The Role of Language in Remote Healthcare Triage: A Meta‐Aggregative Review

ABSTRACT

Aim(s)

To provide an overview of evidence on the role of language in remote healthcare services prioritisation, from now on termed triage. This study synthesises literature, to better understand how language affects triage interactions, aiming to improve these processes.

Design

We conducted a meta‐aggregative review.

Methods

A systematic search in Scopus from inception to September 2023 identified 437 studies on language in remote healthcare triage, of which 23 were included. Information was selected using both inductive and deductive coding, focusing on six linguistic features of interaction that have been described in the literature on studies using conversation analysis: turn‐taking organisation, overall structure, sequence organisation, turn design, lexical choice, and epistemological and other forms of asymmetry. The process followed the RAMESES Publication Standards for Meta‐narrative Reviews.

Results

Two main findings emerged. First, all six linguistic features are present in triage conversations, indicating that language involves more than just what is said. It also matters, for example, how and when a question is asked. Second, computerised decision support systems (CDSS) significantly affect conversation flow and dynamics.

Conclusion

Language in triage involves more than literal speech and is heavily influenced by CDSS.

Implications for the Profession and/or Patient Care

Our study suggests that quality assessments of triage conversations should consider not only what is said but cover all relevant aspects of language. The influence of computerised decision support systems (CDSS) on linguistic features highlights the need for systems to be adaptable, to improve conversation quality and better addressing caller needs rather than focusing on one‐size‐fits‐all questions.

Impact

This review highlights the complex role of language in triage conversations and its impact on interaction. It calls for a broader view of language in quality assessments, recognising that both call‐takers and callers contribute to call quality. Insights from this review can help developers enhance question types, sequence, and delivery methods of computerised decision support systems. Finally, education for call‐takers in healthcare sectors may be improved based on our findings.

Patient or Public Contribution

No patient or public contribution.

Summary.

• What does this paper contribute to the wider global clinical community?

  • ○
    Our study highlights that language in triage conversations involves more than just the spoken words. This insight challenges the current focus on verbatim content in quality assessments and audits, advocating for a broader approach that considers how language is used and its interactional nature. Our findings suggest incorporating the caller's perspective into audits and quality assessments.
  • ○
    Remote triage conversations occur across various healthcare settings, including ambulance services, emergency departments, general practice, and helplines. The insights from this study can improve education on language use for healthcare professional in these diverse areas.
  • ○
    Since computerised decision support systems significantly influence language during triage, our findings provide valuable information for the development and refinement of these systems, ensuring they support effective communication and decision‐making.

1. Introduction

Safety and efficiency are crucial, interconnected aspects of quality of care (Cioffi and Dip 1988; Giesen et al. 2007; Wolfe 2001). Effective triage is essential for achieving both, as it involves determining ‘who should receive which type of care’, optimising the distribution of care within the constraints of healthcare time and budget (Oddsson 2003). In pre‐hospital settings, triage helps decide the most appropriate care and location for a particular patient, whether before a hospital visit or before a general practice consultation (i.e., primary medical care; Blank et al. 2012). This process requires making decisions about whether and which healthcare professional should see the patient and when.

Triage involves interpersonal interaction, so its effectiveness hinges on good communication between healthcare seekers and triage handlers. The quality of triage is thus partly dependent on how well these interactions are managed. Effective communication ensures accurate assessment of healthcare seeker's issues and helps them feel heard (Holmström 2007; Röing and Holmström 2015). Optimising triage quality is critical, as medical errors can arise from ‘missed opportunities’ during triage, such as not listening to the healthcare seeker, poor communication, focusing on irrelevant aspects or asking too few questions (Ernesäter et al. 2012). Previous studies have shown that communication problems account for 38% of disputes during telephone consultations (Holmströ and Holmströ 2007; Rees et al. 2017). Therefore, call‐takers, including nurses, doctors, or non‐medically trained personnel like receptionists, need strong communication skills to manage healthcare seekers with diverse communication abilities (Pettinari and Jessopp 2001; Röing, Rosenqvist, and Holmström 2013).

Remote triage presents additional challenges, as the patient and triage staff are not physically present in the same space. The distance between healthcare seekers and triage staff can vary, and most remote triage occurs via telephone or video calls. These methods provide fewer communicative cues than face‐to‐face interactions (Giesen et al. 2011; Kuriyama, Urushidani, and Nakayama 2017). Both telephone and video consultations tend to be shorter and less informative than face‐to‐face consultations (Hammersley et al. 2019; Hewitt, Gafaranga, and McKinstry 2010). This can complicate the assessment of a patient's condition and the urgency of their needs (Coster et al. 2017; Katz et al. 2008). Call‐takers must weigh hypothetical odds of a diagnosis and its impact, against access to care and risk of calamities—serious or urgent situations such as severe accidents that could significantly affect the patient's safety or outcome—often under time pressure (Edwards 1994; Zhang et al. 2024). They must handle multiple tasks, including gathering information about symptoms, dealing with diverse types of patients, and navigating technology systems like computerised decision support systems (CDSS), which are used in 77%–89% of remote triage cases (Delichatsios, Callahan, and Charlson 1998; Durham et al. 2019; Strasser et al. 1979). While CDSS is designed to assist call‐takers, it can also complicate communication by negatively affecting both interactions and the work process, for example, related to question order or place to enter the healthcare seeker's input (Wouters et al. 2020).

Current audits and quality assessments of triage conversations often focus solely on the call‐taker's actions, overlooking the interactive nature of these conversations (InEen 2017; Morgan and Muskett 2020; Rolink, Bos, and de Boer 2019). The emphasis is usually on what was said, treating triage as a checklist exercise (Chambers et al. 2019; EMD‐Q Performance Standards Edition 10 2018; InEen 2017; Rolink, Bos, and de Boer 2019). However, linguistic research indicates that conversations are constructed with various interrelated linguistic aspects that are in turn interrelated. Thus, in fact, conversational quality involves more than just ‘what is said’ (Drew and Heritage 1992; Greenhalgh and Manzano 2021; Heritage and Clayman 2011). This study examines whether there are existing studies on linguistic features in remote healthcare triage and if so, which linguistic features these are.

To address this, we reviewed literature on the role of language in remote healthcare triage. Our multidisciplinary approach, involving researchers/authors from different disciplines, reflects the growing interest in integrating non‐medical disciplines such as linguistics into healthcare (Mårtensson et al. 2016). Conversation analysts have identified six key linguistic features in institutional interactions (Drew and Heritage 1992; Greenhalgh and Manzano 2021; Heritage and Clayman 2011) (1) turn‐taking organisation (how are turns divided between speakers, who has the rights to speak, how are turns taken), (2) overall structure (how is a conversation in this context commonly structured), (3) sequence organisation (what are the actions done in talk, and how do these relate to each other sequentially), (4) turn design (how are turns formatted), (5) lexical choice (within turns, what word choices are made), and (6) epistemological and other forms of asymmetry (how do participants construct asymmetry in knowledge and participation) (Drew and Heritage 1992; Greenhalgh and Manzano 2021; Heritage and Clayman 2011). This meta‐aggregative review uses these features as a lens to describe the role of language in remote triage.

2. Aim(s)

Recent, studies have underscored the significance of language in triage conversations, yet a comprehensive overview of these studies is still lacking. Our goal is to address the following research questions: ‘What is the role of language in remote healthcare triage conversations?’ and ‘How can understanding this role optimise the triage process?’. By synthesising existing knowledge, identifying knowledge gaps, and exploring implications for education and practice, we aim to enhance the effectiveness and efficiency of remote triage practices. This synthesis will also highlight specific areas where further research is needed.

3. Methods/Methodology

3.1. Design

We used a meta‐aggregative review design (Amog et al. 2022; Hannes and Lockwood 2011). A meta‐aggregative review design is employed to synthesise data from multiple studies by aggregating qualitative findings to identify overarching themes, patterns, or concepts. Unlike traditional meta‐analyses, which primarily focus on quantitative data, meta‐aggregative reviews integrate qualitative insights to provide a more comprehensive understanding of complex phenomena. In our study, we chose this approach to capture a broad spectrum of insights from diverse sources. Although meta‐aggregative reviews typically focus on qualitative data, inclusion of articles using a quantitative approach is allowed.

3.2. Search Methods

We searched the bibliographic database Scopus from inception (as early as possible) to September 2023 for research papers that describe, analyse and/or evaluate a linguistic feature in remote healthcare triage (Ballew 2009). Scopus is a large and accurate database of articles from medical databases, as well as databases in other fields, such as more linguistically‐ or communication‐oriented databases. An overview of our search strategy, based on our inclusion and exclusion criteria, is shown in Table 1. We did not apply publication date limits. Language was restricted to English or Dutch.

TABLE 1.

