Comparative Evaluation of 10 Prehospital Triage Strategy Paradigms for Patients With Suspected Acute Ischemic Stroke.

Background The best strategy to identify patients with suspected acute ischemic stroke and unknown vessel status (large vessel occlusion) for direct transport to a comprehensive stroke center instead of a nearer primary stroke center is unknown. Methods and Results We used mathematical modeling to estimate the impact of 10 increasingly complex prehospital triage strategy paradigms on the reduction of population-wide stroke-related disability. The model was applied to suspected acute ischemic stroke patients in (1) abstract geographies, and (2) 3 real-world urban and rural geographies in Germany. Transport times were estimated based on stroke center location and road infrastructure; spatial distribution of emergency medical services calls was derived from census data with high spatial granularity. Parameter uncertainty was quantified in sensitivity analyses. The mothership strategy was associated with a statistically significant population-wide gain of 8 to 18 disability-adjusted life years in the 3 real-world geographies and in most simulated abstract geographies (net gain -4 to 66 disability-adjusted life years). Of the more complex paradigms, transportation of patients with clinically suspected large vessel occlusion based on a dichotomous large vessel occlusion detection scale to the nearest comprehensive stroke center yielded an additional clinical benefit of up to 12 disability-adjusted life years in some rural but not in urban geographies. Triage strategy paradigms based on probabilistic conditional modeling added an additional benefit of 0 to 4 disability-adjusted life years over less complex strategies if based on variable cutoff scores. Conclusions Variable stroke severity cutoff scores were associated with the highest reduction in stroke-related disability. The mothership strategy yielded better clinical outcome than the drip-'n'-ship strategy in most geographies.

Clinical Perspective

What Is New?

At least 10 different strategy paradigms exist for the prehospital triage of patients with suspected acute stroke.

In most geographic settings, the direct‐to‐comprehensive stroke center triage strategy paradigm (mothership approach) leads to a greater reduction in population‐wide stroke‐related disability and mortality (=gain in disability‐adjusted life years) than transport to the nearest primary stroke center (drip‐‘n'‐ship approach).

With regard to the remaining prehospital triage strategy paradigms, additional gains of disability‐adjusted life years can be achieved with more complex strategies; the greatest benefit is associated with the use of optimal variable (=location‐dependent) ordinal stroke symptom severity scale cutoff scores.

What Are the Clinical Implications?

The optimal prehospital triage strategy paradigm for a given region depends on region‐specific parameters, such as geographic location of primary and comprehensive stroke centers and treatment time performance metrics (door‐to‐needle, door‐to‐groin, door‐in‐door‐out).

The optimal prehospital triage strategy paradigm for a given region can be determined before implementation through the approach presented in this article.

Implementation of the regionally optimal prehospital triage strategy for acute stroke patients shortens prehospital delays to thrombectomy for patients with large vessel occlusions while also considering time‐to‐thrombolysis for patients without large vessel occlusions and is expected to improve outcomes.

Introduction

In patients with acute ischemic stroke (AIS) and cerebral proximal large vessel occlusion (LVO), treatment with mechanical thrombectomy (MT) leads to improved functional outcomes as compared with treatment with intravenous thrombolysis (IVT) alone.1, 2 In the prehospital setting, the presence of LVO cannot be determined reliably given currently available routine diagnostic tools.3, 4 The beneficial effects of both MT and IVT diminish over time5, 6 and not all acute hospitals that offer IVT can also perform MT. Furthermore, a secondary transfer of AIS patients with LVO from a non‐MT‐capable primary stroke center (PSC) to a MT‐capable comprehensive stroke center (CSC) is associated with significant time delays.7, 8 Therefore, the clinical problem of determining the best primary transport destination for patients with suspected AIS and unknown vessel status to achieve optimal outcomes has gained increasing attention over the past 2 years.9, 10, 11 Suggested strategies for prehospital triage have included algorithms based on the additional transport time to reach the nearest CSC12; the severity of stroke symptoms as surrogate marker for the probability of the patient harboring an LVO13, 14; and the estimated functional outcome derived from probabilistic conditional models associated with each available transport option.15, 16, 17, 18, 19, 20, 21 Which of these increasingly complex strategy paradigms should be implemented in a given geographic scenario to achieve optimal results and whether the higher potential costs of the more complex strategy paradigms are offset by a correspondingly larger reduction of stroke‐related disability is not clear.

