‘Knowledge exchange ’ workshops to optimise development of a risk prediction tool to assist conveyance decisions for suspected seizures – Part of the Risk of ADverse Outcomes after a Suspected Seizure (RADOSS) project

Purpose: Suspected seizures present challenges for ambulance services, with paramedics reporting uncertainty over whether or not to convey individuals to emergency departments. The Risk of ADverse Outcomes after a Suspected Seizure (RADOSS) project aims to address this by developing a risk assessment tool utilizing structured patient care record and dispatch data. It proposes a tool that would provide estimates of an individual ’ s likelihood of death and/or recontact with emergency care within 3 days if conveyed compared to not conveyed, and the likelihood of an ’ avoidable attendance ’ occurring if conveyed. Knowledge Exchange workshops engaged stakeholders to resolve key design uncertainties before model derivation. Method: Six workshops involved 26 service users and their significant others (epilepsy or nonepileptic attack disorder), and 25 urgent and emergency care clinicians from different English ambulance regions. Utilizing Nominal Group Techniques, participants shared views of the proposed tool, benefits and concerns, suggested predictors, critiqued outcome measures


Introduction
Suspected seizures are a common presentation to English ambulance services, [1][2][3] and paramedics say it can be challenging to decide which patients presenting with a suspected seizure should be conveyed to hospital.[4][5][6][7] RADOSS [8] aims to address this knowledge gap by developing a risk prediction tool to assist paramedics.In this paper, we report on findings from a series of Knowledge Exchange (KE) workshops attended by stakeholders to optimize tool development.
While suspected seizures can be dramatic and life-threatening, most cases attended by the ambulance service are uncomplicated manifestations of established conditions, mainly epilepsy and non-epileptic attack disorder (NEAD).[1,2,9] Conditions such as syncope, alcohol withdrawal, head injury, and hypoglycaemia account for the remaining minority of suspected seizure incidents.[2].
Despite the typically uncomplicated nature of suspected seizure incidents, [3] ~ 70 % are conveyed to emergency departments (EDs) [1,2,9] and can result in 'avoidable attendances'.Avoidable attendances, as defined by England's National Health Service using O'Keeffe et al.'s criteria [10] (Table 1), are undesirable as they can increase demand on ambulance services, contribute to overcrowding in EDs, cause patient harm and are costly.[11 12,13 14].
Decisions whether to convey a patient to hospital can be influenced by factors beyond clinical need.[15,16] Some paramedics say they struggle to identify cases suitable for non-conveyance and want help in identifying the risks and benefits of conveying and not conveying individuals.[4][5][6][7] RADOSS seeks to address this by developing a risk prediction tool.Such tools use ≥ 2 pieces of patient data to generate a personalised estimate of the likelihood that an individual will experience certain endpoints within a.
specified time frame.RADOSS proposes the new tool provides estimates of two outcomes known to feature in paramedics' decision making [17] namely, likelihood of death and/or recontact with the urgent and emergency care system (within 3 days), as well as the likelihood of an attendance meeting the definition of an avoidable attendance if conveyed.The tool could be deployed when paramedics manage cases that do not present an obvious need for conveyance to ED. [3] Such tools show benefit in analogous situations.Moreover, paramedics use them for other presentations [3] and have expressed a desire for more tools like this to aid decision making.[18].
To support the development of robust risk prediction tools and facilitate use, statistical standards are available.[19] However, health innovations may not be adopted if user and recipient need, preference and workflow are not considered.[19] With this in mind, four aspects needed exploration with stakeholders.
The first was perceived need.Both service users and clinicians have articulated the necessity for change (see [20]).Nevertheless, their perspectives regarding the potential of a risk prediction tool to meet this need, along with any associated concerns and expectations, remained unexplored.
The second was which items to test for inclusion in the tool, here referred to as 'candidate predictors'.Whilst 'on-scene', paramedics lack access to a patient's full medical record.What is available is the information they record within structured fields on the patient's care record (PCR) and what was recorded by call handlers in computer aided dispatch (CAD) systems.Lots of this information could potentially predict outcomes.However, there are statistical limits on how many variables RADOSS can test.[21] Insights from individuals with lived and clinical experiences, including expected strength of association with the outcomes, reliability, and accessibility during conveyance decisionmaking, could inform RADOSS' selection for testing and subsequent prioritization.
The third area pertained to outcome measures.It was necessary to understand whether RADOSS' definitions of them were considered suitable by clinicians to guide their decisions.
Finally, it was necessary to start to understand clinicians' preferences for tool functionality and implementation.This exploration is particularly significant as clinicians' preferences could influence model derivation, including considerations like limits on variables clinicians are willing to input.
To address these knowledge gaps, KE was completed with urgent and emergency care clinicians (comprising of ambulance service paramedics and ED clinicians), service users, and their significant others.

