When has service provision for transient ischaemic attack improved enough? A discrete event simulation economic modelling study

Objectives The aim of this study was to examine the impact of transient ischaemic attack (TIA) service modification in two hospitals on costs and clinical outcomes. Design Discrete event simulation model using data from routine electronic health records from 2011. Participants Patients with suspected TIA were followed from symptom onset to presentation, referral to specialist clinics, treatment and subsequent stroke. Interventions Included existing versus previous (less same day clinics) and hypothetical service reconfiguration (7-day service with less availability of clinics per day). Outcome measures The primary outcome of the model was the prevalence of major stroke after TIA. Secondary outcomes included service costs (including those of treating subsequent stroke) and time to treatment and attainment of national targets for service provision (proportion of high-risk patients (according to ABCD2 score) seen within 24 hours). Results The estimated costs of previous service provision for 490 patients (aged 74±12 years, 48.9% female and 23.6% high risk) per year at each site were £340 000 and £368 000, respectively. This resulted in 31% of high-risk patients seen within 24 hours of referral (47/150) with a median time from referral to clinic attendance/treatment of 1.15 days (IQR 0.93–2.88). The costs associated with the existing and hypothetical services decreased by £5000 at one site and increased £21 000 at the other site. Target attainment was improved to 79% (118/150). However, the median time to clinic attendance was only reduced to 0.85 days (IQR 0.17–0.99) and thus no appreciable impact on the modelled incidence of major stroke was observed (10.7 per year, 99% CI 10.5 to 10.9 (previous service) vs 10.6 per year, 99% CI 10.4 to 10.8 (existing service)). Conclusions Reconfiguration of services for TIA is effective at increasing target attainment, but in services which are already working efficiently (treating patients within 1–2 days), it has little estimated impact on clinical outcomes and increased investment may not be worthwhile.

Strengths and limitations of this study 1 2 • This is the first study to model the impact of routine service provision for patients suffering 3 transient ischemic attack. 4 • The results are likely to be representative of the hospitals studied, but not necessarily other 5 centres in the UK and across the world. 6 • Modelling allows estimation of service 'unknowns' which requires pre-programming and 7 therefore simplification of certain service intricacies. Thus, the services modelled represent 8 stylised versions of the actual services offered by participating hospitals. 9 • Costs relate to standard weekday services and it is not clear whether those actually incurred 10 would be increased by provision of services at the weekend. Transient ischaemic attack (TIA) is common, with incidences of 15-83 patients per 100,000 2 population recorded across the world; [1][2][3][4][5] in the UK, it affects approximately 50 patients per 100,000 3 population. 6 TIA is important because it represents a significant risk factor for future stroke with 4 around 8% suffering an event within 7 days, and 12% within a month without preventative therapy. 7 Rapid recognition and treatment of TIA patients at high risk of having a subsequent stroke is 8 important because simple interventions, such as early prescription of preventative medications, can 9 substantially reduce the risk of stroke following TIA, and existing evidence suggests that earlier 10 intervention maybe better. 9 10 In the UK, guidelines 11 12 recommend that patients at high risk of 11 recurrent stroke (defined using the ABCD 2 score of >4) 10 are seen within 24 hours of symptom 12 onset and all other patients are seen within 7 days. A recent UK audit suggests there has been 13 significant improvement in attainment of these targets: 45% of high risk outpatients and 60% of 14 inpatients now receive treatment on the same day as referral, compared with just 10% and 33% four 15 years earlier. 13 This improved target attainment has been achieved through a variety of approaches, 16 with some centres admitting more patients for assessment and treatment and others designing 17 services with excess routine clinic capacity. 14