In the current study, we use mathematical modeling to estimate and compare the impact of 10 increasingly complex prehospital triage paradigms on the reduction of stroke‐related disability and mortality on the population level in specific real‐world and abstract geographic scenarios.

Methods

Data and Research Materials Transparency

The authors declare that all supporting data are available within the article and the Online Data Supplement.

Model

We used an improved version of a previously published model18, 20 to estimate the impact of 10 different triage strategy paradigms on long‐term disability and mortality for patients with suspected acute stroke and unknown vessel status for whom there is uncertainty about the optimal transport destination as a function of age, sex, stroke severity, time to thrombolysis, and time to groin puncture (Figure 1). Changes in long‐term disability and mortality were expressed as disability‐adjusted life years (DALYs) or disability‐adjusted life days gained or lost. The model applies to unselected patients without prior disability for whom a Code Stroke is activated by emergency medical services (EMS) because of clinical suspicion of acute stroke within 8 hours of symptoms or unknown time of symptom onset. Parameters used in the model are presented in Tables S1 and S2 and Figures S1 through S4. The model assumes that patients with clinical symptoms suggestive of acute stroke that are managed by EMS personnel in the prehospital setting can either have a diagnosis of AIS with LVO, AIS without LVO, hemorrhagic stroke (HS), or stroke mimic. The probability of each diagnosis can be estimated as a function of the rapid arterial occlusion evaluation (RACE) scale, a clinical scale based on the items of the National Institutes of Health Stroke Scale (NIHSS) that quantifies the severity of stroke symptoms on a scale from 0 to 9, with higher scores indicating more severe stroke symptoms and a higher probability of IS with LVO and HS.22 The RACE scale was chosen because it has been validated prospectively in a prehospital cohort of patients with suspected stroke seen by EMS and quantifies the probability of 1 of the 4 final diagnoses (AIS with LVO, AIS without LVO, HS, and stroke mimic) on 10 levels. When used as a dichotomous score (cutoff ≥5), its accuracy is comparable to that of other LVO detection scales.11, 23 For patients with AIS caused by LVO, the model adopts a physiological perspective with a focus on the achievement of recanalization: This can occur either after the administration of IVT,24 or after MT with a procedure success rate of 80%.16 Based on these probabilities, the time delay to IVT at the nearest PSC, the time delay to IVT at the nearest CSC, the transfer time between PSC and CSC, the previously published age‐ and sex‐specific treatment effects of IVT and MT per minute faster treatment,25, 26 and the set of treatment time performance metrics at the stroke centers (door‐to‐needle, door‐out, door‐to‐groin, needle‐to‐groin, groin puncture‐to‐recanalization), we estimated the gain or loss of disability‐adjusted life days associated with direct transport to the nearest CSC as compared with the current standard of care of transporting all patients to the nearest IVT‐ready stroke center. For patients with HS and stroke mimic, the gain/loss of disability‐adjusted life days associated with different transport destinations was assumed to be zero because of lack of evidence of superiority of either transport strategy. Similarly, patients with contraindications to IVT (time from symptom onset >4.5 hours, oral anticoagulation, etc.) or need for advanced imaging (magnetic resonance imaging–based IVT in wake‐up strokes) were assumed to be transported to the nearest CSC and to not derive any benefit from prehospital triage (no equipoise because of lack of treatment options at the PSC). A detailed description of the model in mathematical terms is available in Data S1 with Table S3.

Figure 1

Model structure. CSC indicates comprehensive stroke center; DALD, disability‐adjusted life day; EMS, emergency medical services; IVT, intravenous thrombolysis; LVO large vessel occlusion; MT, mechanical thrombectomy; NIHSS, National Institutes of Health Stroke Scale; PSC, primary stroke center; RACE, rapid arterial occlusion evaluation scale.

Triage Strategy Paradigms

The set of triage strategy paradigms included in our comparison was based on clinical practice, review of the literature, and theoretical considerations. Figure 2 contains a description of the 10 identified paradigms. Paradigm I corresponds to the current standard of care of transporting all patients to the nearest PSC (drip‐‘n'‐ship). For strategy paradigm III that considered solely additional transport time, we used a time cutoff of 20 minutes12 in the base case analysis and performed additional univariate analyses to explore the impact of different time limits. Paradigms VII, VIII, and X are of more theoretical interest because of lack of the necessary technology at the moment. Paradigm IX is currently evaluated in the form of mobile stroke units. Paradigms IX and X involve administration of IVT to eligible patients on scene and are therefore not triage strategy paradigms in the narrow sense, but are included in our analysis for comparison.