Search terms used in the Scopus database. The full search string is available upon request from the authors.

Domain: primary care/emergency care/healthcare	Triage/prioritising	Distance/telephone	Linguistic approach
( {primary care} )	( triag* )	( telephon* )	( conversation W/50 analy* )
( {family medicine} )	( urgency W/50 allocat* )	( phon* )	( {discursive psychology} )
( {general practice} )	( prioritiz* )	( call* )	( {question design} )
( {general medicine} )	( dispatch* )	( remote )	( {interactional activity} )
( {family practice} )	( urgency W/50 determin* )	( video )	( discourse W/50 analy* )
( {general practitioner} OR {general practitioners} )	( urgency W/50 select* )	( distance )	( linguistic W/50 approach* )
( GP OR GPs )	( decision W/1 support W/1 system)	( tele* )	( linguistic W/50 analy* )
( {family physician} OR {family physicians} )	( emergency )	( helpline )	( interaction W/50 analy* )
( nurs* )	( 911 )		( interactional W/50 analy* )
( physician‐on‐call )	( 999 )		( talk‐in‐interaction )
( {physician on call} )	( 112 )		( {talk in interaction} )
( {emergency physician} OR {emergency physicians} )	( 111 )
( ambulance )
( {emergency care} )
( {emergency medicine} )
( {health care} )
( {health‐care} )
( {health service} )
( {health‐service} )
( {medical aid} )
( {medical‐aid} )
( {public health} )
( {public‐health} )
( {medical care} )
( {medical‐care} )
( {community care} )
( {community‐care} )
( {patient care} )
( {patient‐care} )

3.3. Inclusion and/or Exclusion Criteria

We included all original studies, irrespective of type (quantitative, qualitative, and mixed methods) that fulfilled our pre‐defined inclusion and exclusion criteria listed below. Protocol papers and reviews were excluded.

Our inclusion and exclusion criteria are further specified below.

1. Healthcare

   Included studies should have been conducted in healthcare settings where (remote) triage is common such as general practice, or emergency care, irrespective of the caregivers' profession, for example, nurses, paramedics, doctors. Studies conducted in an inauthentic setting such as scenario training were excluded.

2. Triage/prioritising

   There should be some form of triage or prioritisation. This means that helpline studies can also be included if they not only give standard advice, but also prioritise which care is best for that person. Both studies that did and did not use a CDSS for triage were included. We solely focus on the function of a CDSS if this affected language aspects of remote triage and not on the function of this CDSS in general.

3. Remote/telephone

   Triage should be done remotely, by phone or video so that the person seeking help is not in the same place as the healthcare provider. Studies in which only written text was used for triage were excluded.

4. Linguistic approach

   The approach of the included studies had to be linguistic or at least evaluate a linguistic feature. We excluded studies in which people say something about this process but the aspect itself was not analysed in detail, such as interview studies. However, we included these studies if interviews were combined with another type of analysis and then included just the linguistic parts of the analysis in our results.

3.4. Search Outcome

An overview of our search results and selection of articles is illustrated in Figure 1. Our Scopus search yielded 437 articles. The title and abstract of those were screened by Author1, Author2, Author3 and partly by Author7. Screening disagreements were discussed for consensus at a meeting. When doubts remained, the full text of the article was assessed. A total of 401 articles were excluded based on the eligibility criteria, leaving 36 articles for full‐text screening by Author1 and Author2. We excluded a further 13 studies for distinct reasons: no healthcare setting (1 article), lack of triage/prioritisation (6 articles), no remote triage (2 articles), no analysis of a language aspect (3 articles), and analysis of written rather than spoken text (1 article). This left 23 studies suitable for inclusion. There was immediate and unanimous consensus among the authors regarding this selection.

FIGURE 1.

Flow diagram of the search and the selection of included studies.

3.5. Quality Appraisal

Several methods are available for quality appraisal of both quantitative and qualitative healthcare papers (Dixon‐Woods et al. 2004; Higgins et al. 2008). However, conversation analysis, the method used in most of the included papers of this review, cannot be positioned as a quantitative research method, because primary outcomes do not include numbers or statistical data (Heritage and Clayman 2011). Moreover, it also differs from typical qualitative research methods, because it does not analyse cognitions as per interview studies or focus groups (Atkins et al. 2008; Curl and Drew 2008; Heritage and Clayman 2011; Perakyla 1998; Popay 2006). In the absence of suitable quality appraisal methods for appraising our included studies, we instead, as suggested in the literature described (1) the type and amount of data and (2) the analysis in a summary table (Parry and Land 2013).

3.6. Data Abstraction

Using a standardised data extraction form, we extracted the following information from all included studies: (1) article information; title, authors, year of publication, (2) study characteristics; country, language, research aim, research method, setting, people involved, research sample, communication medium, tool for triage and (3) findings; which results were described and which conclusions were drawn and what dimension or aspect of talk were these related to. For the latter, we used the six linguistic features to structure our analysis of the role of language in triage conversations.

3.7. Synthesis

We used QSR's NVivo software to analyse the articles (Thomas and Harden 2008). We combined deductive and inductive coding. We started with a subset of codes for deductive coding based on initial literature readings: (1) background information; observations about triage conversations, role, value or relevance of linguistic approach, (2) setting information; country, language, setting, people involved, research sample, communication medium, method of prioritisation, tool for triage, outcome measure, (3) analysis information; research aim, research method, overview of results, conclusion, implications and suggestions for future research. Author1 first started with line‐by‐line coding of manuscripts. A selection of manuscripts, including those about which Author1 had doubts during coding, was also coded by Author2 and Author3. The purpose of this double coding was to achieve consensus and transparency resulting in a collective agreement in the final coded sections (Friese 2020). When agreement was reached, we organised coding of the result section by using linguistic features for further coding of overview of results, conclusion, implications, and suggestions for future research (Drew and Heritage 1992; Greenhalgh and Manzano 2021; Heritage and Clayman 2011). Following this, inductive coding of all codes (linguistic features and subcodes) was done by Author1, Author2 and Author7 to organise all selected codes. The final codes and subcodes related to our research question were then determined in a consensus meeting with Author1, Author2 and Author6 and Author7.

4. Results/Findings

Table 2 summarises the 23 studies included in our review, which analysed 18 unique datasets (some datasets were used more than once, with a focus of the analysis on different aspects). Nearly all of these studies (22 out of 23) focused on telephone triage. The studies were conducted across eight different settings, with most studies related to ambulance services. Below, we present the findings organised by linguistic features and end with a paragraph on how these features impact triage outcomes.

TABLE 2.

Overview of studies included in the review.