Design
Six KE workshops were conducted.Their design, led by Wilkins and Cooper's [22] KE definition, focused on a two-way exchange between researchers and research users.To secure comprehensive insights [23] and facilitate recruitment, an online group format was chosen.Reporting conforms with the Standards for Reporting Qualitative Research.[24].

Participants
Purposive and diverse samples of 20-30 service users and 20-30 clinicians were sought.
Service users needed to be aged ≥ 18 years and self-report having had contact with the ambulance service during the prior 12 months for a suspected seizure or be a significant other to a person who had (Table 2).They were identified by three patient groups affiliated with the target population circulating recruitment advertisements in March 2023: 'FND Hope' shared the study advert twice on their public social media platforms; 'Epilepsy Action', as well posting on social media twice, advertised the study on their volunteering and study opportunities webpages (see Acknowledgements); and finally, 'Speak up Advocacy' (Sheffield), a group run by and for people with learning disabilities and autistic people, sent the study advert to its members.All persons interested in participating conducted our study team by email or phone.The team confirmed individuals' self-reported eligibility, sent them a Participant Information Sheet, answered any questions they had and secured written, informed consent.No medical records were accessed to confirm a person's eligibility to participate, nor their diagnosis/es.

Table 1
Definition of an 'avoidable attendance' a.

A person has been involved in an avoidable attendance if routine hospital coding for the attendance indicates it:
• did not result in the person being investigated Except b : urinalysis; dementia test; pregnancy test; dental investigation; glucose measurement; peak flow; visual acuity testing

• did not result in the person being treated
Except b : new prescription, medication review, social assessment, psychosocial assessment, recording vital signs, dental treatment, activities of daily living assessment, mobility assessment, closure of a skin wound by tape, gluing of wound, application of a minor dressing; guidance/advice).

• AND the person was discharged
Be it to: home; residential care; custody.Incident could be related to a selfreported diagnosis of either epilepsy and/or NEAD.
Able to provide informed consent and participate in a workshop independently in English; Lives in England.
▪ Project protocol stated intention to recruit persons who had been attended to be an ambulance aged ≥ 16 years.This was to align with national ambulance guidelines [3] which note conveyance is only always required for those aged < 16 years.Ethical approval, however, required persons taking part because they had been attended to by an ambulance needed to be aged ≥ 18.
Significant others were recruited since they can be present when ambulances attend, and it is to them that patients can delegate care decisions.[45] Epilepsy and NEAD account for ~ 70 % of suspected seizure ambulance incidents.[2]

Clinicians
▪ Aged ≥ 18 years (no upper limit); Paramedic, ED doctor or nurse; Works in England; Able to provide informed consent and participate in a workshop independently in English.
▪ Paramedics were recruited since it is their decision making the tool seeks to support.ED clinicians were principally recruited because their expertise in how seizures are managed in ED meant they could help identify challenges of using the avoidable attendance definition for them; Notes: ED, emergency department; NEAD, non-epileptic attack disorder.Eligible urgent and emergency care clinicians, comprised of ambulance service paramedics, or ED doctors and nurses (Table 2).To aid recruitment, England's 10 regional ambulance services and the Royal College of Emergency Medicine Yorkshire and Humber board were asked to circulate an advertisement.Adverts were also placed within Facebook interest groups (Acknowledgements).

Ethics
The Health Research Authority (23/HRA/1439) and University of Liverpool Ethics Committee approved the study (11450).Participants were offered a £10 voucher.