19
Service reconfiguration remains ongoing in routine clinical practice and it is unclear what impact 20 such changes have on service costs and clinical outcomes. This study used discrete event simulation 21 modelling to assess the impact of TIA service reconfiguration in two large urban hospitals with 22 different approaches to service provision. 23 24 Methods 25 An extended methods section can be found in the online data supplement. Setting and service design 1 The model was designed to assess the costs and clinical consequences arising from patients 2 suspected of suffering an acute TIA, referred to two large urban hospitals in the West Midlands 3 region of the UK. Both hospitals ran a specialist TIA outpatient service catering for approximately 4 500 patients with suspected TIA and minor stroke every year. The model parameters were defined 5 by the characteristics of individuals attending these services, recruited to an observational study 15 6 during 2011 (table 1). 7 8 The availability of specialist TIA clinics was modelled on the basis of existing service provision at 9 participating hospital sites during the study period. Patients could either be seen in traditional 10 outpatient clinics, admitted, or seen on the ward on an outpatient basis. The original services were 11 designed within the confines of available clinical staff and in particular a lack of specialist cover at 12 the weekends. Both services subsequently underwent redesign and the impact of these changes on 13 cost and outcomes was examined by modelling both the original and modified service. The number 14 and distribution of clinics in each service are described in table 2. 15 16 The impact of further hypothetical adjustments to the modified service at each site was modelled to 17 replicate the following scenarios: 18 a) The addition of high risk clinic slots for patients presenting on a Saturday and Sunday 19 b) Including a weekend service but reducing the number of routine weekday clinics, by up to 5 20 clinic slots per week. 21 22 Overview of the model 23 A discrete event simulation model was programmed in Delphi version 4 (Borland, San Francisco, symptoms and followed the patient along the clinical pathway from initial presentation to follow-up 2 for subsequent stroke morbidity and mortality. An essential feature of the model was that patients 3 were sharing limited resources in the form of routine clinics. 4 5 Clinical pathways in the model 6 Following onset of an initial event, patients were assumed to contact either their general practitioner 7 (GP) or attend the emergency department (ED) (eFigure 1, online supplemental material). The 8 probability that a patient would choose a specific route was dependent on the type of patient 9 (high/low risk), the time of day, and day of week, estimated using data from the Oxford Vascular 10 Study. 17 Following this initial contact with a healthcare professional, patients were referred to a 11 specialist TIA outpatient clinic, seen on the ward (as an outpatient) or admitted as an inpatient. The 12 type of referral was dependant on the specific hospital service provision, clinic availability and the 13 level of risk of the attending patient, defined according to the ABCD 2 score. 10 During clinic 14 attendance, it was assumed that the appropriate treatment would be initiated (i.e. blood pressure 15 lowering, cholesterol lowering or antiplatelet therapy in accordance with guidelines) and a small 16 proportion of patients (4.1%) would be treated with carotid endarterectomy. 12 18 17 18 Model population 19 Overall rates of patient presentation were based on initial runs of the model and ensured a realistic 20 distribution of final diagnoses which represented the study sample, previous literature 14 and expert 21 opinion (60% TIA mimics, 33% genuine TIA, and 7% minor stroke). The observed ratio of low 22 risk/high risk TIA patients used in the base-case of the model was supplemented in sensitivity 23 analyses by estimates derived from those reported in previous studies 10 19 20 and the experience of The range of these estimates tested in this sensitivity analysis for impact on the model 1 results was: 2.5:1 (high:low risk; base-case) to 1:1, 5:1 and 7:1. 10  Follow-up and risk of repeat events 4 Patients remained in the model until one year from symptom onset or until they suffered a repeat 5 event (within the same year), after which the increased risk of repeat event returns close to normal. 7 6 8 The risk of a repeat event (TIA or disabling stroke) was dependent on the initial event (true TIA 7 or mimic), the severity of that event (measured by ABCD 2 score) and other relevant risk factors 8 such as age, presence of atrial fibrillation and medication prescribed. 10 18 21-26 Risks were modelled 9 using a Weibull distribution for the time to event which allowed for a substantially increased risk in risk patients not seen by a specialist within seven days of referral. 11 12 Further outcomes examined 1 the median time from referral to specialist appointment, the total number of routine outpatient 2 appointments available (used or unused) and any unscheduled outpatient appointments required 3 (where high risk patients were assessed immediately on the ward). service]). 15 16 Potential optimal service provision 17 Modelling the impact of further hypothetical service reconfiguration, including a seven day service, 18 suggested that it was possible to reduce service costs and the number of unused clinic appointments 19 and improve guideline target attainment, but this had little impact on the incidence of disabling 20 stroke post TIA (table 3 and eTable 3). Figure 2 shows the trade-off between routine appointment 21 provision, unscheduled appointments and high/low risk guideline breaches. Assuming unused 22 appointments (pink), unscheduled appointments (in light blue), admissions (purple) and breaches 23 (red [high risk breaches] and light red [low risk breaches]) are preferably avoidable outcomes of 24 service provision, the optimal hypothetical service in hospital 1 involved a weekend service with appointments were incurred, the optimal hypothetical service involved a weekend service with 10-  The aim of this study was to model the impact of TIA service reconfiguration in two large urban 7 hospitals and establish the impact on clinical outcomes (disabling strokes), costs, and clinic usage. 8 Reconfiguration of services was found to be effective and appropriate for reducing resource (clinic 9 slot) waste, and improving guideline target attainment with variable effects on costs. However, in 10 hospitals where services, despite poor target attainment, had minimal patient delays, such as those 11 studied here, significant reconfiguration had little impact on modelled clinical outcomes such as 12 disabling stroke post TIA. These findings suggest that clinicians and service coordinators should be 13 cautious before initiating significant reconfiguration of services which are already seeing high risk 14 patients within 1-2 days, unless there is an obvious potential to reduce costs, such as reducing 15 admissions of high risk TIA patients. Whilst national targets for care are important, consideration 16 should be given to revising them to maximise the benefits of attainment whilst not incentivising 17 small changes that might increase costs for no measurable clinical benefit. 18 19 Strengths and limitations 20 The present study used local hospital data to describe the characteristics of the presenting 21 population and thus the results are likely to be representative of the hospitals studied, but not 22 necessarily generalizable to other centres in the UK and across the world. Despite this, guideline 23 target attainment in hospital 1 (original service configuration) was similar to that reported nationally 24 (31% vs. 37% nationally) 13 which suggests that this hospital was more likely representative of 25 centres across the UK. Modelling allows estimation of service 'unknowns', such as the number of unscheduled clinic 2 appointments required to meet guideline targets for high risk patients. This requires pre- 3 programming with specific rules and therefore simplification of certain service intricacies is 4 necessary. For example, the process by which high risk patients were allocated unscheduled clinic 5 appointments was formalised such that a patient referred by the ED physician was assumed to have 6 been seen by a stroke specialist within two hours. However, patients referred by their GP were 7 assumed to be seen by the stroke specialist on the ward at 10am the following morning or 5pm the 8 same day, whichever was later (but within 24 hours of initial referral). In reality, services are often 9 adapted in response to a specific set of circumstances such as clinic room availability and thus, the 10 services modelled here can best be thought of as stylised versions of the actual services offered by 11 the hospitals in question. 12 13 The costs modelled here were taken from standard NHS reference costs 27 28 and included 14 investigations, treatment and the cost of staff attending patients during each appointment/admission. 15 Costs relate to standard weekday services and it is not clear whether those actually incurred would 16 be increased by provision of services at the weekend. It is therefore possible that the costs of 17 providing a weekend service may have been underestimated in the present analyses. 18 19 Findings in the context of previous literature and implications for practice 20 Economic modelling has previously been used for determining the impact of service reconfiguration 21 in acute stroke 30-33 but studies examining optimal strategies for TIA service provision are less 22 common. Those that exist, focus on whether or not patients should be admitted or be seen as an 23 outpatient. 16 34 35 It is generally accepted that patient admission for TIA is inefficient, 15 Sentinel Stroke National Audit Programme (SSNAP) in the UK. 13 The latter represents hospitals 1 nationally (across the UK) but has only a small focus on TIA service provision and relies on 2 participating hospitals to collect their own data and perform quality checks, introducing the 3 possibility of bias. 4 5 The primary finding of this study was that whilst service reconfiguration can significantly improve 6 guideline target attainment and potentially reduce service costs, such changes have little impact on 7 the estimated prevalence of subsequent stroke, when services are already working efficiently. In the 8 present study, even the largest service reconfiguration (introduction of a 7 day service) only reduced 9 the median time from patient referral to clinic attendance and treatment by 0.3 days (8 hours). This 10 enabled a large number of patients to be seen within 24 hours, improving guideline target 11 attainment but had little impact on stroke risk. The recommendations for a 'see and treat within 24 12 hours' target for high risk patients were originally based on data from the EXPRESS study, 9 which 13 showed that those patients seen and treated within 24 hours of initial referral had significantly better 14 outcomes than those patients who were not. However, in the original service to which this optimal 15 strategy was compared, the average time from referral to clinic appointment was 3 days (IQR 2-5 16 days) and time to first treatment was 20 days (IQR 8-53). The improvement in time to initiation of 17 treatment was therefore 20 days, compared to an improvement of 8 hours seen in the present study, 18 which explains the very different results in terms of clinical outcome. The modified services examined and proposed here may require the capacity to accommodate 1 unscheduled clinic appointments, which is likely to have an opportunity cost affecting the quality of 2 other services running concurrently in the hospital and may result in increased costs with little 3 tangible benefit. In particular, the ability to accommodate such appointments depends on the time of 4 day at which the patient presents, the availability of specialist stroke physicians and scanning 5 equipment required to make an accurate diagnosis, as well as the needs of competing services 6 within the hospital (e.g. acute stroke services). Reconfiguration of specialist services for TIA can be effective and appropriate for reducing costs, 10 reducing the number of unused routine clinic appointments and increasing clinical guideline target 11 attainment. However, where services are already working near optimal, such modification has little 12 impact on clinical outcomes such as subsequent stroke. Clinicians and service coordinators should 13 be cautious before initiating significant reconfiguration of services which are already seeing high 14 risk patients within 1-2 days, as such changes may increase costs with little or no change in 15 outcome. Ethics approval and consent to participate 2 Approval for this project was obtained from the National Research Ethics Service (NRES) 3 Committee, London -Queen Square (reference; 09/H0716/71). 4 5 Data sharing 6 The datasets analysed during the current study are available from the corresponding author on 7 reasonable request. 8 9 Competing interests 10 The authors declare that they have no competing interests  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59    weakness, Duration of symptoms 0-9 mins, not diabetic (resulting in ABCD and ABCD 2 of scores 8 2), AF on warfarin, cholesterol 6.0.  14 Original and modified services (first two bars at each hospital) were those which were actually 15 implemented in each hospital. The remaining weekend services are hypothetical. *Hospital 1 16 included 16 routine clinics, 5 days per week (original service), 17 in the modified service and 19 in 17 the modified service with weekend working. †Hospital 2 included 12 routine clinics, 3 days per 18 week (original service), and 14 in the modified service (which also included weekend working).  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   23   Tables   1  2   Table 1. Patient characteristics as modelled   3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29 30 SD = Standard deviation; TIA = Transient ischemic attack; GP = general practitioner. * Expected 31 numbers are rounded to the nearest integer, the apparent anomaly with the addition results from this 32 rounding. High risk patients defined as an ABCD 2 score of >4  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  Table 2. Pattern of outpatient clinics for suspected TIA based on actual service provision at participating hospitals during patient recruitment period.
*Clinics are divided among 4 specialists, each of whom were absent for approximately 7 weeks a year (annual leave). These clinics are assumed not to take place if the specialist is absent. †Specialists were absent for approximately 7 weeks a year (annual leave). All absent clinicians were replaced by a specialist from another site within the Trust. ‡There is a 50% probability each week that either 4 or 6 clinic slots will be available. All high risk patients are seen on the ward as outpatients as required (including at weekends). Patients referred before 10am are seen at 5pm on the same day, patients referred after 10am are seen at 10am on the following day. All low risk patients seen at the next available clinic in order of referral.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  Table 3. Costs, resource utilisation and outcomes (per year) of modifying TIA service provision in hospital 1. were compared. †High risk breaches were defined as high risk patients not seen by a specialist within 24 hours of initial clinic referral. 12 ǂ Low risk breaches were defined as low risk patients not seen by a specialist within 7 days of initial clinic referral.  High risk patient: male age 70-74, SBP 156mmHg, Speech disturbance without weakness, Duration of symptoms 60+ mins, not diabetic (resulting in ABCD and ABCD2 scores of 5), undiagnosed AF, cholesterol 5.8. Low risk patient: male age 70-74, SBP 115mmHg, Speech disturbance without weakness, Duration of symptoms 0-9 mins, not diabetic (resulting in ABCD and ABCD2 of scores 2), AF on warfarin, cholesterol 6.0.