Authors (year)	Country	Language	Aim	Research method	Setting	People involved	Research sample	Communication medium	Tool for triage
Mellinger (1994)	United States of America	American English	To examine the negotiation of directives in radio calls between paramedics at the scene of a medical emergency and emergency department (MICU) nurses	Conversation Analysis	Emergency medical services—emergency department (=in hospital care)	Paramedic vs. emergency department nurse	Corpus of radio calls between paramedics and emergency department (MICU) nurses. When a paramedic arrives on the scene of an emergency, a radio call is routinely made to the nearest emergency department to describe the victim's symptoms and to receive orders for medical action from the emergency department nurse	Radio call	Not described
Riou, Ball, O'Halloran, et al. (2018) a	Australia	Australian English	To determine whether the type of sentence used by callers in response to the question ‘Is s/he breathing?’ had an impact on call handler recognition of agonal breathing and thus, identification of OHCA	Conversation Analysis and logistic regression with Mann–Whitney U test	Emergency medical services—ambulance (=out of hospital care)	Caller (other than the patient) vs. call‐taker	Selection of 200 OHCA cases (non‐traumatic OHCA in adults (> 13 years) attended by SJA‐WA paramedics during the study period: 100 calls in which cardiac arrest was recognised by the call handler and 100 in which cardiac arrest was not recognised by the call handler. 12 calls were excluded after selection	Telephone	Medical Priority Dispatch System (MPDS)
Riou et al. (2017) a	Australia	Australian English	To explore the impact of the linguistic variations in the way call handlers say the same scripted sentence (the reason‐for‐the‐call prompt). Specifically, the impact of these variations on the way callers subsequently describes the emergency and the timing of calls	Conversation Analysis, prosodic analysis, quantativelyoriented Corpus Linguistics	Emergency medical services—ambulance (=out of hospital care)	Caller (other than the patient) vs. call‐taker	188 emergency calls for paramedic‐confirmed OHCA received at the call centre of St Joh Ambulance Western Australia (SJA‐WA) between 1 January 2014 and 31 December 2015 for the Perth metropolitan area	Telephone	Medical Priority Dispatch System (MPDS)
Riou, Ball, Williams, et al. (2018) a	Australia	Australian English	To analyse how callers' pre‐emption of a reason‐for‐the‐call shapes the trajectory of emergency calls. The focus is on two sequential environments: (1) at the moment of pre‐emption, that is, when call handlers want to attend to something else than the reason‐for‐the‐call and (2) during the reason‐for‐the‐call sequence, that is, when call handlers reach the protocol step when they are supposed to officially request a reason‐for‐the‐call	Conversation Analysis	Emergency medical services—ambulance (=out of hospital care)	Caller vs. call‐taker	66 calls for cardiac arrest received by SJA‐WA between January 2014 and December 2015 in Perth, Western Australia in which the caller pre‐empted a reason‐for‐the‐call	Telephone	Medical Priority Dispatch System (MPDS)
Palma et al. (2014)	Italy	Italian	To analyse factors associated with registered nurse under‐triage of EMS‐calls subsequently found to be associated with deaths, termed ‘green‐black code’ cases	Comparison of call characteristics, and Fele's conversational analysis method	Emergency medical services—ambulance (=out of hospital care)	Caller vs. registered nurse	Random selection of 839 EMS calls that occurred during 2011 at the EOC of Lecce, Italy. Compared to the characteristics of the sample telephone conversations with the characteristics of the 15 green‐black calls occurring in 2011	Telephone	Medical Priority Dispatch System (MPDS)
Hayashi (2019)	Japan	Japanese	To analyse a telephone conversation between a caller and a call handler at a Japanese emergency callcenter. It examines the process of occasioned semantics practiced by the call handler in which the critical features of gait and emesis were constructed. It sheds light on the cause of an incident with fatal consequences after an ambulance request was rejected and claims that the cause could be attributed to the call handler's choice of words, which is related to the categorization of mobility	Ethnomethodological Conversation Analysis and interactional semantic analysis/ occasioned semantics	Emergency medical services—ambulance (=out of hospital care)	Caller vs. call‐taker	1 student who called the 119 emergencies, which was found dead in his apartment 9 days after the date of call. Presumed time of death was the day after the student called the emergency callcenter	Telephone	Not described
Nattrass et al. (2017) b	South Africa	African, isiXhosa, English and other languages spoken in South‐Africa	To (1) describe the nature of place formulation difficulties in EMS‐calls at this site, (2) examine the behaviours of the call handler when establishing location, and (3) consider the implications of these findings for call handler training	Conversation Analysis of calls and thematic analysis of focus groups e	Emergency medical services—ambulance (=out of hospital care)	Caller (patient or proxy) vs. call‐taker	104 audio‐recorded calls made to the EMS centre during 2010–2013 was randomly selected by the call centre manager from various incident categories and made available to the research team for analysis	Telephone	Not described
Penn, Watermeyer, and Nattrass (2017) b	South Africa	African, isiXhosa, English and other languages spoken in South‐Africa	To (1) analyse how language mismatches are managed in emergency calls and (2) analyse what influences code switching behaviour in a multilingual context	Ethnography and Conversation Analysis e	Emergency medical services—ambulance (=out of hospital care)	Caller vs. call‐taker	21 calls made to the call centre between 2010 and 2013 in the following categories: trauma, maternity, patient unresponsive, call backs and teenage pregnancies in which a language mismatch or switch was evident	Telephone	Not described
Booker et al. (2018)	United Kingdom	British English	To explore common features of conversations occurring in a sample of emergency calls that result in an ambulance dispatch for a ‘primary care sensitive’ situation, and better understand the challenges of triaging this cohort	Conversation Analysis	Emergency medical services—ambulance (=out of hospital care)	Caller (patient or proxy) vs. non‐clinically qualified call‐taker	48 calls to one UK Ambulance Service in the period of September 2016 to January 2017 which were included because the caller had dialled the national emergency ‘999’ number and asked for an ambulance, had been triaged to receive an emergency ambulance response, but were subsequently deemed to be for a potentially ‘primary care sensitive’ situation	Telephone	Medical Priority Dispatch System (MPDS)
Allen (2021)	United Kingdom	British English	To analyse the socio‐material practices through which organisational understanding of patients is accomplished to prioritise calls and mobilise emergency medical services at the gateway of the healthcare system	Thematic analysis informed by, but not limited to, Translational Mobilisation Theory (TMT) e	Emergency medical services—ambulance (=out of hospital care)	Caller vs. non‐clinically qualified call‐taker	Data comprises 18 h of fieldwork in one of the three clinical contact centres in Wales using hip fracture as a tracer condition in several sites in a single region: EMS‐control centre, ambulance crews, Accident and Emergency Department, hospital wards, operating theatres and postanaesthetic recovery, community services and outpatient clinics	Telephone	Medical Priority Dispatch System (MPDS)
Tamminen et al. (2020)	Finland	Finnish	To examine the association between true OHCA confirmed by ambulance personnel and laypeople's spontaneous trigger words regarding physiological deterioration of a patient in the context of emergency‐dispatcher‐suspected or EMS‐encountered OHCA	Categorization of trigger words based on word list of Berdowski et al., statistical analysis	Emergency medical services—ambulance (=out of hospital care)	Bystander vs. trained emergency dispatcher	80 emergency calls of dispatcher‐suspected or EMS‐encountered out‐of‐hospital cardiac arrests between January 1, 2017, and May 31, 2017, in a Finnish hospital district	Telephone	Not described
Kevoe‐Feldman and Iversen (2022)	United States and Sweden	US English and Swedish	To compare U.S. 9‐1‐1 emergency calls and Swedish suicide helpline calls to identify practices for assisting callers whose interactional home is in the other arena	Comparative Conversation Analysis	Emergency medical services—ambulance (=out of hospital care) & helpline	Caller vs. call‐taker	37 suicide calls to 9‐1‐1 emergency callcenter; 20 first‐party callers included & 350 calls to suicide helpline of which 47 cases of boundary pushing, concerned with risk assessment and active risk management were included	Telephone	Not described
Leydon, Ekberg, and Drew (2013)	United Kingdom	British English	To gain insights in the process of delivering and seeking cancer related telephone help	Conversation Analysis	Helpline	Caller vs. frontline call‐taker	52 calls to a cancer helpline made between July 2010 and August 2011 in which there are a frontline call, a triage through to a nurse by the frontline call handler and a subsequent conversation between the specialist nurse and the caller	Telephone	Not described
Ericsson et al. (2019)	Sweden	Swedish	To explore the interaction between tele‐nurses and callers when MI patients called a national telehealth advisory service number as their first medical contact	Inductive content analysis	National Telehealth Advisory Service	Caller vs. telenurse	30 calls of patients with an evolving myocardial infarction which had answered in a survey that they had contacted an advisory tele‐nurse as their FMC	Telephone	CDSS
Morgan and Muskett (2020)	United Kingdom	British English	(1) To identify common points within the NHS 111 caLL protocol where the resultant interactions appear vulnerable to misalignment. (2) To explore the consequences of this misalignment for call outcome, specifically the clinical assessment and therefore the risk of system failure	Conversation Analysis	National Telehealth Advisory Service	Caller vs. non‐clinically qualified call‐taker	All serious incidents from the 2 years prior to data collection (from 1st March 2014 to 29th Feb 2016) with associated pathways assessments, event lists, and reports & other calls received from December 1st 2015 to January 31st 2016 (2 months) 10 calls across each of following timepoints (40 in total): ‘in‐hours’ weekday, ‘out‐of‐hours’ weekday, ‘out‐of‐hours’ weekend day, ‘out‐of‐hours’ weekend night	Telephone	NHS Pathways
Ernesäter et al. (2012)	Sweden	Swedish	To describe the characteristics of all malpractice claims following telephone calls to Swedish Healthcare Direct (SHD), including the identified causes, the healthcare providers' measures, and the communication between telenurses and callers in these cases	Qualitative Content Analysis & Roter Interaction Analysis System (RIAS)	National Telehealth Advisory Service	Caller vs. telenurse	33 malpractice claims arising out of telephone calls to Swedish Healthcare Direct (SHD_ during 2003–2010	Telephone	Not described
Young, Rochon, and Mihailidis (2016)	Canada	Canadian English	To (1) develop a PER model; (2) determine categories for the PER model features; (3) identify and calculate measures from call conversations (verbal ability, conversational structure, timing); and (4) examine conversational patterns and relationships between measures and model features applicable for improving the SDS's ability to automatically identify categories within the call model and predict a target response	Content analysis (quantitative conversational analysis)	Personal emergency response service	Personal emergency response service user vs. call operator	109 digitised call recordings of live PER calls between a real caller and a real call centre call handler in Canada collected after 2 years (2008: 52 calls and 2009: 57 calls)	Telephone	Not described
Spek et al. (2023) c	The Netherlands	Dutch	To better understand the interactional implications of discussing concerns during triage conversations between people who called the OHS‐PC for chest discomfort and triage nurses who use the NTS‐tool	Conversation Analysis	Out‐of‐hours primary care	Caller vs. triage nurse	35 triage conversations between people who called the OHS‐PC for chest discomfort and triage nurses who use the NTS‐tool in which concerns were discussed	Telephone	Netherlands Triage Standard (NTS)
Erkelens et al. (2021) c	The Netherlands	Dutch	To explore the implications of multiple‐choice either/or‐questions on the interaction between people who called the OHS‐PC and triage nurses who used the NTS during the calls for triage	Conversation Analysis	Out‐of‐hours primary care	Caller vs. triage nurse	68 telephone triage conversations between callers with chest discomfort and triage nurses at Dutch OHS‐PCs between 2014 and 2016	Telephone	Netherlands Triage Standard (NTS)
Weatherall and Grattan (2023)	New Zealand	New Zealand English	To document the skill and value of medical receptionist work and produce knowledge that can further enhance it	Conversation Analysis	Primary care	Caller vs. medical receptionist	18 calls to three different medical receptionists (all women) and 17 patients (one who had two calls included) between November 2018 through to January 2019.	Telephone	Not described
Leppänen (2010)	Sweden	Swedish	(1) To develop a framework for analysing how power operates in social interaction between nurses and patients. (2) To analyse power in one specific context, namely telephone‐advice nursing in Swedish primary care	Conversation Analysis and thematic analysis of semi‐structured interviews e	Primary care	Caller vs. nurse	276 audio‐recorded telephone calls to 13 different nurses at six primary‐care centres in 1999, and qualitative interviews with 18 nurses, including all the nurses who were recorded	Telephone	CDSS
Murdoch et al. (2014) d	United Kingdom	British English	To compare doctors' and nurses' communication with patients in primary care telephone triage consultations	Coding based on Conversation Analytic findings	Primary care	Patient vs. nurse & patient vs. GP	51 recorded calls for same‐day appointment requests (22 nurse and 29 GP) including 10 video‐recordings of nurses' use of Odyssey CDSS	Telephone	Odyssey by triage nurses; N/A by GPs
Murdoch et al. (2015) d	United Kingdom	British English	To achieve an understanding how nurses coordinate parallel activities of computer‐based activity and talk with patients (or their proxies), focusing on how nurses deployed and integrated CDSS in the delivery of telephone triage for same‐day appointments in primary care	Conversation Analysis	Primary care	Patient vs. nurse	22 audio‐recorded telephone triage nurse‐caller interactions from one GP practice in England, including 10 video‐recordings of nurses' use of CDSS during triage	Telephone	Odyssey