Figure 2

Triage strategy paradigms. †Triage strategy paradigm in the wider sense, included for comparison. CSC indicates comprehensive stroke center; IVT, intravenous thrombolysis; LVO, large vessel occlusion; MT, mechanical thrombectomy; n.a. not applicable; PSC, primary stroke center; RACE, rapid arterial occlusion evaluation scale (score).

Geographic Scenarios

The effects of implementing 1 of the 10 triage paradigms were estimated in 3 specific real‐world and 5000 abstract geographic scenarios (Table S4). The specific real‐word scenarios included 2 urban geographic scenarios based on the city of Berlin, Germany (1 with centralized and 1 with decentralized MT services) and 1 rural geographic scenario based on the federal state of Schleswig‐Holstein in Northern Germany. The age‐ and sex‐specific population distributions according to 447 statistical geographic units in Berlin and 1112 communities in Schleswig‐Holstein were used to model the spatial distribution of stroke incidences and demographic characteristics of stroke patients at specific locations. For the second urban geographic scenario, we defined that MT was only offered at the 3 university hospitals in order to examine the effect of a stroke care infrastructure with centralized MT services (while in reality, MT is offered at up to 11 of all 14 stroke centers, depending on the time of day [de‐centralized MT services]). Transport times were calculated using freely available routing software (OSRM,27 Table S5).

To avoid making interpretations based on only 3 specific geographic scenarios, we analyzed random realizations of urban and rural geographic scenarios with varying numbers of PSCs and CSCs. For these abstract geographic scenarios, between 1 and 5 PSCs and CSCs (maximum total number of stroke centers: 10) were located randomly on a disc of radius 15 km (abstract urban) and 70 km (abstract rural). In the abstract scenarios, we assumed a spatially homogeneous population density and age distribution derived from those of the real‐world scenarios based on Berlin and Schleswig‐Holstein. Transport times between 2 points in the abstract scenarios were first calculated using a Euclidean metric and then transformed to road‐based transport times according to 2 nonlinear relationships estimated from the specific real‐world geographic scenarios (Data S1, Figure S5).

Analysis

For analysis, a total of 10 000 positions were sampled from each geographic scenario (Data S1). The equipoise region was defined as the set of points for which the onset‐to‐thrombolysis time at the nearest PSC (including transport time and door‐to‐needle time) was smaller than the onset‐to‐thrombolysis time at the nearest CSC. For each sampled location in the equipoise region, and each of the 10 triage strategy paradigms, the preferred transport destination (nearest PSC or nearest CSC) was calculated for each of the [age×sex×RACE score] input combinations as shown in Figure 2. These triage strategy paradigm‐specific transport destination decision rules were applied to simulated EMS calls in the examined regions and the results were weighed according to the population‐specific incidence of EMS calls at the given location, stroke symptom severity distribution, and RACE‐score‐dependent probability of the diagnosis of IS with LVO, IS without LVO, HS, and stroke mimic. The incidence of EMS calls was assumed to be proportional to the estimated age‐ and sex‐specific stroke incidence. EMS calls for patients aged 35 years or older with suspected acute stroke were considered in the analysis, because no reliable data for the amount of disability‐adjusted life days saved per minute faster treatment were available for younger patients, and most stroke patients (in our model: 99.6%) are at least 35 years old. Besides the primary outcome of the population‐wide annual gain/loss of DALYs associated with each triage strategy paradigm, we also extracted information on the average time to IVT and MT; the proportion of AIS patients being triaged to the “correct” destination (ie, patients without LVO to the nearest PSC, patients with LVO to the destination associated with better clinical outcome); the total volume of patients being triaged primarily to a PSC and CSC; and the total number of secondary transfers. Uncertainty was quantified in probabilistic sensitivity analyses. For this, each analysis in the 3 specific real‐world geographic scenarios was repeated 1000 times with parameters drawn randomly and independently from their respective distributions and results presented as intervals that contained 95% of the obtained values (equal‐tailed credible intervals [CI]). The effect of changes in the door‐out time at PSCs was examined in separate univariate sensitivity analyses.

All simulations and analyses were performed in MATLAB28 except for the calculation of transport times in real‐world scenarios, which was done in R.29 No ethical approval or informed patient consent were required for this study.