Original
172x160mm (300 x 300 DPI)   The availability of specialist TIA clinics was modelled on the basis of existing service provision at participating hospital sites during the study period. Patients could either be seen in traditional outpatient clinics, admitted, or seen on the ward on an outpatient basis. The original services were designed within the confines of available clinical staff and a lack of specialist cover at the weekends. Both services subsequently underwent redesign and the impact of these changes was examined by modelling both the original and modified service. The number and distribution of clinics in each service are described in Table 2.
The impact of further adjustments to the modified service at each site was modelled to replicate the following scenarios: a) The addition of high risk clinic slots for patients presenting on a Saturday and Sunday b) Including a weekend service but reducing the number of routine weekday clinics, by up to 5 clinic slots per week.  The model estimated the impact of existing service provision, the effect of recent service change and the impact of further alterations to clinic availability on service costs/provision and subsequent stroke events, from a health services perspective.

Overview of the model
Clinical pathways in the model Following onset of an initial event, patients were assumed to contact either their general practitioner (GP) or attend the emergency department (ED) (eFigure 1, online supplemental material). The probability that a patient would choose a specific route was dependent on the type of patient (high/low risk), the time of day, and day of week, estimated using data from the Oxford Vascular Study. 3 Time taken to reach the service depended on the route chosen; where patients directly attended the ED, the time was sampled from a Weibull distribution fitted to data from the Newcastle Stroke Service. 2 Patients attending their GP were assumed to arrive within daytime clinic hours. 3 Following this initial contact with a healthcare professional, patients were referred to a specialist TIA outpatient clinic, seen on the ward (as an outpatient) or admitted as an inpatient. The type of referral was dependant on the specific hospital service provision, clinic availability and the level of risk of the attending patient, defined according to the ABCD 2 score. 4 During clinic attendance, it was assumed that the appropriate treatment would be initiated (i.e. blood pressure lowering, cholesterol lowering or antiplatelet therapy in accordance with guidelines) and a small proportion of patients (4.1%) would be deemed suitable for carotid endarterectomy. 5,6 Model population Overall rates of patient presentation were based on initial runs of the model and ensured a realistic distribution of final diagnoses which represented the study sample, previous literature 7 and expert opinion (60% TIA mimics, 33% genuine TIA, and 7% minor stroke). In the absence of robust data for the distribution of new referrals through the week, it was assumed that new cases arrived uniformly.
Due to missing risk data in the sample population, the observed ratio of low risk/high risk TIA patients used as basecase in the model was supplemented in sensitivity analyses by estimates derived from those reported in previous studies 4,8,9

Follow-up and risk of repeat events
Patients remained in the model until one year from symptom onset or until they suffered a repeat event (within the same year). Repeat TIA events meant restarting the clinical pathway, subsequent stroke was counted as an outcome and costed accordingly. The risk of a repeat event (TIA or major stroke) was dependent on the initial event (true TIA or mimic), the severity of that event (measured by ABCD 2 score) and other relevant risk factors such as age, presence of atrial fibrillation and medication prescribed. 4, 6, 10-15 Risks were modelled using a Weibull distribution for the time to event which allowed for a substantially increased risk in the short term, followed by a decreasing risk over time. 16 Patients were considered to have a risk of "other cause" death, on the basis of age group and gender, based on national statistics for "all cause" death adjusted for stroke deaths, [17][18][19] allowing a date of death to be sampled for each individual patient. In the vast majority of cases, the date of other cause death was beyond the follow-up time in the model.
The model was run 100 times for a total simulated time of 12 years in each run: the first year of the run was regarded as a "warm up" period to allow the system to reach a steady state before collecting data. Patients entering during the next 10 years were included in the model results, with the final year being necessary to allow for a one-year follow up time for all included patients. Results for patients entering the model during this last "follow up" period were not included, but such patients were modelled to ensure that the constraints on clinic availability were maintained. Results are presented in terms of annual costs and other outcomes.

Costs and outcomes
The model included costs of GP visits, transport by ambulance to the ED, ED attendance, outpatient clinics, hospital admission for TIA and stroke, surgery and therapy (eTable 1 and 2, online data supplement). Costs included in the model were taken from a combination of NHS reference costs, 20,21 and drug costs from the British National Formulary. 22 The price year for all costs was 2011-12.
The primary outcome of this study was number of major strokes occurring post TIA. Secondary outcomes included the overall service costs per year and attainment of national targets for TIA service provision. Target attainment was defined as the number of high and low risk 'breaches' which occurred in each service per year: high risk breaches were defined as a high risk patient not seen by a specialist within 24 hours of initial referral. 5,23 Low risk breaches were defined as low risk patients not seen by a specialist within seven days of referral. 5,23 Further outcomes examined the median time from referral to specialist appointment, the total number of routine outpatient  99% was chosen because of multiple values were compared †High risk breaches were defined as high risk patients not seen by a specialist within 24 hours of initial clinic referral. ǂ Low risk breaches were defined as low risk patients not seen by a specialist within 7 days of initial clinic referral.    Identify the study as an economic evaluation or use more specific terms such as "cost-effectiveness analysis", and describe the interventions compared. Abstract 2 Provide a structured summary of objectives, perspective, setting, methods (including study design and inputs), results (including base case and uncertainty analyses), and conclusions.