^a Studies used the same data.

^b Studies used the same data.

^c Studies used the same data.

^d Studies used the same data.

^e For completeness, we mention that this multiple‐/mixed‐method study also used focus groups or (semi‐structured) interviews but we did not include the results of this part in our analysis.

Table 3 summarises the results of this review.

TABLE 3.

Summary of results. The headings correspond to the headings in the results section. The findings presented in this table are described in more detail in the results section.

Turn‐taking organisation	Overall structure	Sequence organisation	Turn design
Call‐takers usually take the lead in initiating sequences and dividing turns. As a result, call‐takers (1) have more turns at talk, (2) control the agenda of the conversation and (3) commonly invite callers for extended turn, rather than callers taking extended turns on their own initiative.	The overall structure of telephone triage conversations comprises several phases.	A key activity by the caller is presenting the problem.	Turn design differences can sometimes be linked to who is calling or taking the call.
The length of a turn and the number of turns can also deviate from the regular pattern depending on the caller type, call‐taker type, and conversation phase.	The opening phase of the conversation is important because (1) a smooth opening positively affects the triage conversation and (2) the structure at the beginning of the call differs from an ordinary telephone conversation	The key activity of call‐takers is to grant or defer the caller's request for help.	Caller turns primarily take the form of reports and narratives.
	The nature of interaction between callers and call‐takers can differ between phases.	Call‐takers must balance actions expected by the CDSS with those that facilitate the conversation, such as reassuring the caller or asking further questions.	Emotions of fear or anxiety influence turn design in triage conversations.
	The structure of telephone triage is strongly influenced by the CDSS, or the dispatch protocol used.		Call‐takers' turn design is influenced by the computerised CDSS in three ways: (1) question design, (2) returning to the reason for calling, and (3) responding to additional information. First, the CDSS influences question design by dictating the type and grammatical tense of questions. Second, the CDSS affects how call‐takers return to the reason for the call when it was discussed before. Third, call‐takers' responses to additional information provided by callers are also influenced by the CDSS.

Lexical choice	Epistemological and other forms of asymmetry	Influence on triage outcome
Call‐takers sometimes rephrase specific words uttered by the caller.	Callers often treat call‐takers as having medical expertise, cautiously offering a candidate diagnosis, reinforcing medical knowledge asymmetry.	Although all linguistic features have been described in the included articles, the direct influence of these elements on the accuracy of urgency determination has not been detailed for all linguistic features.
A lexical item used by the caller may have different implications depending on the call's context.	Call‐takers assert their authority by taking the interactional lead when callers treat call‐takers as not the experts they need.	Regarding risk assessment, linguistic features turn‐taking organisation, turn design, and lexical choice were investigated for their potential to indicate urgency, but the aspects examined were not suitable for urgency determination.
Using two words interchangeably by call‐takers can be problematic for callers due to their different associations.	Asymmetry in knowledge of care organisation and triage processes affects interactions.	Missing elements of the dispatch protocol may lead to inadequate urgency determination.
	The CDSS introduces asymmetry as callers are usually unaware of its presence.	Towards the end of the conversation, missing information may cause callers not to follow the call‐taker's advice, often due to poor interaction or misunderstanding.
	Third‐party callers introduce epistemic asymmetry.	Checking for understanding is essential to ensure information is correctly received.
	Asymmetry in goal orientation is overarching to all the aforementioned asymmetries.	Call‐takers should be vigilant when repeated callers, reassessing symptoms rather than relying on previous conversations, as repeated contact increases the likelihood of and urgent conditions.
		Regarding turn design, callers describing symptoms as non‐urgent can lead to undertriage as urgency allocation likely to be strongly influenced by callers.
		Regarding overall structure, call duration does not correlate with urgency risk.

4.1. Turn‐Taking Organisation: How Are Turns Divided Between Speakers, Who Has the Rights to Speak, and How Are Turns Taken?

Multiple studies reveal that call‐takers in triage conversations typically take the lead in initiating interactions and managing turn‐taking. As a result, call‐takers (1) have more turns at talk, (2) control the agenda of the conversation and (3) commonly invite callers for extended turns, rather than callers taking extended turns on their own initiative. First, because call‐takers almost always deliver the first part of a question‐answer turn taking sequence and open and close the conversation, they tend to have slightly more turns than callers (Ernesäter et al. 2012; Young, Rochon, and Mihailidis 2016). Second, besides having more turns, call‐takers control the agenda during emergency calls (Kevoe‐Feldman and Iversen 2022; Spek et al. 2023). Helpline calls differ in this respect; call‐takers are supposed to wait for callers to set the topic and agenda (Kevoe‐Feldman and Iversen 2022). In contrast, many emergency calls conversations follow the ‘question‐answer sequences’, with questions often derived from the CDSS, asked by the call‐taker, answered by the caller, and followed by a response or new question from the call‐taker (Spek et al. 2023). Although call‐takers generally control the conversation agenda, they can deviate from the routine way of asking questions (Kevoe‐Feldman and Iversen 2022; Leppänen 2010). Call‐takers can do so by inviting the caller to take an extended turn that is not directly part of the question‐answer sequence at hand. Such deviation may reduce resistance when the call‐taker asks the next question, increasing the chance of receiving an answer (Kevoe‐Feldman and Iversen 2022). Callers can also deviate, for instance, by asking permission to ask a question, indicating that it is not ‘normal’ for callers to do so. Callers can also claim interactional space by providing background information until they state the question they are permitted to ask (Leppänen 2010). Thus, although call‐takers have a prominent role in turn‐taking organisation, situational constraints or requirements can instigate deviations from this pattern.

The length of a turn and the number of turns can also deviate from the regular pattern depending on the caller type, call‐taker type, and conversation phase (Morgan and Muskett 2020; Murdoch et al. 2014; Young, Rochon, and Mihailidis 2016). Regarding the conversation phase, in the last phase of the call, the caller's turns tend to be shorter and less frequent (Morgan and Muskett 2020). Comparing care providers, such as homecare nurses, calling on behalf of a patient to patients themselves, care providers had significantly longer ‘first turn length in words’ and seemed to have longer but fewer turns than patients (Young, Rochon, and Mihailidis 2016). When comparing telephone triage led by nurses to that conducted by GPs, the average call duration was similar. However, nurses using a CDSS asked patients almost three times as many questions as GPs (Murdoch et al. 2014). This difference may arise not only from the use of CDSS but also from variations in professional identity and communication style, where nurses may adopt a more detailed questioning approach compared to GPs (Anderson, Birks, and Adamson 2020; Graversen et al. 2020).