Procedure
Workshops for service users and clinicians ran separately.All were facilitated by BM, a qualitative health services researcher.LJB was present at clinician workshops to assist with statistical questions.AN was at service user workshops to offer support.
Topic guides were developed on the basis of the literature [25] and uncertainties regarding the tool's future implementation, [26] and were piloted with 8 service users (Supplementary File 1).Nominal Group Techniques [27] were used to facilitate open and constructive sharing and discussion of views.
Fig. 1 shows workshop structure.They started with a presentation to orientate participants to the proposed tool (Supplementary File 2).
To gather predictor suggestions, service users and clinicians were presented with a slide displaying the information routinely recorded within structured data-fields and accessible to paramedics whilst 'on scene'.Service users were asked which might indicate care needs following a seizure; clinicians were asked which might predict the likelihood of the outcomes.They could make suggestions for the outcomes separately or collectively.Groups were also asked to provide suggestions not included on the slide.All service user responses were recorded 'live' on screen by AN within a table (Supplementary File 1).
With participants' consent, workshops were audio-recorded and transcribed.Participants did not review transcripts.

Candidate predictors
The objective was to generate a list of candidate predictors prioritised by service users and a list for clinicians.Thus, for service users BM extracted the tables generated by their workshops and merged them.For the clinician list, BM reviewed transcripts, extracting candidate suggestions.All suggestions were classified according to whether they are captured within structured ambulance electronic PCR/ CAD data-fields (Table 3 footnotes gives further detail).

Perceived need, candidate justification and views on functionality/ implementation
Qualitative findings on these topics needed to be analysed with high methodological rigour but swiftly enough to inform model derivation.Thus, we employed Hamilton's Rapid Analysis method.[28] Instead of line-by-line coding, analysts review transcripts, populating a 'summary template' for each data collection episode with relevant data segments, creating 'episode profiles'.
For RADOSS, Rapid Analysis involved BM and AN independently populating a summary template for each workshop and then holding consensus meetings to consolidate their work and create a final, populated template for each.Templates were generated a priori (Supplementary File 3) and included domain headings based on the topic guide.
Matrices helped to support the identification of similarities, differences and trends across workshops and informants.[29] Aggregated results were presented to the wider team to facilitate cross-checking before finalisation and decisions being taken about their implications for the project.
Illustrative quotations from participants (P) are provided, with Supplementary File 4 providing additional ones.There has been minor editing of some to preserve anonymity and ensure clarity.

Participants
From 47 eligible service users responding to the advert, 33 were available for, and booked in to attend a workshop.A total of 26 ultimately participated (17 females, 9 males; aged 18 to 85).They included, 18 people with diagnosed seizure disorders and 8 significant others.Service users came from 7 of England's 10 ambulance regions.Fig.
shows the recruitment process, workshop details and composition of the groups.
From 60 clinicians contacting us, 32 were available for, and booked in to attend a workshop.A total of 25 ultimately participated (15 females, 10 males).They included 19 paramedics (from 6 ambulance regions) and 5 ED consultant doctors and 1 ED doctor in training (from EDs).
Service user workshops occurred in April 2023, lasting 74 to min, while clinician workshops occurred in June 2023, lasting 82 to min.

Anticipated benefits
Service user and clinician participants supported the proposed tool's development.Clinicians were keen for additional support for what they identified as complex decisions: "We have to make decisions based on a patient presentation who has been with us for 30 min… in ED …they get…longer…It can be difficult."(P31; Male, paramedic) Service users shared instances of important care interventions by the ambulance service.Most though, said their prevailing experience was transport to ED when they believed it was not required.They felt the tool might mitigate against knowledge gaps they perceived some paramedics to have which contributed to this: " I agree I think it's a good thing…The number of paramedics that have pulled a phone out and 'Googled' what is NEAD…is frankly alarming… some…have been a bit shaky with epilepsy as well."(P7; Female, person with epilepsy and NEAD) Service users emphasised ED visits when not required were not 'riskfree'.Indeed, some said there was a need to "de-medicalise" uncomplicated seizures: "Need to recognise that going to ED when not necessary can be stressful which itself can elicit further seizures.When I wake up and find myself in ED it can be really distressing…" (P21; Male, person with epilepsy)