Background and objectives
3 Provide an explicit statement of the broader context for the study. Present the study question and its relevance for health policy or practice decisions. Describe what outcomes were used as the measure(s) of benefit in the evaluation and their relevance for the type of analysis performed.

Measurement of effectiveness
11a Single study-based estimates: Describe fully the design features of the single effectiveness study and why the single study was a sufficient source of clinical effectiveness data. 11b Synthesis-based estimates: Describe fully the methods used for identification of included studies and synthesis of clinical effectiveness data. Measurement and valuation of preference based outcomes 12 If applicable, describe the population and methods used to elicit preferences for outcomes.

Estimating resources and costs
13a Single study-based economic evaluation: Describe approaches used to estimate resource use associated with the alternative interventions. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs. 13b Model-based economic evaluation: Describe approaches and data sources used to estimate resource use associated with model health states. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs. Currency, price date, and conversion 14 Report the dates of the estimated resource quantities and unit costs. Describe methods for adjusting estimated unit costs to the year of reported costs if necessary. Describe methods for converting costs into a common currency base and the exchange rate. Describe all analytical methods supporting the evaluation. This could include methods for dealing with skewed, missing, or censored data; extrapolation methods; methods for pooling data; approaches to validate or make adjustments (such as half cycle corrections) to a model; and methods for handling population heterogeneity and uncertainty.

Study parameters 18
Report the values, ranges, references, and, if used, probability distributions for all parameters. Report reasons or sources for distributions used to represent uncertainty where appropriate. Providing a table to show the input values is strongly recommended. Incremental costs and outcomes 19 For each intervention, report mean values for the main categories of estimated costs and outcomes of interest, as well as mean differences between the comparator groups. If applicable, report incremental cost-effectiveness ratios. Characterising uncertainty 20a Single study-based economic evaluation: Describe the effects of sampling uncertainty for the estimated incremental cost and incremental effectiveness parameters, together with the impact of methodological assumptions (such as discount rate, study perspective). 20b Model-based economic evaluation: Describe the effects on the results of uncertainty for all input parameters, and uncertainty related to the structure of the model and assumptions. Characterising heterogeneity 21 If applicable, report differences in costs, outcomes, or costeffectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.

Discussion
Study findings, limitations, generalisability, and current knowledge 22 Summarise key study findings and describe how they support the conclusions reached. Discuss limitations and the generalisability of the findings and how the findings fit with current knowledge.

Other
Source of funding 23 Describe how the study was funded and the role of the funder in the identification, design, conduct, and reporting of the analysis. Describe other non-monetary sources of support.

Conflicts of interest 24
Describe any potential for conflict of interest of study contributors in accordance with journal policy. In the absence of a journal policy, we recommend authors comply with International Committee of Medical Journal Editors recommendations.

10
Outcome measures: The primary outcome of the model was the prevalence of major stroke after 11 TIA. Secondary outcomes included service costs (including those of treating subsequent stroke) and 12 time to treatment and attainment of national targets for service provision (proportion of high risk 13 patients [according to ABCD 2 score] seen within 24 hours). Modelling the impact of further hypothetical service reconfiguration, including a seven day service, 1 suggested that it was possible to reduce service costs and the number of unused clinic appointments 2 and improve guideline target attainment, but this had little impact on the incidence of major stroke 3 post TIA or post-stroke deaths (table 3 and eTable 3). Figure 2 shows the trade-off between routine 4 appointment provision, unscheduled appointments and high/low risk guideline breaches. Assuming

15
The aim of this study was to model the impact of TIA service reconfiguration in two large urban 16 hospitals and establish the impact on clinical outcomes (major strokes), costs, and clinic usage. 17 Reconfiguration of services was found to be effective and appropriate for reducing resource (clinic 18 slot) waste, and improving guideline target attainment with variable effects on costs. However, in 19 hospitals where services, despite poor target attainment, had minimal patient delays, such as those 20 studied here, significant reconfiguration had little impact on modelled clinical outcomes such as 21 major stroke post TIA. These findings suggest that clinicians and service coordinators should be 22 cautious before initiating significant reconfiguration of services which are already seeing high risk 23 patients within 1-2 days, unless there is an obvious need to free up resources which could be used 24 for another purpose (e.g. managing other types of patients whilst not incentivising small changes that might increase costs for no measurable clinical benefit. 2 3 Strengths and limitations 4 The present study used local hospital data to describe the characteristics of the presenting 5 population and thus the results are likely to be representative of the hospitals studied, but not 6 necessarily generalizable to other centres in the UK and across the world. Despite this, guideline 7 target attainment in hospital 1 (original service configuration) was similar to that reported nationally 8 (31% vs. 37% nationally) 13 which suggests that this hospital was more likely representative of 9 centres across the UK. 10 11 Modelling allows estimation of service 'unknowns', such as the number of unscheduled clinic 12 appointments required to meet guideline targets for high risk patients. This requires pre- 13 programming with hypothetical patients and specific rules; therefore simplification of certain 14 service intricacies is necessary. For example, the process by which high risk patients were allocated 15 unscheduled clinic appointments was formalised such that a patient referred by the ED physician 16 was assumed to have been seen by a stroke specialist within two hours. However, patients referred 17 by their GP were assumed to be seen by the stroke specialist on the ward at 10am the following 18 morning or 5pm the same day, whichever was later (but within 24 hours of initial referral). In 19 reality, services are often adapted in response to a specific set of circumstances such as clinic room 20 availability and thus the services modelled here, and the patients attending these services, can best 21 be thought of as stylised versions of the actual services available and patients attending the hospitals 22 in question. 23 24 The risk of recurrent stroke attributed to hypothetical patients in the model was based on an 25 estimated ABCD 2 score. 10 The accuracy of this score for predicting recurrent stroke has been called 26 at low and high risk of early (7 day) stroke. 30 However, this was the score used and recommended 2 in practice at the time these services were modelled and was used for all service configurations 3 examined, so our relative comparisons are likely to remain valid, even if the absolute numbers may 4 be subject to change.