4.2. Overall Structure: How Is a Conversation in This Context Commonly Structured?

The overall structure of telephone triage conversations generally comprises several phases (Allen 2021; Ericsson et al. 2019; Erkelens et al. 2021; Hayashi 2019; Leydon, Ekberg, and Drew 2013; Mellinger 1994; Morgan and Muskett 2020; Palma et al. 2014; Riou, Ball, O'Halloran, et al. 2018; Spek et al. 2023). These phases have developed over time as a common practice among call‐takers and are partly convention. Early in the call, call‐takers obtain the caller's contact details in case the connection is unintendedly disconnected. This is often followed by gaining more insight into the symptoms and ends with sharing the action plan or next steps (Allen 2021; Ericsson et al. 2019; Erkelens et al. 2021; Hayashi 2019; Morgan and Muskett 2020; Riou, Ball, O'Halloran, et al. 2018; Spek et al. 2023). Triage conversations may evolve into a counselling conversation when the outcome of the triage requires telephone advice only, thus adding an additional phase (Spek et al. 2023).

The opening phase of the conversation receives particular attention (Leydon, Ekberg, and Drew 2013; Mellinger 1994; Morgan and Muskett 2020; Palma et al. 2014; Riou, Ball, O'Halloran, et al. 2018). Its importance is evident, as a smooth opening generally positively affects the triage conversation (Ericsson et al. 2019; Leydon, Ekberg, and Drew 2013). This part of the conversation is notable because the structure at the beginning of the call differs from an ordinary telephone conversation, lacking mutual greetings. Typically, callers state the reason for their call early in the interaction, whereas in triage conversations, it is expected to be discussed later, usually after providing contact details (Morgan and Muskett 2020). Callers often attempt to ‘hijack’ the dispatch protocol by pre‐empting the reason for the call early in the interaction (Riou, Ball, O'Halloran, et al. 2018).

When triage conversations involve more than one call‐taker, such as when a healthcare professional with different expertise is needed, the opening by the second call‐taker is crucial for a smooth transition (Leydon, Ekberg, and Drew 2013). This should include recognition, (self‐)identification, formulation of the reason for calling, and a request to elaborate (Leydon, Ekberg, and Drew 2013). A scripted call opening with all relevant aspects, including known information and what is expected from the caller now, would thus be useful to ensure a smooth transition (Leydon, Ekberg, and Drew 2013).

The nature of interaction between callers and call‐takers can differ between phases. In the opening phase, the caller's symptoms and context could be described in detail and structured, or vaguely and unstructured. In the orienting phase, the four types of interaction—distinctive, reasoning, indecisive and irrational—determine the progress of the conversation (Ericsson et al. 2019). In the closing phase, the type of interaction determines whether the situation is resolved productively or not (Ericsson et al. 2019). Distinctive interaction is characterised by concise and driven communication. Reasoning interaction involves collaborative reflection to achieve mutual understanding. Indecisive interaction is marked by ambiguous and vague dialogue, in terms of the precision of communication. Irrational interaction is characterised by incoherent connection between the call‐taker and the caller.

The structure of telephone triage is strongly influenced by the CDSS, or the dispatch protocol used (Allen 2021; Ericsson et al. 2019; Erkelens et al. 2021; Hayashi 2019; Leydon, Ekberg, and Drew 2013; Morgan and Muskett 2020; Murdoch et al. 2014, 2015; Palma et al. 2014; Riou, Ball, O'Halloran, et al. 2018; Spek et al. 2023). The type and sequence of questions differ in triage conversation with and without the CDSS (Murdoch et al. 2014, 2015). Problems arise when the protocol or the required categorization in the system does not allow enough room for entering information that patients provide (Allen 2021; Hayashi 2019; Morgan and Muskett 2020; Murdoch et al. 2014, 2015; Palma et al. 2014; Riou, Ball, O'Halloran, et al. 2018). Call‐takers, however, have some leeway to deviate from questions in the CDSS and the order in which they are asked. Yet, when high urgency becomes evident early in the call, they rarely deviate from questions generated by the CDSS (Allen 2021; Murdoch et al. 2014; Spek et al. 2023).

4.3. Sequence Organisation: What Are the Actions Done in Talk, and How Do These Relate to Each Other Sequentially?

The review identifies three key activities regarding sequence organisation: (1) problem presentation, (2) granting the caller's request for help, and (3) balancing between actions expected from the CDSS, and actions that facilitate the conversation.

A key activity for callers is presenting the problem. The initial presentation sets the context for the rest of the conversation, including determining urgency and guiding the questions that should be asked and influences the decisions made by the call‐taker (Booker et al. 2018; Leppänen 2010). Callers are generally expected to present their medical issue immediately after the call's opening (Leppänen 2010). If callers do not present their problem, nurses explicitly ask for the medical reason for calling. Callers might briefly describe their medical issue before directly requesting a consultation with a GP when calling primary care (Leppänen 2010). This differs from calling ambulance services, where the caller already indicates an urgent need for medical attention, presenting a ‘doctorable’ problem without making a direct request (Booker et al. 2018).

The key activity for call‐takers is addressing the caller's request for help. Acknowledging the caller's situation can help in either deferring or granting their request. Granting the request is preferred as it aligns with the caller's needs and increases patient satisfaction (Mellinger 1994; Weatherall and Grattan 2023). Not granting the request is less preferred but can occur due to practical issues, such as an order that cannot be implemented (e.g., injecting an infusion by an on‐site nurse when no veins are visible). Presenting added information that questions the previous directive or the paramedic's comfort with it may lead to its withdrawal and replacement with another directive (Mellinger 1994). Other reasons for not granting the request could be that a nurse should perform the action rather than a medical receptionist, or the issue requiring more urgent attention than initially requested (Weatherall and Grattan 2023).

Call‐takers must balance actions expected by the CDSS with those that facilitate the conversation, such as reassuring the caller or asking further questions (Ericsson et al. 2019; Ernesäter et al. 2012; Mellinger 1994; Murdoch et al. 2014; Spek et al. 2023; Young, Rochon, and Mihailidis 2016). Nurse‐led triage differs from GP‐led triage as nurses, focused on the CDSS, may pay less attention to the patient's explanation or perspective, often triaging more in a checklist style (Murdoch et al. 2014). Overlooking these perspectives may result from multitasking, such as following the CDSS's structure, formulating questions, interpreting answers, asking about the patient's location, entering responses into the computer, and maintaining contact with the caller (Mellinger 1994; Weatherall and Grattan 2023). However, attention to the caller's perspective is crucial. One approach is asking the ‘what matters to you’‐question (Spek et al. 2023). Other methods include using distinctive or reasoning interaction (see Section 4.2) or exploring why the caller believes the requested care is needed (Ericsson et al. 2019; Ernesäter et al. 2012).

4.4. Turn Design: How Are Turns Formatted?

Turn design differences can sometimes be linked to who is calling or taking the call (Mellinger 1994; Murdoch et al. 2014; Young, Rochon, and Mihailidis 2016). Compared to care provider callers, older adults tend to have shorter turns, speak more slowly, exhibit more linguistic disfluencies, and use one‐word utterances more frequently (Young, Rochon, and Mihailidis 2016). In contrast, healthcare provider callers typically have longer turns (Young, Rochon, and Mihailidis 2016). Caller characteristics also vary, resulting in different symptom presentations (Allen 2021; Ericsson et al. 2019). For example, callers may use specific terms like ‘weird’ or ‘disgusting’ to indicate the need for acute care, or use a qualified yes‐answer to a breathing question to indicate the need for acute care as well (Ericsson et al. 2019; Riou, Ball, Williams, et al. 2018; Riou et al. 2017). Language mismatches can occur when callers speak different languages, but these are often resolved smoothly when call‐takers adapt to the caller's preferred language (Penn, Watermeyer, and Nattrass 2017).

As noted in Section 4.3, nurse‐led telephone triage using a CDSS differs from GP‐led telephone triage without such systems (Murdoch et al. 2014). Nurses tend to use a checklist style, focusing on confirming symptoms for urgency assessment and risk management. In contrast, GPs consider the patient's own explanation, broader medical history, and familiar consultation behaviour.

Caller turns often take the form of reports and narratives (Riou et al. 2017). Report responses are concise and factual, while narrative responses provide broader context and details. Report responses are preferred in triage as they result in faster dispatch times, though efficient responses from the caller's perspective can be problematic for call‐takers aiming for quick triage (Riou et al. 2017). At the same time, paying attention to concerns is seen as something that takes time. (Spek et al. 2023).