Results

Transport Destination Decision Rules

First, we calculated transport destination decision rules for each of the 10 prehospital triage strategy paradigms according to geographic location, age, sex, and stroke symptom severity (RACE score). For an exemplary 70‐year‐old male patient, Figure 3 shows the RACE cutoff scores at or above which a patient should be transported to the nearest CSC for stroke triage paradigms based solely on additional transport time (III) and optimal variable cutoff scores (VI). For the remaining currently available paradigms, drip‐‘n'‐ship (I), mothership (II), and fixed cutoff score (IV), the transport destination rules are independent of the estimated transport times to the nearest stroke enters. For paradigm V (fixed cutoff score with probabilistic outcome determination), the RACE cutoff score was equal to 5 at all positions of the examined specific scenarios. The relative size of the equipoise region in the specific real‐world geographic scenarios (ie, the region where a triage decision is necessary because the time‐to‐IVT at the nearest PSC is shorter than the time‐to‐IVT at the nearest CSC, calculated with regard to the estimated annual number of EMS calls for suspected acute stroke) was 30% in the urban scenario based on the city of Berlin with decentralized MT services, 81% in the urban scenario based on the city of Berlin with centralized MT services, and 61% in the rural scenario based on the state of Schleswig‐Holstein. Transport destination decision rule maps for abstract geographic scenarios are presented in Figure S6.

Figure 3

Prehospital triage strategy paradigm‐associated transport destination decision rules in specific real‐world urban and rural geographic scenarios. Color‐coded are the RACE cutoff scores at or above which a 70‐year‐old male patient seen by emergency medical services personnel for suspected acute stroke should be transported to the nearest CSC instead of the nearest PSC for a triage strategy based solely on a maximum additional transport time of 20 minutes (paradigm III, left) and optimal variable stroke symptom severity cutoff scores (paradigm VI, right). A dash “‐” signifies transport of all patients to the nearest CSC because of lack of equipoise because of a shorter transport time (light color) or to the nearest PSC (dark color). CSC indicates comprehensive stroke center; MT, mechanical thrombectomy; PSC, primary stroke center; RACE, rapid arterial occlusion evaluation scale.

Specific Real‐World Geographic Scenarios

We next applied the 10 transport destination decision rule maps to simulated EMS calls for suspected acute stroke in specific urban and rural real‐world geographic scenarios (Berlin I [“as is,” decentralized MT services], Berlin II [“theoretical,” centralized MT services], and Schleswig‐Holstein, Figure 4). In all 3 examined specific scenarios, transporting all patients to the nearest PSC irrespective of transport times and stroke symptom severity (drip‐‘n'‐ship, paradigm I) was associated with a slightly shorter onset‐to‐thrombolysis time for patients with AIS and a significantly longer onset‐to‐groin puncture time for AIS patients with LVO compared with all other currently available paradigms (II to VI). In comparison to the drip‐‘n'‐ship approach (paradigm I), the remaining currently available triage strategy paradigms II to VI were associated with an estimated population‐wide annual gain of DALYs of between 8 and 18 DALYs.

Figure 4

Impact of prehospital triage strategy paradigms on patient‐centered outcome parameters in specific real‐world geographic scenarios. Boxplots show data for prehospital triage strategy paradigms I to X from probabilistic sensitivity analyses; vertical extent of the boxes represents the interquartile range, the horizontal line represents the base case result, and the whiskers extend to include 95% of all values. Currently available triage strategy paradigms (I–VI) are shown in blue, the remaining paradigms (VII–X) in shades of red. Gain of DALYs is calculated with reference to triage strategy paradigm I (drip‐‘n’‐ship approach). The last row depicts the additional gain in DALYs associated with each triage strategy paradigm over and above all less complex triage strategy paradigms. For a description of triage strategy paradigms, see Figure 2. DALY indicates disability‐adjusted life year; IVT, intravenous thrombolysis; MT, mechanical thrombectomy.