6
The costs modelled here were taken from standard NHS reference costs 27 28 and included 7 investigations, treatment and the cost of staff attending patients during each appointment/admission. 8 Costs relate to standard weekday services and increased rates for weekend services were 9 unavailable for this analysis. It is therefore likely that the costs of providing a weekend service may 10 have been underestimated in the present analyses. 11 12 Findings in the context of previous literature and implications for practice 13 Economic modelling has previously been used for determining the impact of service reconfiguration 14 in acute stroke 31-34 but studies examining optimal strategies for TIA service provision are less 15 common. Those that exist, focus on whether or not patients should be admitted or be seen as an 16 outpatient. 16 35 36 It is generally accepted that patient admission for TIA is inefficient and not cost- the Sentinel Stroke National Audit Programme (SSNAP) in the UK. 13 A recent review identified six 20 models of care designed to avoid patient admission for TIA. These ranged from patient assessment 21 and triage in the ED, a 24-hour nurse-led telephone advice service and the primary care triage 22 system studied here (+/-electronic decision support). 41 The appropriate model of care will depend 23 on the local healthcare system, but consideration of how such a model is implemented in terms of 24 service configuration is warranted, regardless of the system used.  The primary finding of this study was that whilst service reconfiguration can significantly improve hours' target for high risk patients were originally based on data from the EXPRESS study, 9 which 8 showed that those patients seen and treated within 24 hours of initial referral had significantly better 9 outcomes than those patients who were not. However, in the original service to which this optimal 10 strategy was compared, the average time from referral to clinic appointment was 3 days (IQR 2-5 11 days) and time to first treatment was 20 days (IQR 8-53). The improvement in time to initiation of 12 treatment was therefore 20 days, compared to an improvement of 8 hours seen in the present study, 13 which explains the very different results in terms of clinical outcome. 14 15 Recent results from SSNAP show that 45% of high risk outpatients and 60% of high risk inpatients 16 are seen and treated within 24 hours of referral, suggesting there is plenty of room for improvement. 17 However, with a median time to first clinic appointment (and treatment) for all patients of just 2 18 days, the present analyses would suggest further improvements in guideline target attainment may 19 have little impact on major stroke following TIA. 20 21 The modified services examined and proposed here may require the capacity to accommodate 22 unscheduled clinic appointments, which is likely to have an opportunity cost affecting the quality of 23 other services running concurrently in the hospital and may result in increased costs with little 24 tangible benefit. In particular, the ability to accommodate such appointments depends on the time of 25 day at which the patient presents, the availability of specialist stroke physicians and scanning Data sharing 6 The datasets analysed during the current study are available from the corresponding author on 7 reasonable request. 8 9 Competing interests 10 The authors declare that they have no competing interests          week (original service), and 14 in the modified service (which also included weekend working).   Table 2. Pattern of outpatient clinics for suspected TIA based on actual service provision at participating hospitals during patient recruitment period.
*Clinics are divided among 4 specialists, each of whom were absent for approximately 7 weeks a year (annual leave). These clinics are assumed not to take place if the specialist is absent. †Specialists were absent for approximately 7 weeks a year (annual leave). All absent clinicians were replaced by a specialist from another site within the Trust. ‡There is a 50% probability each week that either 4 or 6 clinic slots will be available.

Hospital and service
Number of routine clinic slots available by day of the week All high risk patients are seen on the ward as outpatients as required (including at weekends). Patients referred before 10am are seen at 5pm on the same day, patients referred after 10am are seen at 10am on the following day. All low risk patients seen at the next available clinic in order of referral.

Setting and service design
The model was designed to assess the costs and clinical consequences arising from patients suspected of suffering an acute TIA and referred to two large urban hospitals in the West Midlands region of the UK. Both hospitals ran a specialist TIA outpatient service catering for approximately 500 patients with suspected stroke and TIA every year. The model parameters were defined by the characteristics of individuals attending these services and recruited to an observational study 1 during 2011 (table 1).
The availability of specialist TIA clinics was modelled on the basis of existing service provision at participating hospital sites during the study period. Patients could either be seen in traditional outpatient clinics, admitted, or seen on the ward on an outpatient basis. The original services were designed within the confines of available clinical staff and a lack of specialist cover at the weekends. Services were redesigned at each site during the study period and the impact of these changes was examined by modelling both the original and modified service. The number and distribution of clinics in each service are described in Table 2.
The impact of further adjustments to the modified service at each site was modelled to replicate the following scenarios: a) The addition of high risk clinic slots for patients presenting on a Saturday and Sunday b) Including a weekend service but reducing the number of routine weekday clinics, by up to 5 clinic slots (one patient per slot) per week.

Overview of the model
A discrete event simulation model was programmed in Delphi version 4 (Borland, San Francisco, CA, USA). It was adapted from a previously developed model and further detail regarding the general modelling framework can be found in the original report. 2 Briefly, the model generated a number of possible (virtual) patient histories which began at the onset of TIA (or TIA like) symptoms and followed the patient along the clinical pathway from initial presentation to referral, specialist TIA clinic attendance, treatment and lifetime follow-up for subsequent stroke morbidity and mortality. An essential feature of the model was that patients were sharing limited resources in the form of routine clinics. The model estimated the impact of existing service provision, the effect of recent service change and the impact of further alterations to clinic availability on service costs/provision and subsequent stroke events, from a health services perspective.

Clinical pathways in the model
Following onset of an initial event, patients were assumed to contact either their general practitioner (GP) or attend the emergency department (ED) (eFigure 1, online supplemental material). The probability that a patient would choose a specific route was dependent on the type of patient (high/low risk), the time of day, and day of week, estimated using data from the Oxford Vascular Study. 3 Time taken to reach the service depended on the route chosen; where patients directly attended the ED, the time was sampled from a Weibull distribution fitted to data from the Newcastle Stroke Service. 2 Patients attending their GP were assumed to arrive within daytime clinic hours. 3 Following this initial contact with a healthcare professional, patients were referred to a specialist TIA outpatient clinic, seen on the ward (as an outpatient) or admitted as an inpatient. The type of referral was dependant on the specific hospital service provision, clinic availability and the level of risk of the attending patient, defined according to the ABCD 2 score. 4 During clinic attendance, it was assumed that the appropriate treatment would be initiated (i.e. blood pressure lowering, cholesterol lowering or antiplatelet therapy in accordance with guidelines) and a small proportion of patients (4.1%) would be deemed suitable for carotid endarterectomy. 5,6 Patients were assumed to take this treatment as prescribed and gain the full benefit in terms of stroke prevention.

Model population
All patients in the model were hypothetical, although they we based on data from real patients and hospitals. Overall rates of patient presentation were based on initial runs of the model and ensured a realistic distribution of final diagnoses which represented the study sample, previous literature 7 and expert opinion (60% TIA mimics, 33% genuine TIA, and 7% minor stroke). High and low risk TIAs were defined according to the ABCD 2 score. 4 In the absence of robust data for the distribution of new referrals through the week, it was assumed that new cases arrived uniformly.
Due to missing risk data in the sample population, the observed ratio of low risk/high risk TIA

Follow-up and risk of repeat events
Hypothetical patients remained in the model until one year from symptom onset, after which the increased risk of repeat event returns close to normal, 10,11 unless they died or suffered a non-fatal disabling stroke. No distinction was made in the model between fatal strokes and non-fatal disabling strokes: these were labelled "major strokes". The risk of a repeat event (TIA or stroke) was dependent on the initial event (minor stroke, true TIA or mimic), the severity of that event (measured by ABCD 2 score) and other relevant risk factors such as age, presence of atrial fibrillation and medication prescribed. 4,6,12-17 Following a minor (non-disabling) stroke, patients remained in the model, but with an additional risk of mortality that could be reduced by appropriate treatment. Additional deaths from this cause were estimated and labelled in the model outputs as "post-stroke deaths". Modelled outputs are therefore derived from risk profile of the hypothetical patients, adjusting for the effects of treatments which would start at varying times in the different scenarios. Risks were modelled using a Weibull distribution for the time to event which allowed for a substantially increased risk in the short term, followed by a decreasing risk over time. 18 Examples of the modelled risk for two patients (with high and low risk characteristics) are given in figure 1.
The model was run 100 times for a total simulated time of 12 years in each run: the first year of the run was regarded as a "warm up" period to allow the system to reach a steady state before collecting data. Patients entering during the next 10 years were included in the model results, with the final year being necessary to allow for a one-year follow up time for all included patients. Results for patients entering the model during this last "follow up" period were not included, but such patients were modelled to ensure that the constraints on clinic availability were maintained. Results are presented in terms of annual costs and other outcomes.