Emotions of fear or anxiety influence turn design in triage conversations (Allen 2021; Ernesäter et al. 2012; Kevoe‐Feldman and Iversen 2022; Leppänen 2010; Spek et al. 2023; Weatherall and Grattan 2023). In emergency calls, obtaining information often takes precedence over addressing emotional expressions, potentially missing opportunities to help callers (Kevoe‐Feldman and Iversen 2022). However, briefly addressing concerns before returning to the main points of the dispatch protocol ensures a smooth conversation flow (Spek et al. 2023). Concerns are more frequently expressed when a call‐taker explicitly advises consulting a doctor, such as directly recommending a doctor consultation than when a consultation is more subtly or indirectly, such as by saying ‘let me have a look’, typing audibly on their computer, or asking for a date of birth. These actions may implicitly signal to the caller that an appointment is being arranged (Leppänen 2010). Explicit expressions occur particularly in cases involving children or potentially serious medical problems, often prompted by the call‐taker.

Call‐takers' turn design is influenced by the computerised CDSS in three ways: (1) question design, (2) returning to the reason for calling, and (3) responding to additional information.

First, the CDSS influences question design by dictating the type and grammatical tense of questions (Booker et al. 2018; Erkelens et al. 2021; Ernesäter et al. 2012; Leppänen 2010; Murdoch et al. 2014; Riou et al. 2017; Young, Rochon, and Mihailidis 2016). Questions are often treated as a checklist rather than a tool to clarify the caller's specific problems (Murdoch et al. 2014). Alternative question designs, such as either‐or‐questions, are asked by call‐takers because they appear in that form in the scripted CDSS, while polar questions, which can be answered with a simple ‘yes’ or ‘no’, are easier for callers to answer, but require call‐takers to deviate from the question form of the CDSS (Booker et al. 2018; Erkelens et al. 2021; Ernesäter et al. 2012). However, polar questions can also be problematic if they imply a preferred ‘no problem’ answer, limiting the interactional space to not give ‘no problem’ answer (Murdoch et al. 2014). For example, in the paper from Murdoch et al. (2014), the call‐taker asks, ‘Has she been vomiting at all?’. This question is designed to elicit a ‘no problem’ response, as the use of the negative polarity item ‘at all’ presupposes the absence of vomiting. The preferred response would be a simple ‘no’ to indicate that there is no problem. If the patient would have been vomiting, the caller would need to do more work to ‘go against’ the inferred ‘no’‐response. Murdoch et al. show that more than half of the polar questions asked by call‐takers in their study includes such preference (Murdoch et al. 2014). While call‐takers typically use closed questions, open‐ended questions can provide more insight into caller's perspectives, but could be problematic because callers are given little guidance on what answer to give which may cause delays (Erkelens et al. 2021; Ernesäter et al. 2012).

The tense of questions also affects callers' responses (Riou et al. 2017). Replacing the scripted ‘what happened’ (simple past tense) with the non‐scripted ‘what has happened’ (present perfect tense) increases the likelihood report‐style answer rather than a narrative one. Callers' narratives are typically structured in two parts: first an introduction to the problem, followed by details of the problem from the beginning and then chronologically to the present (Leppänen 2010). While narratives are time‐consuming, they allow callers to share their own account of illness and should not be overlooked (Erkelens et al. 2021; Leppänen 2010; Riou et al. 2017).

Second, the CDSS affects how call‐takers return to the reason for the call when it was discussed earlier. Callers usually state this reason at the beginning of the conversation, but the CDSS asks for it later, which can be problematic (Booker et al. 2018; Ernesäter et al. 2012; Riou, Ball, O'Halloran, et al. 2018; Riou et al. 2017). Using a pre‐emption to refer to earlier information avoids repetition, checks understanding, and indicates the need for additional information but is rarely done due to system constraints (Riou, Ball, O'Halloran, et al. 2018). This approach is also effective in repeated contacts to check for changes rather than continuing with assumptions made in the previous triage conversation (Ernesäter et al. 2012).

Third, call‐takers' responses to additional information provided by callers are also influenced by the CDSS. Callers may add extra information when answering a question from a call‐taker, but this extra information is often overlooked as call‐takers focus on entering answers in the CDSS (Morgan and Muskett 2020; Riou, Ball, Williams, et al. 2018; Spek et al. 2023). ‘Over‐answering’ the question reflects a social norm where it is typical to clarify or qualify responses to polar questions to show that you are a knowledgeable answerer (Morgan and Muskett 2020). However, this can complicate the triage process. Additional information may be unrelated or only loosely related to the question, for example, when it is an expression of concerns. These concerns are important for the caller to discuss but often unrelated to the previous question (Spek et al. 2023). Qualified yes‐answers, which are affirmative responses accompanied by additional information or conditions, can be misleading in triage contexts (Riou, Ball, Williams, et al. 2018). For example, if a caller answers ‘yes’ but adds that they are experiencing severe discomfort or new symptoms, this additional information might actually indicate a more serious issue than initially suggested by a simple ‘yes’. In some cases, what might seem like an affirmative answer could effectively align more closely with a ‘no’ or indicate a different level of urgency. Consequently, call‐takers should be cautious with such responses, as they may require a more nuanced evaluation or a different, potentially more urgent, response. Thus, additional information can potentially affect the triage process (Morgan and Muskett 2020).

Location information that cannot be entered into the CDSS poses another difficulty, requiring repetition, spelling out or explanation from a co‐present person to resolve (Nattrass et al. 2017).

4.5. Lexical Choice: Within Turns, What Word Choices Are Made?

Review findings on lexical choice in triage conversations are limited. Some studies examine how call‐takers rephrase specific words uttered by the caller (Hayashi 2019; Nattrass et al. 2017; Young, Rochon, and Mihailidis 2016). Rephrasing can elicit more accurate information by the call‐taker, such as a specific location (Nattrass et al. 2017). However, rephrasing with incorrect associations can be risky and potentially lead to incorrect urgency allocation. For instance, walking implies movement, but being able to move does not necessarily mean that the caller can walk (Hayashi 2019). The meaning given to a word used by the caller can vary depending on the context of the call (Hayashi 2019). Hayashi describes a situation where a word used at the beginning of the call was used to select the entrance complaint in the CDSS, but the same word could be noted as a symptom at the end of the call. Additionally, using two words interchangeably by call‐takers, such as ambulance (vehicle) and paramedic (personnel of the ambulance), can be problematic for callers due to their different associations (Young, Rochon, and Mihailidis 2016).

4.6. Epistemological and Other Forms of Asymmetry: How Do Participants Construct Asymmetry in Knowledge and Participation?

In this section, we address the concept of asymmetry in conversations, particularly focusing on how imbalances in knowledge and participation impact triage interactions.

Epistemological asymmetry refers to the imbalance in knowledge between participants. For instance, in a triage call, the call‐taker typically has more medical knowledge than the caller. Conversely, the caller often possesses more detailed information about their own symptoms and current situation. This imbalance can affect how information is exchanged and interpreted.

Interactional asymmetry involves differences in how participants engage in the conversation. This includes variations in the level of control or influence each participant has over the interaction. For instance, a call‐taker may control the pace and content of the conversation by directing questions and initiating the flow of information, while the caller provides responses based on their own perspective. Despite the call‐taker's expertise, the caller's detailed knowledge about their symptoms and situation is crucial for accurate triage.

Included papers indicate various types of asymmetry between caller and call‐taker in triage calls, affecting the call and its outcome: epistemic asymmetry, interactional asymmetry, differences in knowledge of care organisation and triage processes, the CDSS, and asymmetry related to third‐party callers (Allen 2021; Booker et al. 2018; Ericsson et al. 2019; Ernesäter et al. 2012; Kevoe‐Feldman and Iversen 2022; Leppänen 2010; Leydon, Ekberg, and Drew 2013; Morgan and Muskett 2020; Murdoch et al. 2014, 2015; Nattrass et al. 2017; Riou, Ball, O'Halloran, et al. 2018; Riou, Ball, Williams, et al. 2018; Riou et al. 2017; Weatherall and Grattan 2023).

First, callers often treat call‐takers as having medical expertise, and only cautiously suggest a possible diagnoses themselves, reinforcing medical knowledge asymmetry (Booker et al. 2018; Leppänen 2010; Riou, Ball, Williams, et al. 2018). When unsure which symptoms, observations or features are important, callers tend to provide elaborate accounts of events without linking them to diagnoses or medical consequences, thus positioning themselves as knowledgeable on their own situation but leaving the problem unclear to call‐takers (Booker et al. 2018). This epistemic asymmetry involves callers presenting their problems to call‐takers by emphasising worsening symptoms, which may affect the call‐taker's assessment of the urgency and necessary actions (Leppänen 2010). While the call‐taker retains the ultimate decision‐making power, the way callers frame their issues can influence the interaction and potentially affect the call‐taker's perception of the situation.