When considering whether a triage strategy paradigm should be implemented, the complexity of each paradigm needs to be taken into account. We therefore also analyzed the additional gain of DALYs associated with each triage strategy paradigm over and above the best performing less complex paradigm. Here we found that the mothership approach (paradigm II) was associated with a statistically significant gain of DALYs over the drip‐‘n'‐ship approach (strategy I) in all examined specific geographic scenarios. In the specific real‐world urban geographic scenario with decentralized MT services (Berlin I, “as is”), none of the remaining, currently available more complex triage strategy paradigms (III–VI) provided an additional clinical benefit. In the specific real‐world urban geographic scenario with centralized MT services (Berlin II, “theoretical”) and in the specific real‐world rural scenario, a triage strategy based on optimal variable stroke severity cutoff scores (VI) offered a statistically significant additional benefit of 1.6 DALY per year (95% CI: 0.0–2.4 DALYs) and 1.1 DALY per year (95% CI: 0.2–2.0 DALYs), respectively. In addition, in the specific real‐world rural scenario, a triage strategy based on a fixed stroke severity cutoff without consideration of transport times (IV) was associated with a not statistically significant additional benefit in comparison to less complex strategies (2.12 DALYs [−0.5 to 3.2 DALYs]).

In addition, we investigated how prehospital triage using any of the 10 paradigms would affect the proportion of patients triaged primarily to a CSC and a PSC, as well as the number of secondary transfers. Results are displayed in Figure 5. Of note, the proportions of AIS patients transported to the “correct” destination were significantly higher with strategy paradigms based on clinical stroke severity scales (IV, V; 75%% [95% CI: 73–77%]) than with the mothership strategy (II; 29% [95% CI: 27–31%]).

Figure 5

Impact of prehospital triage strategy paradigms on health system–related outcome parameters in specific real‐world geographic scenarios. Boxplots show data for prehospital triage strategy paradigms I to X from probabilistic sensitivity analyses; vertical extent of the boxes represents the interquartile range, the horizontal line represents the base case result, and the whiskers extend to include 95% of all values. Currently available triage strategy paradigms (I–VI) are shown in shades of blue, the remaining paradigms (VII–X) in shades of red. For a description of triage strategy paradigms, see Figure 2. CSC indicates comprehensive stroke center; MT, mechanical thrombectomy; PSC, primary stroke center.

For comparison, we also analyzed the potential gain of DALYs associated with triage strategy paradigms based on technology that is currently not available (optimal LVO detection device [paradigms VII and VIII], mobile MT unit [X]), or has limited availability (mobile IVT unit [IX]). As shown in Figures 4 and 5 (red bars), such novel technologies have the potential to achieve clinically significant reductions of stroke‐related disability in addition to what can be attained with currently available triage strategy paradigms. Numerical results of all examined outcome measures are presented in Tables S6 through S8.

Abstract Geographic Scenarios

The impact of prehospital triage of patients with suspected AIS depends not only on the absolute number of PSCs and CSCs (ie, the overall spatial density of stroke centers) but also on the CSC‐to‐PSC ratio and the relative location of the stroke centers to each other, all of which directly influence the size of the equipoise region. We therefore chose to examine the effect of prehospital triage strategy paradigms in abstract urban and rural scenarios according to the size of the equipoise region (Figure 6). Similar to the results obtained in the specific real‐world geographic scenarios, we found that the mothership approach (paradigm II) offered significant additional clinical benefit over the drip‐‘n'‐ship approach (I) in nearly all random abstract scenarios, with the difference in disability ranging from zero of 66 DALYs per year in the urban and from −4 to 31 DALYs per year in the rural scenarios. A strategy based on optimal variable cutoff scores (VI) was associated with a small additional benefit of up to ≈0 to 4 DALYs per year over all other paradigms. In rural, but not in urban scenarios, fixed cutoff scores (paradigm IV) offered some additional benefit (up to 12 DAYLs). Strategies only considering additional transport time (III) and strategies based on a fixed cutoff score with probabilistic outcome determination (V) did not offer additional benefit over less complex paradigms, except for a few rural scenarios with a large equipoise region. In summary, results in abstract scenarios confirmed the findings from the three specific real‐world geographic scenarios.

Figure 6

Impact of prehospital triage strategy paradigms on the reduction of stroke‐related disability in abstract geographic scenarios. Boxplots show results for prehospital triage strategy paradigms I to X from repeated random generation of abstract rural and urban geographic scenarios with between 1 and 5 primary stroke centers and 1 and 5 comprehensive stroke centers according to the relative size of the equipoise region (ER). Vertical extent of the boxes represents the interquartile range, the horizontal line the mean, and the whiskers extend to include 95% of all values. Currently available triage strategy paradigms (I–VI) are shown in shades of blue, the remaining paradigms (VII–X) in shades of red. In the first row, gain of DALYs is calculated with reference to triage strategy paradigm I (drip‐‘n’‐ship approach). The second row depicts the additional gain in DALYs associated with each triage strategy paradigm over and above all less complex triage strategy paradigms. For a description of triage strategy paradigms, see Figure 2. DALY indicates disability‐adjusted life year.

