged showing age, gender, and education as significant predictors of self-triage decision accuracy.5 10 Furthermore, in agreement with the gold standard, (self-) triage accuracy was found to be highest among individuals aged 26–50 years.5 Moreover, women, compared with men, exhibited greater risk aversion and were more likely to overtriage.10
In addition, health anxiety, often used interchangeably with hypochondriasis in the literature and defined by excessive fear or belief of having a serious illness, has been identified as a meaningful predictor associated with the use of self-triage tools and healthcare utilisation. A recent cross-sectional study conducted in Germany revealed that hypochondriasis is a significant user characteristic, influencing the utilisation of self-triage tools (OR range: 1.24–1.26 (95% CI 1.1 to 1.4)).11 Beyond this, evidence suggests a positive link between health anxiety and increased healthcare utilisation;12–17 although, one study reported no significant relationship between health anxiety and service use.18 These studies typically draw from the general population or specific patient cohorts and often rely on retrospective patient accounts, which are prone to recall bias. Moreover, an epidemiological study conducted in Australia showed that approximately 5.7% of the Australian population experiences health anxiety at some point in their lifetime.15 The same study further confirmed that individuals with health anxiety during the study period were found to access mental healthcare services at a significantly greater rate, compared with those without health anxiety. However, this study used brief screening questions to assess health anxiety, and these questions were not specifically structured to evaluate all Diagnostic and Statistical Manual of Mental Disorders-IV criteria needed for a diagnosis of hypochondriasis. Additionally, a previous study investigated factors affecting individuals’ ability to self-triage; however, this study failed to consider whether other factors such as health anxiety (or hypochondriasis) influence the self-triage decision mechanism.10 Thus, given the importance of this topic, further research in this domain is warranted.
In response, our study contributes to the limited research on health anxiety and self-triage by exploring the relationship between health anxiety and self-triage accuracy among patients who visited the ED for non-urgent health conditions. In terms of health anxiety, previous studies19 20 have discussed the distinction between health anxiety and hypochondriasis proposed by the diagnostic manuals and provided evidence that does not support this separation. Consistent with this work, we define health anxiety as a broader concept that extends beyond the specific diagnostic criteria. Also, it can manifest as hypochondriasis at its most severe form in the spectrum. In addition, we focus on this distinctive patient group in our study, because they independently decide to seek care in the ED. Furthermore, their decision-making process may differ from that of the general population or other patient cohorts who may not experience the same level of urgency. Importantly, in the Australian context, it is widely discussed and emphasised that ED non-urgent patients can be adequately treated by primary healthcare providers.21–23 Therefore, understanding the association between health anxiety and self-triage decisions is vital for enhancing patients’ care-seeking behaviours and designing effective interventions to reduce the misuse of health services.
Methods
Study design, participants and collection procedures
We conducted an anonymous, scenario-based survey using the Qualtrics online platform. This non-probability sampling involved an exhaustive recruitment strategy, aiming to include non-urgent patients who visited the ED at a single centre, the Princess Alexandra Hospital in Brisbane, Australia, between 13 December 2022 and 30 August 2023. The survey, designed in collaboration with ED nurse educators and nursing academics, was tailored for comprehensibility at a 12th-grade reading level or below. It included questions about demographics, cognitive biases, self-rated risk-taking willingness24 and risk-taking behaviour using a balloon analogue risk task25 and six hypothetical medical scenarios for self-triage. Additionally, the survey featured a health anxiety test (using the Whiteley Index (WI)-6)26 and the Big Five personality traits Mini-Marker. The study protocol was registered with the Open Science Framework (https://osf.io/5ysjc).
Patients were eligible to participate in the study if they belonged to the Australasian Triage Scale (ATS) category four or five—non-urgent, were 18 years or older, visited the ED any day of the week between 6:00 a.m. and 23:00 p.m., spoke English and were physically and mentally capable of participating in the survey. Exclusion criteria included being in ATS categories one to three, experiencing significant pain or discomfort, requiring translator or interpreter services, having a referral from a primary healthcare provider, and being under the influence of alcohol or drugs. We further excluded patients with primary or severe mental health issues. Participants could only complete the survey once, regardless of multiple ED visits during the study period.