Second, callers sometimes treat call‐takers, especially nurses, as not the experts they need, requesting a GP consultation directly (Leppänen 2010). Yet, call‐takers assert their authority by taking the interactional lead, for example, asking necessary questions according to the triage protocol. This interactional initiative is crucial, as indecisive interactions may result in failing to lead the conversation and maintain authority (Ericsson et al. 2019).

Third, asymmetry in knowledge of care organisation and triage processes affects interactions (Booker et al. 2018; Ericsson et al. 2019; Kevoe‐Feldman and Iversen 2022; Leppänen 2010; Leydon, Ekberg, and Drew 2013; Riou, Ball, O'Halloran, et al. 2018; Weatherall and Grattan 2023). Callers may contact the wrong organisation, such as a helpline when an emergency department was more appropriate, or an ambulance service when GP care was more appropriate (Booker et al. 2018; Kevoe‐Feldman and Iversen 2022). This might cause goal misalignment, creating tension where call‐takers could miss opportunities to assist callers, while balancing the risk between callers losing faith in the helpline and the risk of callers being physically harmed and not having direct assistance from the emergency department but solely from a helpline (Kevoe‐Feldman and Iversen 2022). It could also result in call‐takers prioritising the caller's description of the situation rather than effectively gathering necessary information to decide on the level of urgency in providing emergency services. Call‐takers need to be aware of this tension and navigate these differential expectations, acting beyond the boundaries of their domain to maintain call progress without losing the caller (Kevoe‐Feldman and Iversen 2022). When emergency call‐takers cross their institutional boundary by engaging in emotional support talk, callers are more receptive to accepting emergency help. Explaining the triage process and providing necessary information can mitigate these issues (Weatherall and Grattan 2023). Furthermore, it is important for the healthcare professional to make explicit what information was requested from the previous healthcare professional when switching from one healthcare professional to another (Leydon, Ekberg, and Drew 2013).

Fourth, the CDSS introduces asymmetry as callers are often unaware of its presence (Allen 2021; Morgan and Muskett 2020; Murdoch et al. 2014, 2015; Nattrass et al. 2017; Riou, Ball, O'Halloran, et al. 2018). Call‐takers juggle interaction and CDSS navigation, potentially overlooking concerns (Murdoch et al. 2015). Callers may struggle with the order of information or to understand the CDSS's questions (Morgan and Muskett 2020; Murdoch et al. 2015; Nattrass et al. 2017; Riou, Ball, O'Halloran, et al. 2018). Healthcare professionals calling instead of patients may overload call‐takers with unordered information, which cannot be entered in the system and thus complicating interactions (Allen 2021). Greater flexibility and community awareness of the CDSS could help (Allen 2021; Murdoch et al. 2014).

Fifth, third‐party callers introduce epistemic asymmetry (Allen 2021; Booker et al. 2018; Ernesäter et al. 2012; Riou, Ball, O'Halloran, et al. 2018). The fact that malpractice claims often involve triage conversations with third‐party callers highlights how problematic these interactions can be. These third‐party callers may lack knowledge of the patients themselves or the patients' situations, or they may not always be present with the patient (Allen 2021; Booker et al. 2018; Ernesäter et al. 2012; Riou, Ball, O'Halloran, et al. 2018; Young, Rochon, and Mihailidis 2016). To compensate for this, third‐party tend to provide detailed but time‐consuming descriptions (Booker et al. 2018). Therefore, call‐takers should ask questions aligned with the caller's own epistemic domain, which refers to the caller's knowledge and understanding of the situation, or attempt to speak directly to the patient (Booker et al. 2018; Ernesäter et al. 2012).

Asymmetry in goal orientation is overarching to all the aforementioned asymmetries. Misalignment of agenda or goals between the caller and call‐taker causes interactional difficulties (Booker et al. 2018; Ernesäter et al. 2012; Kevoe‐Feldman and Iversen 2022; Riou et al. 2017). Call‐takers aim to determine urgency and provide the right help quickly, while callers often have broader goals in addition to getting help, including describing the situation, providing their perspective, and eventually addressing emotions. This problem is exacerbated with third‐party callers who also have their own goals and agendas (Booker et al. 2018; Ernesäter et al. 2012). Non‐alignment can lead to tension and result in a troubled exchange (Booker et al. 2018; Kevoe‐Feldman and Iversen 2022).

4.7. Influence on Triage Outcome

Although all linguistic features have been described in the included articles, the direct influence of these elements on the accuracy of urgency determination has not been detailed for all linguistic features. At the same time, it is plausible that a smoother call benefits urgency determination, as the call‐taker can receive the necessary information more quickly, and the caller can better understand and apply the advice given. Thus, although not explicitly studied in all papers, these features may influence the accuracy of triage outcomes.

Regarding risk assessment, certain linguistic features such as turn‐taking organisation, turn design, lexical choice and overall structure were investigated for their potential to indicate urgency. For turn‐taking organisation, the risk appears to depend on the caller type, making it hard to identify a unified pattern (Young, Rochon, and Mihailidis 2016). Furthermore, the duration of speech intervals and the average turn length did not differ between urgent and non‐urgent diagnoses (Tamminen et al. 2020; Young, Rochon, and Mihailidis 2016). For turn design, metrics such as average turn length, linguistic disfluencies percentages, number of responses, questions, and words in the first sequence did not differ across risk groups, though high‐risk calls had fewer total number of words (Young, Rochon, and Mihailidis 2016). For lexical choice, predicting medical outcomes based on spontaneous trigger words from third‐party callers proved difficult (Tamminen et al. 2020).

Issues with overall structure could affect risk assessment (Ernesäter et al. 2012; Morgan and Muskett 2020; Murdoch et al. 2014; Palma et al. 2014; Young, Rochon, and Mihailidis 2016). Missing elements of the dispatch protocol may lead to inadequate urgency determination (Palma et al. 2014). Missing information influence messages coherence and comprehensibility and could therefore negatively influence triage outcome (Allen 2021; Ericsson et al. 2019; Ernesäter et al. 2012; Kevoe‐Feldman and Iversen 2022; Leppänen 2010; Morgan and Muskett 2020; Spek et al. 2023; Weatherall and Grattan 2023). Towards the end of the conversation, missing information may cause callers not to follow the call‐taker's advice, often due to poor interaction or misunderstanding (Ericsson et al. 2019). An example could be a safety advice not effectively communicated or misunderstood such as the need to contact again if symptoms worsen (Morgan and Muskett 2020). Customising the CDSS could help, such as adding a thoughtful message in the CDSS to alert the caller to the seriousness of the situation when the call‐taker was struggling to be heard or adding specific pronounced explanations to motivate the caller to follow the advice (Ericsson et al. 2019). For call‐takers, it is important to take sufficient time to listen to the caller's answers, especially for crucial questions, to recognise urgent circumstances (Riou, Ball, Williams, et al. 2018).

Checking for understanding is essential to ensure information is correctly received (Ernesäter et al. 2012; Mellinger 1994; Morgan and Muskett 2020; Riou, Ball, O'Halloran, et al. 2018). Call‐takers more frequently checked their own understanding of callers than they did the callers' understanding of the advice given (Ernesäter et al. 2012). The latter almost never happened. Such checks could be achieved through repetition of the caller's previous statements (Riou, Ball, O'Halloran, et al. 2018). Additionally, call‐takers should be vigilant with repeated callers, reassessing symptoms rather than relying on previous conversations, as repeated contact increases the likelihood of urgent conditions (Ernesäter et al. 2012). Regarding turn design, callers describing symptoms as non‐urgent can lead to undertriage, as urgency allocation is likely to be strongly influenced by callers (Palma et al. 2014). Regarding overall structure, call duration does not correlate with urgency risk (Young, Rochon, and Mihailidis 2016).

We should realise that triage conversations should be as short as possible to ensure that callers in need of urgent care receive this as soon as possible. At the same time, it is important that all necessary information is obtained to ensure accurate urgency allocation and therewith avoid both overtriage and undertriage.

5. Discussion

Our review of the role of language in remote healthcare triage leads to two main conclusions. First, the significance of language extends beyond the mere content of what is said. Current audits and quality assessments typically focus on the content call‐taker's content and whether they follow triage protocols (Morgan and Muskett 2020; Riou, Ball, O'Halloran, et al. 2018). However, evidence shows that deviating from the protocol can sometimes enhance the flow of the call (Riou et al. 2018a; Riou et al. 2017). Evaluating triage conversations based solely on the content of the call‐taker assumes callers are passive participants, whereas they are actually active contributors to the interaction (Morgan and Muskett 2020). Thus, incorporating an interactional perspective in audits and quality assessments could offer a more complete understanding of triage effectiveness.