Potential concerns
Service users and clinicians did not want risk estimate/s from the proposed tool to dictate care choices.Rather, they wanted them to be one, albeit important, piece of information accounted for.Service users wanted their preferences to be heeded: "the experience of the persons experiencing the seizures and the carers should not be dismissed… if paramedics had more experience than me, I would fall off my chair…" (P5; Female, significant other to person with epilepsy and NEAD).
Most clinicians said paramedics needed to still be permitted to use their judgement since there may be additional factors of importance:  "You still need to be allowed to use judgement…the tool assumes there is a safe alternative for those who on paper don't need to have come to ED…it isn't always available or possible…" (P41; Female, ED clinician)

Candidate predictors 3.3.1. Variables routinely available to paramedics at the time and recorded within structured fields
Stakeholders together suggested ten variables or variable 'families' which might be predictive of the likelihood of death/ recontact and occurrence of an avoidable attendance (Fig. 3).These included, but were not limited to, vital signs, presence of a learning disability, incident time, location, who made the call and treatments provided.There was partial overlap in clinician and service user suggestions.Table 3 shows the reasons participants offered for their potential importance.
In certain instances, the variable was suggested to indicate a need for additional investigation or treatment in ED (e.g., significant injury, neurological deficit).In other cases, it complicated paramedics' understanding of the person's needs (e.g., caller identity) or limited their ability to provide care beyond ED (e.g., time of day).Although clinicians were encouraged to propose predictors for the two outcomes separately, most did not.

Variables not routinely available to paramedics at the time and/or not recorded within structured fields
Participants suggested 16 variables not currently captured within structured PCR data fields that might also be predictive (Fig. 3).They included seizure type, presence of a witness, availability of alternative care options and the attending paramedic's experience.Table 3 shows stakeholders' reasons.
Clinicians mentioned that paramedics could record information on some of the variables in PCR free-text sections.The implications of RADOSS not accessing this information were debated.Some considered it a limitation, while others saw it as less significant.The latter argued that it does not reflect what is available to paramedics during conveyance decisions.
"…think how paramedics fill in the PCR…demographic information gets 'pulled' straight away… next thing is…the observations and…primary survey….These might not be the most useful variables, but they're the ones filled in…when…deciding whether to take them to hospital… If you are getting to the stage where you're writing your free text up… you're already in the 'headspace' of your…conveyance decision" (P51; Male, paramedic)

Matters complicating deriving a predictive tool
Clinicians and service users identified things they said might make it challenging to derive a model able to make precise predictions.
The first was some patients may be given inappropriate care.Thus, the information recorded on the PCR about the incident may not reflect patient need.The example participants gave was medication being inappropriately given to people with NEAD:

"….with the FND [functional neurological disorder] seizures I've been given diazepam a few times…that just makes it worse…" (P15; Female, person with epilepsy and NEAD)
Another reason the PCR might not reflect patient need was measures used by the ambulance service might not be sufficiently sensitive to differentiate between presentation severity:

No -
Notes: A&E, accident and emergency department; comms, communications; ED, emergency department; GCS, Glasgow Coma Scale score; HDU, high dependency unit; ICU, intensive care unit; NEAD, non-epileptic attack disorder; neuro, neurology; P, participant; PTSD, posttraumatic stress disorder.We note that some of the variables might not be directly recorded within a structured data field but may be discernible from data recorded within one or more other structured fields (e.g., service user having a learning disability might be indicated by the CAD fields, 'Where call has originated from' and 'Pickup Location Type' or the PCR field 'Incident Location Type').In such instances, we have classified these variables as being available from a structured data field and accessible for derivation.
a The grouping of the candidate predictors according to whether or not they are recorded within structured data fields was supported by a data-dictionary, generated by the Yorkshire Ambulance Service (YAS).It indicates the data its staff record within structured Patient Care Record (PCR) or computer aided dispatch (CAD) data fields for incidents and which will be accessible for the purposes of derivation via the 'CURED+' data set.CURED + will be used by RADOSS to derive and validate the prediction model.Further details on CURED + are available via RADOSS' protocol.[20] In brief, CURED + is being developed by the University of Sheffield's Centre for Urgent and Emergency Care Research.It maps urgent and emergency care use by individuals served by the YAS from 2011 to 2022 and will contain records of all ambulance contacts which will be linkable to any subsequent ambulance, hospital (ED, inpatient) and death records.b Information on comorbidities is potentially available via different sources.However, it is not routinely available for all patients, nor recorded for all.c Within England, guidelines state all adults with epilepsy should have an agreed and comprehensive written epilepsy care plan.[46] One section it should include is information on "first aid, safety and injury prevention at home and at college or work".[47] Care plansor pertinent information derived from them -are not routinely accessible to urgent and emergency care staff for people with seizure disorders.d This is potentially derivable for all, but it is usually only available for select patients, such as 'frequent callers', those with calls within the last 24 h.