Patient eligibility was determined with the help of emergency triage nurses on duty each day. Once identified, the first author approached potential participants to explain the study details. If patients agreed, they were given a tablet to complete the survey, with responses directly recorded on an online server. Patients could choose not to answer any question or discontinue participation at any time if they felt discomfort. Submission of the completed survey served as consent to participate. Recruitment was conducted individually, and as a token of appreciation, each participant received a gift voucher worth up to AU$6.
Health anxiety
We measured health anxiety using the WI,26 a validated tool initially comprising 14 questions that assess health anxiety on a dichotomous scale. Prior research indicates that a subset of six questions suffices for English-speaking participants.27 28 We thus employed the six-item version (WI-6), focusing on health anxiety and bodily preoccupation. Responses ranged from ‘not at all’ to ‘a great deal’ and scored from 1 (lowest) to 5 (highest), with total scores ranging from 6 to 30. Incomplete responses at random were excluded. Reliability analysis showed a Cronbach’s alpha coefficient of 0.86. Scores were categorised with a score of 6 indicating ‘no health anxiety’; scores from 7 to 30 were divided into four quartiles reflecting varying levels of health anxiety, with the first quartile showing lower levels and subsequent quartiles indicating higher levels. This approach enhanced our understanding of how varying levels of health anxiety influence self-triaged decisions.
Outcomes
The primary outcome was the number of accurately self-triaged decisions, derived from responses to six hypothetical medical scenarios. Each scenario had a correct response as detailed in the appendix (online supplemental table 1), and the accuracy was confirmed with the help of ED nurse educators and nursing academics. The primary outcome was further divided into two subgroups based on the acuity: severe and minor health conditions.
Supplemental material
The secondary outcomes included counts of inaccurately self-triaged decisions, divided into overtriaged and undertriaged categories. Overtriaged decisions occurred when patients overestimated the necessity of care compared with the verified appropriate response for the given scenario. Conversely, undertriaged decisions were those in which patients underestimated the required level of care. Similar to accurately self-triaged decisions, undertriaged decisions were divided into two subgroups (severe and minor health conditions). Notably, there were no overtriage options for scenarios classified as severe health conditions. We performed a pairwise correlation between overtriaged and undertriaged decisions and found a negative correlation (−0.41, p <0.0001).
Statistical analysis
Given the lack of previous research on the relationship between health anxiety and self-triage decisions, we assumed that a sample of 400 non-urgent patients would be sufficient to detect a small effect size (Cohen’s 
of 0·20) with a significance level (
a
) of 0·05 and a power 
of 0·90. The effect size was assumed based on previous studies that explore health anxiety and avoidance/reassurance-seeking behaviour and factors influencing ED use.29 30
Non-urgent patients’ characteristics were summarised using the ranges, means, SD and percentages, and associations were visually presented with a graph.
In the pre-registration, logistic regression analysis was planned because the outcome was initially conceptualised as a binary variable. However, in the actual analysis, we treated the outcome as a count variable (ie, the number of accurate or inaccurate self-triage decisions), which provided richer insights. Postestimation checks revealed that logistic regression was not a good fit for this approach. Consequently, multivariable Poisson regression analyses with robust standard errors were performed separately for each outcome to investigate the association with health anxiety. A sequential model specification approach was used to build the model, incrementally adding theoretically or empirically relevant control variables to examine the robustness of the relationship between health anxiety and self-triaged decisions (online supplemental tables 2-4). All analyses were adjusted for patients’ socio-demographic characteristics, distance to the closest ED from the current residing suburb, the number of GP clinics per 1000 people in each suburb, and time effects (date, time, week and month of survey submission). Marginal effects at the means were calculated for each model. A comprehensive description of each variable is provided in the appendix (pp. 2–3).
Subgroup analyses by acuity were conducted to examine the associations between health anxiety and self-triaged decisions for severe and minor health conditions. In addition, separate analyses stratified by age and sex were performed to investigate further how health anxiety might influence self-triaged decisions across different demographic cohorts. These analyses provided a detailed exploration of the association between health anxiety and self-triaged decisions across different subgroups.