Costs and outcomes
The model included costs of GP visits, transport by ambulance to the ED, ED attendance, outpatient clinics, hospital admission for TIA and stroke, surgery and therapy (eTable 1 and 2, online data supplement). Costs included in the model were taken from a combination of NHS reference costs, 19,20 and drug costs from the British National Formulary. 21 The price year for all costs was 2011-12.
The primary outcome was the number of expected major strokes occurring post TIA, based on risk analysis. Secondary outcomes included the overall service costs per year and attainment of national targets for TIA service provision. Target attainment was defined as the number of high and low risk 'breaches' which occurred in each service per year: high risk breaches were defined as a high risk patient not seen by a specialist within 24 hours of initial referral. 5,22 Low risk breaches were defined as low risk patients not seen by a specialist within seven days of referral. 5,22 Further outcomes examined the median time from referral to specialist appointment, the total number of routine outpatient appointments available (used or unused) and any unscheduled outpatient appointments required (where high risk patients were assessed immediately on the ward).
All data are presented as means or medians ± standard deviation (SD), inter-quartile range (IQR) or 99% confidence intervals (CI), chosen because multiple values are compared. Percentages are given for the total population unless otherwise stated.  99% was chosen because of multiple values were compared †High risk breaches were defined as high risk patients not seen by a specialist within 24 hours of initial clinic referral. These could not occur with the service pattern, as all high risk patients were admitted in the original service, and seen on the ward in the modified service. ǂ Low risk breaches were defined as low risk patients not seen by a specialist within 7 days of initial clinic referral.  If applicable, report differences in costs, outcomes, or costeffectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.  2 3 Modelling changes to service provision for transient ischemic attack 4 5 Barton, P., 1 Sheppard, J.P., 2 Penaloza-Ramos, M.C., 1 Jowett, S., 1 Ford, G.A., 3 Lasserson, D., 4 Mant,6 J., 5 Mellor, R.M., 6 Quinn, T., 7 Rothwell, P.M., 8 Sandler, D., 9 Sims, D., 10     Participants: Patients with suspected TIA were followed from symptom onset to presentation, 7 referral to specialist clinics, treatment and subsequent stroke. Interventions: Included existing versus previous (less same day clinics) and hypothetical service 9 reconfiguration (7-day service with less availability of clinics per day).

10
Outcome measures: The primary outcome of the model was the prevalence of major stroke after 11 TIA. Secondary outcomes included service costs (including those of treating subsequent stroke) and 12 time to treatment and attainment of national targets for service provision (proportion of high risk 13 patients [according to ABCD 2 score] seen within 24 hours). • This study focused on two hospitals within the West Midlands, UK, so the results are likely to 3 be representative of these hospitals, but not necessarily other centres in the UK and across the 4 world.

5
• Modelling allows estimation of service 'unknowns' which requires pre-programming and 6 therefore simplification of certain service intricacies. Thus, the services modelled represent 7 stylised versions of the actual services offered by participating hospitals. 8 • Costs relate to standard weekday services and it is not clear whether those actually incurred 9 would be increased by provision of services at the weekend. 10 • The potential costs of implementation of new services, for example those for weekend services, 11 were unavailable and so the costs presented here may have been underestimated. Rapid recognition and treatment of TIA patients at high risk of having a subsequent stroke is 8 important because simple interventions, such as early prescription of preventative medications, can 9 substantially reduce the risk of stroke following TIA, and existing evidence suggests that earlier 10 intervention may be better. 9 10 In the UK, guidelines 11 12 recommend that patients at high risk of 11 recurrent stroke (defined using the ABCD 2 score of >4) 10 are seen within 24 hours of symptom 12 onset and all other patients are seen within 7 days. A recent UK audit suggests there has been 13 significant improvement in attainment of these targets: 45% of high risk outpatients and 60% of 14 inpatients now receive treatment on the same day as referral, compared with just 10% and 33% four 15 years earlier. 13 This improved target attainment has been achieved through a variety of approaches, 16 with some centres admitting more patients for assessment and treatment and others designing
The availability of specialist TIA clinics was modelled on the basis of existing service provision at 9 participating hospital sites during the study period. Patients could either be seen in traditional 10 outpatient clinics, admitted, or seen on the ward on an outpatient basis. The original services were 11 designed within the confines of available clinical staff and in particular a lack of specialist cover at 12 the weekends. Services were redesigned at each site during the study period and the impact of these 13 changes on cost and outcomes was examined by modelling both the original and modified service. 14 The number and distribution of clinics in each service are described in table 2.

16
The impact of further hypothetical adjustments to the modified service at each site was modelled to 17 replicate the following scenarios:  Overview of the model 23 A discrete event simulation model was programmed in Delphi version 4 (Borland, San Francisco, 24 CA, USA). It was adapted from a previously developed model and further detail regarding the 25 general modelling framework can be found in the original report. 16 Briefly, the model generated a  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  symptoms and followed the patient along the clinical pathway from initial presentation to follow-up 2 for subsequent stroke morbidity and mortality. An essential feature of the model was that patients 3 were sharing limited resources in the form of routine clinics. 4 5 Clinical pathways in the model 6 Following onset of an initial event, patients were assumed to contact either their general practitioner 7 (GP) or attend the emergency department (ED) (eFigure 1, online supplemental material). The 8 probability that a patient would choose a specific route was dependent on the type of patient 9 (high/low risk), the time of day, and day of week, estimated using data from the Oxford Vascular 10 Study. 17 Following this initial contact with a healthcare professional, patients were referred to a 11 specialist TIA outpatient clinic, seen on the ward (as an outpatient) or admitted as an inpatient. The 12 type of referral was dependant on the specific hospital service provision, clinic availability and the 13 level of risk of the attending patient, defined according to the ABCD 2 score. 10 During clinic 14 attendance, it was assumed that the appropriate treatment would be initiated (i.e. blood pressure 15 lowering, cholesterol lowering or antiplatelet therapy in accordance with guidelines) and a small 16 proportion of patients (4.1%) would be treated with carotid endarterectomy. 12 18 Patients were 17 assumed to take this treatment as prescribed and gain the full benefit in terms of stroke prevention. 18 19 Model population 20 All patients in the model were hypothetical, although they were based on data from real patients and 21 hospitals. Overall rates of patient presentation were based on initial runs of the model and ensured a 22 realistic distribution of final diagnoses which represented the study sample, previous literature 14 and 23 expert opinion (60% TIA mimics, 33% genuine TIA, and 7% minor stroke). High and low risk 24 TIAs were defined according to the ABCD 2 score. 10 The observed ratio of low risk/high risk TIA scenarios. Risks were modelled using a Weibull distribution for the time to event which allowed for 18 a substantially increased risk in the short term, followed by a decreasing risk over time. 9 Examples 19 of the modelled risk for two patients (with high and low risk characteristics) are given in figure 1.