"….[measures like the] Glasgow Coma Scale are not always the best tools…someone who's post-ictal could have a GCS hovering around 13,14… for say 20 min. During that 20 min you are…seeing them becoming less postictal, but the GCS is still hovering at 14." (P44; Male, paramedic).
A third challenge was potential inaccuracy of some information recorded within structured PCR/CAD fields.Inaccuracies could arise from information offered by bystanders and patients themselves: "…people misinterpret, misunderstand, particularly observers…I've asked people at work [witnessing the same seizure] to describe it… You wouldn't believe the variation you get in stories……" (P22; Female, person with epilepsy) "…I'll often be saying something in the post-ictal phase, but it's not what I think I've said."(P24; Male, person with epilepsy) A final issue participants said might complicate matters was that certain presentation features have different implications for care needs, depending on the person's seizure disorder.An example provided by participants was the meaning of seizure activity: "…With dissociative seizures I've had a seizure for 45 min…if that was epilepsy you would go straight to hospital whereas you wouldn't necessarily need to go if it is a dissociative seizure…Its not just a list of things."(P14; Female, person with epilepsy and NEAD) For these reasons, participants discussed whether the tool should consider a person's seizure diagnosis initially.However, most acknowledged that paramedics typically lack access to a person's medical history for this purpose.Some users shared their efforts to overcome this, while those with recent diagnoses were surprised by the lack of routine access.3′s note provides details on the grouping of candidate predictors according to their availability via structured data fields.

A) Candidate predictors available from structured data fields B) Candidate predictors not available from structured data fields
"Is that not something they [the ambulance service] have in place?I would have thought that would have been in place 'donkeys years' ago… that sounds crazy… (P23; Female, person with NEAD)

Outcome measures
Clinicians confirmed the outcomes' importance.However, they raised issues with regards their proposed definitions.

Death and recontact
They advised that the significance of death and service recontact is not equivalent and so alongside the combined risk estimate, they said it would be prudent to articulate the proportion experiencing each element: "…I think if you're gonna use in the tool the words 'death within three days' it suddenly adds a massive emphasis…on risk…there's a… difference between recontact and death…."(P44; Male, paramedic) With regards service recontact, clinicians advised that it would be important for those using the tool to know what types of service recontact are and are not included (e.g., whether it needs to be seizure related).
They debated the suitable time frame for events post-incident, focusing on the duration within which recontact could be reasonably linked to previous care.Opinions varied.Some suggested 24 h, others 3 days, and a minority 30 days.Some paramedics hesitated to offer suggestions, citing limited feedback on outcomes after ambulance care, hindering the formation of opinions.It was proposed that a range of estimates might be appropriate.
"…it would bother me less if they…recontacted at 3 days…But yeah, 24 h and under…I would say I would probably think that I've…made a mistake."(P31; Male, paramedic) "…3 days in the kind of the grand scheme…is probably quite a short amount of time…for us to potentially being basing our decision making on it….I think…30 day[s]…" (P38; Female, paramedic)