Second, CDSS significantly influences remote triage. CDSS primarily structures the triage process but can create epistemological differences (differences in knowledge about the protocol, or the interaction between callers and call‐takers); callers may be unaware of its impact. It may also disrupt sequence organisation when callers discuss topics beyond the CDSS's scope and it shapes turn design. While CDSS can improve clinical management and patient outcomes, according to Fernandes et al., focussing to much on the system may neglect the sequential nature inherent in triage conversations, leading to interactional difficulties, and potentially jeopardising triage accuracy (Fernandes et al. 2020). Rather than eliminating CDSS, we suggest increasing its flexibility to accommodate the dynamic nature of interactions, allowing call‐takers to input information from patients into the system as it emerges. This suggestion is in line with psychological research that highlights the importance of adaptability in various communication contexts dealing with protocolized interaction (Barlow et al. 2023; Pitts and Harwood 2015; Pitts and Giles 2008). Keeping a focus on the interactive element of language also applies to the development of chatbots. Chatbots are digital or telephone computerised questionnaires without the patient speaking to a person. Due to staffing and time constraints, these chatbots are described by some people as the future of remote triage processes. However, we need to realise that these bots should not just be just symptom checkers but should take into account the interactional context of a conversation and pay attention to both the what and the how of utterances (Ben‐Shabat et al. 2022; Sarbay et al. 2023).

In addition to these two overarching conclusions, our review identified two notable findings: (1) the opening of the conversation matters, and (2) epistemological and other forms of asymmetry are interrelated with other linguistic features. First, effective introductions by call‐takers are crucial, especially when the caller has been transferred from another call‐taker. Effective introductions should include recognition, (self‐)identification, the reason for calling, and a request for elaboration (Leydon, Ekberg, and Drew 2013). Research has shown that clearer opening questions improve caller responses (Koole and Verberg 2017). For example, Koole and Verberg showed that changing the question from ‘Emergency call 112, who do you want to speak to?’ to ‘Emergency call 112, do you want police, fire brigade or ambulance?’ has proven to be more understandable for callers. Second, the linguistic feature epistemological and other forms of asymmetry is interconnected with other linguistic features, highlighting the fragile nature of remote healthcare triage interactions.

The outcomes of the included studies are rarely linked to their impact on triage outcomes. Some studies even find it challenging to reflect direct effects on triage outcomes (Booker et al. 2018; Ernesäter et al. 2012; Spek et al. 2023). This may be due to the detailed, context‐specific nature of conversation analytic studies (most of the included studies), which often have smaller sample sizes due to feasibility. As a result, it is often not possible to combine it with studies on the effect on triage outcomes. Moreover, conversational analysts are convinced that everything is context commanded so no guarantees can be given about outcomes. Thus, it is difficult to link specific language phenomena to conversation outcomes causally. However, capturing detail‐level interactional aspects in (large) quantitative studies is also challenging. Therefore, a multidisciplinary approach combining different research methods is crucial for improving communicative aspects in remote triage.

A recent review on prehospital and emergency care communication overlaps with our findings, despite our more focused scope allowing us to delve more extensively and in detail into the described information (Riou 2024). Both reviews emphasise the importance of call opening, obtaining patient's location, misalignment, asymmetry, emotions, and language barriers. Riou highlights the potential of AI‐driven triage, noting its superior recognition of out‐of‐hospital cardiac arrests compared to human call‐takers (Blomberg et al. 2019; Riou 2024). However, our review underscores that triage involves more than a checklist; it requires attention to interactional aspects that current AI may not fully address. Thus, caution is warranted before fully adopting AI‐driven triage systems, ensuring they consider both the content and the nuances of communication such as how and when something is said.

Although conversation analytic research is inherently context‐dependent, the findings of this review are likely broadly applicable beyond their specific setting. For instance, while language mismatches have been a prominent issue in the South African context due to its diverse linguistic landscape, such challenges are increasingly relevant in other countries as well. The growing trends of migration and multiculturalism are making these issues more pertinent globally. Thus, even though language mismatches initially appeared specific to South Africa, their implications are becoming more universally significant as societies around the world become more diverse.

5.1. Strengths and Limitations

This review has several strengths. First, our multidisciplinary research team, comprising experts in medicine, education, and linguistics, provided a comprehensive analysis of linguistic features in triage conversations. This collaborative approach aligns with recent findings that highlight the benefits of collaboration between conversation analysts and medical researchers (Riou 2024). By using a framework common in linguistics, particularly conversation analysis, we were able to classify results according to key interactional linguistic features (Drew and Heritage 1992; Heritage and Clayman 2011). This framework, though less familiar to clinicians, allowed us to explore the results from a broad perspective. Consequently, we identified factors influencing triage outcomes, such as miscommunication, and highlighted interactional aspects not (yet) linked to outcomes, beyond what is typically considered in quality assessments.

However, there are some limitations to note. First, we included only articles published in Dutch or English, and while no articles in other languages were found, search terms in other languages might have identified additional articles. Despite this, our broad search strategy minimises the likelihood of missing relevant articles. A recent review on prehospital and emergency care communication included a similar number of articles, though analysed with a different focus (Riou 2024). Unlike that review, we examined language as a broad phenomenon and incorporated various research methods, not limiting our analysis to conversation analysis alone. Third, classifying results into linguistic features was challenging due to their interrelated nature, with results sometimes fitting into multiple linguistic features. Nonetheless, these features served to ensure a comprehensive examination of linguistic elements, rather than being an end goal in themselves. Our qualitative approach focused on substantive content rather than precise quantification, so classification challenges did not prevent us from identifying overarching themes. Thus, our objective of highlighting these themes was not impeded by difficulties in categorization.

5.2. Conclusion

By analysing studies of triage conversations through linguistic features, we demonstrated that remote triage is an interactive process where both content, manner and form of speech are crucial. This analysis shows that language encompasses more than what is said verbatim, which is the current focus of audits and quality assessments. Additionally, these linguistic features can be consequential for conversation outcomes. These insights are vital for designing future studies and shaping education in this field.

5.3. Relevance to Clinical Practice

As described in our results, we identified several important aspects relevant to clinical practice. These aspects apply to callers, call‐takers, quality assessors, and the CDSS. While informing the general population about the need and process of adequate triage would be beneficial, it is more feasible and effective to train call‐takers to guide callers during the time of calling, helping them know what to expect.

For call‐takers, our study identified several trainables for education and clinical practice: (1) scripted call openings are necessary when switching from a first call‐taker to a second, covering known information and expectations to ensure a smooth transition, (2) avoid alternative and either‐or‐questions, using polar questions instead, (3) ask polar questions in a way that does not prefer a ‘no problem’ answer, thus allowing interactional space to give another answer, (4) ask ‘what has happened’ instead of ‘what happened’, (5) be alert to qualified yes‐answers, as they may indicate a ‘no’, (6) ask all elements of the dispatch protocol, (7) be especially attentive to repeated contacts, as they indicate a higher likelihood of urgent conditions, and (7) check for understanding of both received information and given advice.

The above trainables are skills, but there are also effective learning objectives that influence call‐takers' mindsets (Booker et al. 2018; Erkelens et al. 2021; Riou, Ball, Williams, et al. 2018; Spek et al. 2023; Wei, Saab, and Admiraal 2021). Call‐takers should realise that information beyond the CDSS scope can be crucial for conversation progression and triage outcomes.

We recommend incorporating all relevant linguistic aspects in quality assessments and audits, as they can influence conversation outcomes. Our review showed that triage conversations are interactional: callers' actions affect call‐takers' actions and vice versa. Thus, policy makers should adopt this broader perspective in quality assessments rather than focusing only on what is literally said by the call‐taker. Additionally, since many linguistic features are influenced by the CDSS, developers should incorporate flexibility into the system. This flexibility allows call‐takers to focus on interactional aspects, making conversations flow more naturally and addressing the callers' needs, rather than using the CDSS merely as a checklist (Murdoch et al. 2014, 2015). This influence on multiple linguistic features also limits the applicability of (digital) triage via (AI‐driven) chatbots or checklists, which are increasingly being considered due to staff and time constraints.

Author Contributions

E.d.G. is the lead investigator who conceived the research idea and methodology. M.S., M.v.B., D.C.A.E. and E.d.G. did the search and selection of the articles. M.S., M.v.B., D.C.A.E. and E.d.G. performed the analyses. M.S., wrote the original draft of the manuscript. M.v.B., D.C.A.E., F.H.R., R.P.V., D.L.Z. and E.d.G. critically reviewed the analytic process and manuscript. All approved the final version of the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Peer Review

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer‐review/10.1111/jan.16528.

Supporting information

Data S1.

Data Availability Statement

The authors have nothing to report.