20
The model was run 100 times for a total simulated time of 12 years in each run. 21 22 Costs and outcomes 23 The model included costs of any GP visit (from presentation to referral), or transport by ambulance 24 to the ED and ED attendance, outpatient clinics or hospital admission for TIA and stroke, surgery 25 and therapy (eTables 1 and 2, online data supplement). Pre-hospital and treatment related costs were  were defined as a high risk patient not seen by a specialist within 24 hours of initial referral. 11 12 11 Low risk breaches were defined as low risk patients not seen by a specialist within seven days of 12 referral. 11 12 Further outcomes examined the median time from referral to specialist appointment, the 13 total number of routine outpatient appointments available (used or unused) and any unscheduled 14 outpatient appointments required (where high risk patients were assessed immediately on the ward). 15 16 All data are presented as means or medians ± standard deviation (SD), inter-quartile range (IQR) or 17 99% confidence intervals (CI), chosen because multiple values are compared. Percentages are given 18 for the total population unless otherwise stated. 19 20

21
TIA service usage 22 The model estimated a total of 490 patients would be referred to specialist TIA clinics each year at   Potential optimal service provision 4 Modelling the impact of further hypothetical service reconfiguration, including a seven day service, 5 suggested that it was possible to reduce service costs and the number of unused clinic appointments 6 and improve guideline target attainment, but this had little impact on the incidence of major stroke 7 post TIA or post-stroke deaths (table 3 and eTable 3). Figure 2 shows the trade-off between routine 8 appointment provision, unscheduled appointments and high/low risk guideline breaches. Assuming

19
The aim of this study was to model the impact of TIA service reconfiguration in two large urban 20 hospitals and establish the impact on clinical outcomes (major strokes), costs, and clinic usage. 21 Reconfiguration of services was found to be effective and appropriate for reducing resource (clinic whilst not incentivising small changes that might increase costs for no measurable clinical benefit. 5 6 Strengths and limitations 7 The present study used local hospital data to describe the characteristics of the presenting 8 population and thus the results are likely to be representative of the hospitals studied, but not 9 necessarily generalizable to other centres in the UK and across the world. Despite this, guideline 10 target attainment in hospital 1 (original service configuration) was similar to that reported nationally 11 (31% vs. 37% nationally) 13 which suggests that this hospital was more likely representative of 12 centres across the UK. 13 14 Modelling allows estimation of service 'unknowns', such as the number of unscheduled clinic 15 appointments required to meet guideline targets for high risk patients. This requires pre- 16 programming with hypothetical patients and specific rules; therefore simplification of certain 17 service intricacies is necessary. For example, the process by which high risk patients were allocated 18 unscheduled clinic appointments was formalised such that a patient referred by the ED physician 19 was assumed to have been seen by a stroke specialist within two hours. However, patients referred 20 by their GP were assumed to be seen by the stroke specialist on the ward at 10am the following 21 morning or 5pm the same day, whichever was later (but within 24 hours of initial referral). In The risk of recurrent stroke attributed to hypothetical patients in the model was based on an 2 estimated ABCD 2 score. 10 The accuracy of this score for predicting recurrent stroke has been called 3 into question with a recent systematic review showing it to have poor discrimination between those 4 at low and high risk of early (7 day) stroke. 30 However, this was the score used and recommended 5 in practice at the time these services were modelled and was used for all service configurations 6 examined, so our relative comparisons are likely to remain valid, even if the absolute numbers may 7 be subject to change. The costs modelled here were taken from standard NHS reference costs 27 28 and included 10 investigations, treatment and the cost of staff attending patients during each appointment/admission.

11
Costs relate to standard weekday services and increased rates for weekend services were 12 unavailable for this analysis. It is therefore likely that the costs of providing a weekend service may 13 have been underestimated in the present analyses. 14 15 Findings in the context of previous literature and implications for practice 16 Economic modelling has previously been used for determining the impact of service reconfiguration 17 in acute stroke 31-34 but studies examining optimal strategies for TIA service provision are less 18 common. Those that exist, focus on whether or not patients should be admitted or be seen as an 19 outpatient. 16 35 36 It is generally accepted that patient admission for TIA is inefficient and not cost- the Sentinel Stroke National Audit Programme (SSNAP) in the UK. 13 A recent review identified six 23 models of care designed to avoid patient admission for TIA. These ranged from patient assessment 24 and triage in the ED, a 24-hour nurse-led telephone advice service and the primary care triage 25 system studied here (+/-electronic decision support on the local healthcare system, but consideration of how such a model is implemented in terms of 1 service configuration is warranted, regardless of the system used. 2 3 The primary finding of this study was that whilst service reconfiguration can significantly improve hours' target for high risk patients were originally based on data from the EXPRESS study, 9 which 11 showed that those patients seen and treated within 24 hours of initial referral had significantly better 12 outcomes than those patients who were not. However, in the original service to which this optimal 13 strategy was compared, the average time from referral to clinic appointment was 3 days (IQR 2-5 14 days) and time to first treatment was 20 days (IQR 8-53). The improvement in time to initiation of 15 treatment was therefore 20 days, compared to an improvement of 8 hours seen in the present study, 16 which explains the very different results in terms of clinical outcome. 17 18 Recent results from SSNAP show that 45% of high risk outpatients and 60% of high risk inpatients 19 are seen and treated within 24 hours of referral, suggesting there is plenty of room for improvement. 20 However, with a median time to first clinic appointment (and treatment) for all patients of just 2 21 days, the present analyses would suggest further improvements in guideline target attainment may 22 have little impact on major stroke following TIA. 23 24 The modified services examined and proposed here may require the capacity to accommodate 25 unscheduled clinic appointments, which is likely to have an opportunity cost affecting the quality of 26 Data sharing 6 The datasets analysed during the current study are available from the corresponding author on 7 reasonable request. 8 9 Competing interests 10 The authors declare that they have no competing interests      1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59   week (original service), and 14 in the modified service (which also included weekend working).  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  Table 2. Pattern of outpatient clinics for suspected TIA based on actual service provision at participating hospitals during patient recruitment period.

Original
The impact of further adjustments to the modified service at each site was modelled to replicate the following scenarios: a) The addition of high risk clinic slots for patients presenting on a Saturday and Sunday b) Including a weekend service but reducing the number of routine weekday clinics, by up to 5 clinic slots (one patient per slot) per week.

Overview of the model
A discrete event simulation model was programmed in Delphi version 4 (Borland, San Francisco, CA, USA). It was adapted from a previously developed model and further detail regarding the general modelling framework can be found in the original report. 2 Briefly, the model generated a number of possible (virtual) patient histories which began at the onset of TIA (or TIA like) symptoms and followed the patient along the clinical pathway from initial presentation to referral, specialist TIA clinic attendance, treatment and lifetime follow-up for subsequent stroke morbidity and mortality. An essential feature of the model was that patients were sharing limited resources in the form of routine clinics.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  The model estimated the impact of existing service provision, the effect of recent service change and the impact of further alterations to clinic availability on service costs/provision and subsequent stroke events, from a health services perspective.