We conducted sensitivity analyses by controlling for patients’ risk-taking willingness because health decisions are often made under risk and uncertainty in which the probability of outcomes is not known.31 32 Risk-taking willingness was assessed using a self-rated question on an 11-Likert scale ranging from 0 being not at all willing to take risks and 10 being very willing to take risks. Patients were asked: how do you see yourself? Are you generally a person who is very willing to take risks or do you try to avoid taking risks?24 Additionally, we recategorised our health anxiety total scores (ranging from 6 to 30) based on a prior study’s14 classification and reassessed its relationship with self-triage outcomes to investigate whether our main findings are robust. Results from these sensitivity analyses were reported in the appendix.
A total of 27 responses for the WI-6 were missing at random, with no identifiable patterns or characteristics distinguishing these cases from the rest of the sample. These patients who did not answer all six health anxiety questions were excluded from the analyses. In addition, data from other variables were also missing at random. No imputation methods were employed, and only completed observations were included in our analyses. All regression results are reported with 95% CIs, and a p value below 0·05 was considered statistically significant. Data analyses were conducted using STATA software, V.17·0 (StataCorp).
Patient and public involvement
No patients were directly involved in the study design, developing research questions, data collection, data analysis, interpretation of results and reporting of this study. However, during the recruitment process, the first author did discuss the study with patients. Additionally, prior to designing the survey instrument, the first author engaged in multiple discussions with ED triage nurse educators to get a comprehensive understanding of the types of patients present in the ED and their health concerns. We plan to disseminate this paper widely to members of the public after the publication.
Results
Of the 527 non-urgent patients who met the eligibility criteria and were approached, 122 declined to participate; five initially agreed but had to discontinue as they were called in for treatment by an emergency physician; 400 patients consented and submitted the survey, yielding a participation rate of 75·9%.
The characteristics of our study participants are shown in table 1. The mean (SD) age of participants was 39·3 (15·2) years, ranging from 18 to 86 years old. A slight majority of the participants were female (52·2%). Nearly half of the participants were in a relationship (49·3%), and a significant proportion held an undergraduate qualification (30%). Among the participants, 64.5% were Australians. 66 participants (16·5%) reported an average annual income above AU$100 001. Of the 400 participants, 234 (58·5%) did not have private health insurance, 72·9% of participants visited the ED once and 30% of participants visited other health clinics or services once within 6 months before participating in the survey. The mean (SD) number of GP clinics per 1000 people was 0·3 (0·2) and the mean (SD) distance to the closest ED from participants’ currently residing suburb was 7·2 (7·1) km.
Characteristics of Non-urgent Patients
Figure 1 displays the percentages of accurately and inaccurately self-triaged decisions made by non-urgent patients for six given hypothetical medical scenarios. Overall, a majority of patients underestimated both severe and minor health conditions, leading to inaccurate decisions. Specifically, only 58.9% of patients accurately self-triaged for the sore throat scenario.
Self-triaged decisions made by non-urgent patients for six medical scenarios.
Figure 2 shows the relationship between health anxiety and self-triaged decisions among non-urgent patients. No relationships were observed between health anxiety and accurately self-triaged decisions or undertriaged decisions. We observed a slightly positive relationship between health anxiety and overtriaged decisions.
The relationship between health anxiety and self-triaged decisions.
The results from the multivariable Poisson regression models, which present the marginal effects of factors associated with accurately self-triaged decisions, are reported in table 2. Notably, no statistically significant association was found between health anxiety and accurately self-triaged decisions (model 1). However, we found that age and sex were significantly associated with accurately self-triaged decisions (model 1). Furthermore, subgroup analyses by acuity revealed that age, sex and income were associated with accurately self-triaged decisions under severe health conditions (model 2).
Factors associated with accurately self-triaged decisions
Table 3 shows the results of the marginal effects of factors associated with overtriaged and undertriaged decisions from multivariable Poisson regression models. Patients across all WI-6 quartiles (ie, first to fourth) exhibited higher tendencies in making overtriaged decisions, relative to those exhibiting no health anxiety (model 1). Notably, the marginal effects for patients in the third and fourth quartiles (relative to patients with no health anxiety) were 0·29 (95% CI 0·09 to 0·50) and 0·25 (95% CI 0·07 to 0·44), respectively. In contrast, no statistically significant association was found between health anxiety and undertriaged decisions (model 2), except for patients in the fourth quartile of the WI-6 (−0·93, 95% CI −1·61 to −0·24). Subgroup analyses by acuity revealed negative associations between health anxiety and undertriaged decisions (models 3 and 4).