Avoidable attendance
Concerning avoidable attendance, clinicians highlighted that a potential drawback in O'Keeffe et al.'s [10] criteria was its assumption that all recorded investigations, treatments, and admissions from ED were clinically 'appropriate.'They observed that some tests be administered for reasons beyond necessity for seizures, particularly blood tests and intracranial imaging.The former might be routinely collected to try to reduce patient waiting time.The latter could be overused due to staff misunderstanding.Clinicians stressed the importance of exploring and accounting for these issues if necessary.
"…tests can be done purely because somebody has arrived at the front door…might be because there's 'preloading'….mightbe because an F2 [trainee doctor] sees that patient rather than a consultant….Also, there's variation up and down the country…" (P27; Male, ED clinician) To a lesser extent, it was also noted that the time and day on which a person visits an ED might be important, since it might impact likelihood of hospital admission: "I think time of day people attend…and um day of week people attend… It's far harder to discharge someone at 3:00…in the morning, then is it at 3:00…in the afternoon."(P33; Female, ED clinician)

Tool functionality and implementation
Clinicians provided a range of insights about their preferences.Table 4 details them.
Regarding functionality, a crucial aspect was the number of variables paramedics would input willingly, with the upper limit being ~ 10.This, however, depended on input complexity.Integrating the tool into electronic PCR systems, they noted, could eliminate any cap by allowing for 'auto-population.'.
Clinicians recommended additional features for the tool's estimates, noting that the risk tolerance of clinicians might differ from that of service users.They suggested contextualizing risk estimates, colorcoding them with 'traffic lighting,' and prompting clinicians to discuss these estimates with service users.
"…as clinicians we're quite risk averse…the risks…we would take on behalf of patients are smaller than the risks patients…take for themselves, they're more likely to say, yeah, that's alright…."(P41; Female, ED clinician) After development, clinicians suggested actions to increase uptake, emphasizing the importance of national ambulance guideline recommendation and regional service endorsement.They stressed the need to optimize alternative care pathway provision due to existing disparities among areas, potentially hindering widespread use."…what's good is that if somebody comes to my ED with a seizure…they get a referral to our neurology centre and they get reviewed…you don't want to set something up that is then detrimental to care…."(P27; Male, ED clinician)

Main findings and implications
Novel KE workshops were conducted with stakeholders.These provided important insights into service user and clinician priorities for the proposed risk prediction tool and indicated actions RADOSS can take to optimise how it intends to develop it.

Need and anticipated benefits
The first piece of important evidence related to need and anticipated benefits.Given tools should be developed to meet an identified need, [30] it was important to find that, in line with the preliminary evidence instigating RADOSS, that stakeholders supported tool development.They articulated anticipated benefits.Paramedics hoped it could enhance support for a complex decision.Service users believed it could address knowledge gaps they perceived paramedics had which contributed to perceived over-conveyance.

Candidate predictors
Stakeholders were shown the information routinely recorded within structured data-fields and accessible to paramedics whilst 'on scene'.They discussed which items should be considered for testing.Ten were suggested, which clinicians also identified as being the maximum number it would be reasonable to expect a clinician to input.Some were proposed by both users and clinicians, others were not.Some might have been possible to identify as candidates on the basis of the literature (e.g., vital signs, significant injury), [10,15,31] others could not have been (e.g., who made the call, presence of learning disability).This emphasizes the value of engaging with stakeholders and learning from their lived and clinical experience.RADOSS will seek to test all these suggestions, not least because a tool's adoption can be hindered if it fails to account for factors users consider important.[32,33].

Concerns
The workshops surfaced potential concerns.A starting assumption of RADOSS, aligned with the tenets of 'person-centred care', was that the proposed tool would support care decisions, not impose them.This was shared with participants via the presentation.Nonetheless, stakeholders primary concern was whether clinical judgement and patient preference would still factor in decisions.This emphasises the issue's importance to stakeholders and that going forward, RADOSS needs to better communicate the tool's assistive role.
Another reason such messaging is important relates to care plans and the seizure management instructions they contain.Should a paramedic find themselves with access to both the tool and a patient's care plan, the latter should be prioritized as it will be more individualised.However, it is noteworthy that care plans, especially ones accessible to paramedics, remain uncommon.Paramedics describe finding such items as "like striking gold".[6].