Clinical pathways in the model
Following onset of an initial event, patients were assumed to contact either their general practitioner (GP) or attend the emergency department (ED) (eFigure 1, online supplemental material). The probability that a patient would choose a specific route was dependent on the type of patient (high/low risk), the time of day, and day of week, estimated using data from the Oxford Vascular Study. 3 Time taken to reach the service depended on the route chosen; where patients directly attended the ED, the time was sampled from a Weibull distribution fitted to data from the Newcastle Stroke Service. 2 Patients attending their GP were assumed to arrive within daytime clinic hours. 3 Following this initial contact with a healthcare professional, patients were referred to a specialist TIA outpatient clinic, seen on the ward (as an outpatient) or admitted as an inpatient. The type of referral was dependant on the specific hospital service provision, clinic availability and the level of risk of the attending patient, defined according to the ABCD 2 score. 4 During clinic attendance, it was assumed that the appropriate treatment would be initiated (i.e. blood pressure lowering, cholesterol lowering or antiplatelet therapy in accordance with guidelines) and a small proportion of patients (4.1%) would be deemed suitable for carotid endarterectomy. 5,6 Patients were assumed to take this treatment as prescribed and gain the full benefit in terms of stroke prevention.

Model population
All patients in the model were hypothetical, although they were based on data from real patients and hospitals. Overall rates of patient presentation were based on initial runs of the model and ensured a realistic distribution of final diagnoses which represented the study sample, previous literature 7 and expert opinion (60% TIA mimics, 33% genuine TIA, and 7% minor stroke). High and low risk TIAs were defined according to the ABCD 2 score. 4 In the absence of robust data for the distribution of new referrals through the week, it was assumed that new cases arrived uniformly.
Due to missing risk data in the sample population, the observed ratio of low risk/high risk TIA patients used as basecase in the model was supplemented in sensitivity analyses by estimates derived from those reported in previous studies 4,8,9 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Hypothetical patients remained in the model until one year from symptom onset, after which the increased risk of repeat event returns close to normal, 10,11 unless they died or suffered a non-fatal disabling stroke. No distinction was made in the model between fatal strokes and non-fatal disabling strokes: these were labelled "major strokes". The risk of a repeat event (TIA or stroke) was dependent on the type of initial event (minor stroke, true TIA or mimic), the ABCD 2 based risk prediction of a subsequent event and other relevant risk factors such as age, presence of atrial fibrillation and medication prescribed. 4,6,12-17 Following a minor (non-disabling) stroke, patients remained in the model, but with an additional risk of mortality that could be reduced by appropriate treatment. Additional deaths from this cause were estimated and labelled in the model outputs as "post-stroke deaths". Modelled outputs are therefore derived from risk profile of the hypothetical patients, adjusting for the effects of treatments which would start at varying times in the different scenarios. Risks were modelled using a Weibull distribution for the time to event which allowed for a substantially increased risk in the short term, followed by a decreasing risk over time. 18 Examples of the modelled risk for two patients (with high and low risk characteristics) are given in figure 1.
The model was run 100 times for a total simulated time of 12 years in each run: the first year of the run was regarded as a "warm up" period to allow the system to reach a steady state before collecting data. Patients entering during the next 10 years were included in the model results, with the final year being necessary to allow for a one-year follow up time for all included patients. Results for patients entering the model during this last "follow up" period were not included, but such patients were modelled to ensure that the constraints on clinic availability were maintained. Results are presented in terms of annual costs and other outcomes.

Costs and outcomes
The model included costs of any GP visit (from presentation to referral), or transport by ambulance to the ED and ED attendance, outpatient clinics or hospital admission for TIA and stroke, surgery and therapy (eTables 1 and 2, online data supplement). Pre-hospital and treatment related costs were the same for all options considered in this study since the focus was to compare comparative differences between modelled services caused by different clinic configurations. Costs included in the model were taken from a combination of NHS reference costs, 19,20 and drug costs from the British National Formulary. 21 The price year for all costs was 2011-12.
The primary outcome was the number of expected major strokes occurring post TIA, based on risk analysis. Secondary outcomes included the overall service costs per year and attainment of national targets for TIA service provision. Target attainment was defined as the number of high and low risk 'breaches' which occurred in each service per year: high risk breaches were defined as a high risk patient not seen by a specialist within 24 hours of initial referral. 5,22 Low risk breaches were defined as low risk patients not seen by a specialist within seven days of referral. 5,22 Further outcomes examined the median time from referral to specialist appointment, the total number of routine outpatient appointments available (used or unused) and any unscheduled outpatient appointments required (where high risk patients were assessed immediately on the ward).

Target population and subgroups 4
Describe characteristics of the base case population and subgroups analysed, including why they were chosen. Setting and location 5 State relevant aspects of the system(s) in which the decision(s) need(s) to be made. Study perspective 6 Describe the perspective of the study and relate this to the costs being evaluated. Comparators 7 Describe the interventions or strategies being compared and state why they were chosen. Time horizon 8 State the time horizon(s) over which costs and consequences are being evaluated and say why appropriate. Discount rate 9 Report the choice of discount rate(s) used for costs and outcomes and say why appropriate. Choice of health outcomes 10 Describe what outcomes were used as the measure(s) of benefit in the evaluation and their relevance for the type of analysis performed. Measurement of effectiveness 11a Single study-based estimates: Describe fully the design features of the single effectiveness study and why the single study was a sufficient source of clinical effectiveness data. 11b Synthesis-based estimates: Describe fully the methods used for identification of included studies and synthesis of clinical effectiveness data. Measurement and valuation of preference based outcomes 12 If applicable, describe the population and methods used to elicit preferences for outcomes.

Estimating resources and costs
13a Single study-based economic evaluation: Describe approaches used to estimate resource use associated with the alternative interventions. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs. 13b Model-based economic evaluation: Describe approaches and data sources used to estimate resource use associated with model health states. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs. Currency, price date, and conversion 14 Report the dates of the estimated resource quantities and unit costs. Describe methods for adjusting estimated unit costs to the year of reported costs if necessary. Describe methods for converting costs into a common currency base and the exchange rate. Describe all analytical methods supporting the evaluation. This could include methods for dealing with skewed, missing, or censored data; extrapolation methods; methods for pooling data; approaches to validate or make adjustments (such as half cycle corrections) to a model; and methods for handling population heterogeneity and uncertainty.

Study parameters 18
Report the values, ranges, references, and, if used, probability distributions for all parameters. Report reasons or sources for distributions used to represent uncertainty where appropriate. Providing a table to show the input values is strongly recommended. Incremental costs and outcomes 19 For each intervention, report mean values for the main categories of estimated costs and outcomes of interest, as well as mean differences between the comparator groups. If applicable, report incremental cost-effectiveness ratios. Characterising uncertainty 20a Single study-based economic evaluation: Describe the effects of sampling uncertainty for the estimated incremental cost and incremental effectiveness parameters, together with the impact of methodological assumptions (such as discount rate, study perspective). 20b Model-based economic evaluation: Describe the effects on the results of uncertainty for all input parameters, and uncertainty related to the structure of the model and assumptions. Characterising heterogeneity 21 If applicable, report differences in costs, outcomes, or costeffectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.

Discussion
Study findings, limitations, generalisability, and current knowledge 22 Summarise key study findings and describe how they support the conclusions reached. Discuss limitations and the generalisability of the findings and how the findings fit with current knowledge.

Other
Source of funding 23 Describe how the study was funded and the role of the funder in the identification, design, conduct, and reporting of the analysis. Describe other non-monetary sources of support.

Conflicts of interest 24
Describe any potential for conflict of interest of study contributors in accordance with journal policy. In the absence of a journal policy, we recommend authors comply with International Committee of Medical Journal Editors recommendations.