Factors associated with inaccurately self-triaged decisions
Analyses stratified by sex showed no significant association between health anxiety and accurately self-triaged decisions (online supplemental table 5). However, female and male patients in the highest quartile of the WI-6 (fourth quartile) were significantly associated with overtriaged decisions, compared with those with no health anxiety. Furthermore, the marginal effects for female and male patients in the fourth quartile (relative to those with no health anxiety) were 0·26 (95% CI 0·02 to 0·49) and 0·27 (95% CI 0·05 to 0·48), respectively (online supplemental table 6). In contrast, a statistically significant negative association was observed between health anxiety and undertriaged decisions among only female patients (online supplemental table 7). Further analyses conducted stratifying by age uncovered important results regarding the relationship between health anxiety and inaccurately self-triaged decisions (online supplemental tables 8-10). Notably, the association between patients in the first to fourth quartiles of the WI-6 within the older age group (35–79 years) and overtriaged decisions was statistically significant (online supplemental table 9).
Additional analyses using two self-reported behavioural proxies of healthcare utilisation—the number of ED visits and the number of other health clinics in the last 6 months—confirmed similar trends (online supplemental table 11). Compared with patients with no health anxiety, those with the highest WI-6 scores (fourth quartile) exhibited positive associations with the number of ED visits (0·43 (95% CI 0·10 to 0·77)) and the number of visits to other health services (1·46 (95% CI 0·80 to 2·11)) in the last 6 months.
In our sensitivity analyses, after controlling for self-rated risk-taking willingness, the results remained consistent for both accurately and inaccurately self-triaged decisions (online supplemental tables 12 and 13). In addition, rerunning the analyses with the reconstructed health anxiety variable indicated similar results as shown in our main results (online supplemental tables 14 and 15).
Discussion
To the best of our knowledge, this is the first cross-sectional study examining the relationship between health anxiety and self-triage decisions among ED non-urgent patients. The exploratory nature of this study facilitated identifying statistically significant associations between health anxiety and inaccurately self-triaged decisions, affecting both overtriaged and undertriaged decisions.
The reported total health anxiety score in our study ranged from 6 to 30, with a mean (SD) of 12·4 (5·1). Existing studies conducted in different contexts using the WI-6 reported a mean (SD) score of 15·8 (6·6) in US patients at a community health centre.33 These differences arise from variations in study settings and the sociodemographic characteristics of the studied population. Nevertheless, the distribution of health anxiety scores in our study aligns closely with those from previous studies involving primary healthcare patients.12 34 This underscores its relevance to our ED non-urgent patients, who might also be treated by primary healthcare professionals.
Our results revealed no significant association between health anxiety and accurately self-triaged decisions. This could be due to other factors, such as previous similar experiences, health literacy and/or cognitive coping strategies, which may have played a role in the observed associations, rather than health anxiety alone. However, information related to these factors was not collected in this study. In contrast, the findings support the hypothesis that there is a directional relationship between health anxiety and self-triaged decisions, that is, individuals with higher levels of health anxiety are more likely to overtriage and less likely to undertriage. A possible explanation could be that anxious individuals may be more likely to overestimate serious, negative health threats.35 Consequently, this can lead to misuse of health services more frequently. These findings echo limited prior research exploring the nexus between health anxiety or hypochondriasis and health service utilisation.14 16
In addition, the exploratory design of our study enabled us to get a nuanced understanding of the varying relationship between health anxiety and self-triaged decisions among different patient subgroups. Consistent with a previous study12 that showed a link between health anxiety and healthcare utilisation among older adults, our subgroup analyses, stratified by age, revealed a positive association between health anxiety and overtriaged decisions, particularly among older non-urgent patients (aged between 35 years and 79 years). Evidence suggests that anxiety affects how older adults perceive their physical health.36 Furthermore, this implies that deteriorating health in older adults may be a stressor to make overtriaged decisions, leading to inefficient use of health services.