Univariate Sensitivity Analysis

In univariate sensitivity analyses assuming a shorter door‐out time of 15 minutes, we found an overall diminished magnitude of the effect of the mothership strategy (II) on the gain of DALYs. In addition, the mothership approach was no longer superior to the drip‐‘n'‐ship approach (I) in a relevant proportion of abstract rural geographic scenarios with a large equipoise region. Instead, in abstract rural geographic scenarios, triage strategy paradigms based solely on transport times (III) and a fixed cutoff score with probabilistic outcome determination (V) were associated with a modest additional clinical benefit as compared with less complex strategy paradigms (additional DALYs gained up to 6 and 4, respectively). The additional benefit of optimal variable cutoff scores (triage paradigm VI) was similar to the base case results in urban, and decreased by ≈50% in rural scenarios (Figures S7 through S9, Tables S6 through S8).

In a second univariate sensitivity analysis, we assessed the impact of different time limits for strategies considering only additional transport time (paradigm III). Regarding the impact on the reduction of stroke‐related disability and mortality, we found that in urban scenarios, most DALYs were gained with a time limit of 20 minutes; higher time limits provided similar benefit (identical to mothership approach for time limit ≥30 minutes). In rural scenarios, the optimal transport time limit was 30 to 40 minutes; when higher time limits were used, the benefit of triage started to decrease. In absolute terms, the differences between triage strategies using different time limits were modest (up to 5 DALYs in all 3 real‐world geographic scenarios).

Discussion

Main Findings

We estimated the effect of 10 increasingly complex prehospital triage strategy paradigms for patients with suspected acute stroke in a probabilistic conditional model. In our model, for patients with suspected acute stroke and unknown vessel status for whom a Code Stroke is activated by EMS, direct transportation to the nearest CSC (mothership approach) instead of the nearest PSC (drip‐‘n'‐ship) was associated with a net gain of DALYs. The total amount of DALYs gained ranged from 8 to 18 in the specific real‐world geographic scenarios and from −4 to 66 in the abstract geographic scenarios. Adjusting prehospital triage algorithms to include stroke symptom severity irrespective of expected transport times was associated with an additional gain of DALYs in rural scenarios, particularly in those with a relatively large equipoise region (ie, fewer stroke centers with centralized MT services). Of the triage strategy paradigms based on probabilistic conditional modeling, use of optimal variable cutoff scores, (ie, consideration of vessel status on an ordinal scale) yielded an additional gain of DALYs over and above less complex triage strategy paradigms in all scenarios. On the other hand, triage strategy paradigms based on a fixed cutoff score (ie, consideration of vessel status on a dichotomous scale) were associated with additional benefit only in rural scenarios under the assumption of a short door‐out time of 15 minutes.

Previous Studies

Our study is the first to systematically collect a list of conceivable prehospital triage strategy paradigms for patients with suspected AIS and to compare the consequences of their implementation in specific real‐world geographic scenarios in a single model. Hereby, our aim was not to compare the accuracy of different individual prehospital stroke symptom severity scales, but to evaluate a set of conceptually different triage strategy paradigms. Our study is also the first to estimate population‐wide effects, which are ultimately the driving force for decisions for or against the implementation of a given triage strategy. For this, we aggregated the estimated outcomes of individual patients while taking into account the spatial distribution of stroke incidence, spatially heterogeneous demographics, and the distribution of stroke severity. Previously published reports using mathematical modeling for the evaluation of prehospital triage decision algorithms have been performed in simplified abstract geographic scenarios with only 2 stroke centers without considering spatially heterogeneously distributed population characteristics and true transport times, or have only analyzed the impact of 1 single triage strategy (see Holodinsky et al21 for a recent review). Apart from mathematical modeling, robust evidence from real‐world studies is still scarce for most of the examined triage strategy paradigms. In line with results of our study, there is some evidence from clinical studies that the mothership strategy (ie, direct transportation of all patients with suspected stroke to the nearest CSC irrespective of stroke symptom severity and transport times) might be beneficial if the additional transport time is below 30 to 45 minutes.9 Regarding the benefit of prehospital triage strategy paradigm IV (ie, transportation of all patients with suspected LVO as determined by a higher score on a clinical prehospital stroke symptom severity scale to a CSC irrespective of transport times), a randomized controlled trial is currently ongoing (NCT02795962)13 with results expected for 2020.