Outcome definitions
Clinicians clarified what adjustments should be made.Concerning death and recontact, RADOSS proposed estimating their incidence within 3 days.Since death is rare (<1% of cases), [34] for statistical reasons, we anticipated combining it with service contact (~12 % of cases within 3 days).[35] Clinicians advised that the tool should, in providing the combined estimate, also state the proportions experiencing the two types of events to guard against inadvertently facilitating risk averse decisions.
No consensus on the suitable prediction timeframe for death/service recontact emerged.To address varying perspectives, stakeholders recommended a risk dashboard with different time frames.Consequently, the RADOSS will estimate death/recontact within 3 days via its primary analysis and 1 and 30 days via secondary analyses.
Regarding avoidable attendances, O'Keeffe's [10] criteria assumes all ED investigations and treatments are appropriate.Stakeholders noted potential overuse of intracranial imaging and full blood panel testing.The latter has not been identified before for UK EDs.[36] RADOSS needs to address these investigations' use to prevent under classification of attendances as avoidable, compromising the tool's utility.RADOSS will thus compare the proportion of conveyed incidents in the derivation dataset classified as "avoidable" when said investigations are considered versus when they are not.If a substantial difference (≥15 %) is observed, the avoidable attendance definition will be amended to exclude these investigations.

Functionality/implementation
If RADOSS can create a validated tool, stakeholders provided insights on desired functionality and strategies to enhance uptake.Suggestions encompassed embedding the tool within electronic PCR systems and utilizing colour coding for interpretation.While more detailed input, such as preferences for risk information visualization, is needed before finalizing the tool, [25] the obtained evidence is proportionate to the development stage.

Future research
Within the wider literature, information routinely recorded by paramedics within structured fields has been found to help predict other outcomes (e.g., [37]).Whether it can help predict the outcomes of interest to RADOSS remains to be determined.Stakeholders did pinpoint challenges for RADOSS in deriving a precise prediction model.However, they also proposed additional factors, currently absent from structured data fields, for future research consideration to potentially enhance the

Fig. 2 .
Fig. 2. Recruitment flow diagram for, and composition of, workshops for (A) service users and (B) clinicians.

Fig. 3 .
Fig. 3. Candidate predictor suggestions made by workshop participants Notes: HDU, high dependency unit; ICU, intensive care unit.Table3′snote provides details on the grouping of candidate predictors according to their availability via structured data fields.

Table 2
Inclusion and exclusion criteria for participant Knowledge Exchange workshops and rationale.
▪ Aged ≥ 18 years (no upper age limit); Attended to by an ambulance during prior 12 months for a suspected seizure/ s OR A significant other to such a person (e.g., family member, friend) who was aged ≥ 16 years;

Table 3
Candidate predictors suggested by workshop participants and their reasons.
(continued on next page) A.J.Noble et al.

Table 3
(continued ) (continued on next page) A.J.Noble et al.

Table 4
Clinicians views on tool functionality and implementation considerations.verynice to have kind of objective tools that give you an answer…rather than just saying now use clinical knowledge…[that] just feels like 'oh what's the point using the tool' if…it gets thrown back onto us kind of thing".(P32,Female, paramedic)"…suggestions should be made, but they remain guidance, because at the end of the days it should still be your clinical decision… you've still got a choice…."(P39, Female, paramedic) Contextualise

estimate against risk in other populations "
…you can provide… the background risk of that person having an adverse outcome in those three days or 30 days regardless of them having had a fit anyway.So what's the chance of the general public attending the ED 3 days after they've attended once already…" (P41, Female, ED) "It'll be really important that that tool is…in the JRCALC …it needs to be more than just put into the ether …it needs to be kind of 'rubber stamped'(P31,Male, paramedic)

Included in local ambulance service tool repository "
…I find it a bit overwhelming like how many different tools there are… … each condition has got something extra to it that you have to remember…So I think for me like just make it really accessible… my ideal for me would be like integrated into existing systems."(P49,Female,paramedic)"…in [our ambulance trust] we've got a repository…you'll get it literally on your phone…I would say they are referred to far more…" (P31,Male, paramedic)