Implications
The implications of our findings are significant for healthcare policy and practice. Health anxiety in patients can lead to substantial economic burdens on healthcare systems due to unnecessary use of services. This highlights the need for routine diagnosis, treatment and management of severe health anxiety (hypochondriasis) to reduce disease burden and curb healthcare costs.17 Our research also underscores the need for critical considerations in the design and development of digital symptom checkers and self-triage decision support tools37 to account for behavioural and psychological factors. These tools can incorporate user-friendly, tailored diagnostic questions based on demographic characteristics (ie, age, sex), which lessen cognitive overload and guide potential users toward accurate self-triage decisions. Also, leveraging health anxiety assessments and health literacy aids in decision support tools can further enhance the accuracy of self-triage decisions. Consequently, these strategies will ensure that self-triage tools are not only technologically robust but also align with users’ behavioural and psychological needs. However, to effectively address patients’ health anxiety in EDs, clinicians, nurses, social health workers and mental health professionals should adopt a collaborative approach. This multidisciplinary effort can help to ensure that patients receive more holistic care, including appropriate referrals to primary healthcare providers. Moreover, untreated severe health anxiety may lead to other severe consequences such as diminished quality of life, increased sick leaves, higher rates of disability pension awards38 39 and long-term medical conditions such as ischaemic heart disease.40 Addressing the needs of severe health anxiety among patients is therefore imperative.
Conclusions
The chronic misuse of health services is a significant public health challenge worldwide. This exploratory study investigated the association between health anxiety and self-triaged decisions among ED non-urgent patients. The results suggested that health anxiety was positively associated with overtriaged decisions. These findings align with the established positive association between health anxiety and increased utilisation of health services. However, as this study is exploratory, further confirmatory research is required to validate our findings. The results of this study can guide clinicians, health promotion professionals and policymakers in designing interventions to reduce unnecessary care-seeking behaviour of potential patients and navigate patients to appropriate care more promptly.
Data availability statement
No data are available. The study protocol, survey instrument and STATA do files can be made available upon request. However, the deidentified individual patient data used in this study are the property of the Queensland University of Technology and cannot be made available. Please contact the corresponding author for further information.
Ethics statements
Patient consent for publication
Ethics approval
The study protocol received ethics approval from the Metro South Health Human Research Ethics Committee (HREC/2022/QMS/84820) and the Queensland University of Technology Human Research Ethics Committee (HE-AdRev 2022-6321-10997). Additionally, the Metro South Health Research Governance Office approved the site-specific assessment (SSA/2022/QMS/84820).
Acknowledgments
This study is a part of TNEK’s PhD thesis. TNEK received a grant (Rethinking Emergency Medicine Research Grant - 2022) from the PA Research Foundation, Australia to conduct this work and the findings of this study were included in the final report submitted to the PA Research Foundation. We thank the research manager (Dr Rob Eley) in the PA emergency department for his guidance and support for this project, the Head of the emergency department and the director of Emergency Medicine Research for their technical support. We also thank emergency nurse educators for their support in designing the survey instrument and triage nurses, physicians, allied health professionals and administrative staff for their support during the recruitment process and the non-urgent patients who participated in our survey. Finally, we thank attendees of the 15th workshop on the Economics of Health and Well-Being (19–21 Feb 2024 in Melbourne, Australia) for their comments.
References
Footnotes
Contributors TNEK, BT, HFC, JH and SW conceptualised the study. TNEK, HFC, JH, BT and SW contributed to the methodology. TNEK collected the data and administered the project. TNEK performed the formal analysis and wrote the original draft. All authors reviewed and edited the original draft and agreed to submit the manuscript for publication. TNEK is responsible for the overall content as guarantor-2025.
Funding This study was supported by a grant (Rethinking Emergency Medicine Research Grant - 2022) from the PA Research Foundation, Australia. The funder had no role in the study design, data collection, analysis, interpretation of data and reporting of this study.
Competing interests Thilini Nisansala Egoda Kapuralalage had financial support from the PA Research Foundation for the submitted work. All other authors (HFC, JH, BT, SW) have no competing interest to declare.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.