Clinical Implications

When planning the implementation of a prehospital triage strategy for patients with suspected AIS in order to reduce the time delay to the most adequate and effective reperfusion treatment, decision makers need to consider the impact of the intervention on clinical outcome, but also the cost of the intervention. In contrast to previous studies,15, 16, 18 we chose to quantify the clinical impact of prehospital triage not as the probability of good functional outcome at 90 days but as the long‐term reduction of disability and mortality, a more generic measure that permits a direct comparison with the effectiveness of other healthcare interventions.30 When selecting one of the many available triage strategies, the additional benefit over less complex strategies needs to be weighed against the increasing cost of setting up and maintaining the triage strategy. For example, implementation of a strategy based on optimal variable cutoff scores would require dedicating resources to train EMS personnel to reliably use an ordinal stroke symptom severity scale such as the RACE scale and to maintain an updated online service to allow the real‐time prediction of the expected outcome based on the stroke severity scale score and the expected transport times for each individual patient. On the other hand, the unconditional mothership approach would be easier to implement and maintain, but at the same time would be associated with greater shifts of patient volumes between hospitals and slightly smaller reductions of stroke‐related disability and mortality. A formal cost‐effectiveness analysis addressing these questions, which is beyond the scope of this article, is currently planned.

As shown previously for selected prehospital triage strategy paradigms, the impact of prehospital triage is strongly influenced by performance time metrics (door‐to‐needle time, door‐out‐time) at PSCs.15, 18 In our study, assuming a shorter door‐out‐time was associated with a lower proportion of abstract rural geographical regions in which patients would benefit from unconditional transportation to the nearest CSC (mothership). The decision for or against implementation of a specific prehospital triage strategy paradigm should therefore be preceded by an estimation of the expected impact considering regional performance time metrics at participating PSCs. In particular, our results indicate that better performance time metrics at PSCs should translate directly into higher patient volumes at PSCs, and vice versa.

In addition to its potential in improving patient‐related outcomes, prehospital triage of patients with suspected acute stroke also affects health system–related parameters. Our current study confirms the findings of previous studies of increased numbers of patients managed in emergency departments of CSCs, lower patient volume in PSCs, and a lower number of secondary transfers.19, 31 These shifts would require providers to adapt their services over time to cope with the higher or lower volume. Although in our opinion, the most relevant parameter for decision making is the improvement of functional outcome and the reduction of disability of stroke patients, secondary and higher‐order ramifications of shifts in patient volume should not be ignored. At the level of PSCs, such consequences may include efforts to offer MT in order to attract more patients with uncertain consequences for the quality of MT services offered, but also the establishment of policies and protocols to ensure rapid IVT‐to‐door‐out times. Similarly, higher patient volumes at CSCs could lead to further streamlined processes with shorter pretreatment delays or, when no adequate resources can be made available, to a decrease in quality because of overcrowding. In comparison to the mothership strategy, use of clinical stroke symptom severity scales to inform triage decisions would be associated with a similar (sometimes even larger) reduction of stroke‐related disability while at the same time causing a smaller shift of patients away from PSCs to CSCs.

Strengths and Limitations

Theoretical models offer the opportunity to answer questions that are difficult to examine in clinical trials, such as a direct comparison of 10 different triage strategy paradigms, and to vary input parameters over a wide range, in our case to examine several geographic scenarios and stroke care infrastructure settings simultaneously. However, we are aware that models to analyze complex decision problems represent simplified abstractions from reality whose results are influenced by the assumptions made when building the model. In the current study, we addressed some of the weaknesses of previous studies. In particular, we constructed our model to represent the real clinical scenario of patients with stroke symptoms, but unknown final diagnosis. In addition, we applied our model to specific real‐world geographic scenarios for which demographic data were available at a high spatial granularity and derived the key parameters for the model from a large prospective prehospital cohort of patients with suspected acute stroke managed by EMS that is representative of the target population of our model.22 Data on the time‐dependent effectiveness of IVT and MT stratified by age, sex, and stroke symptom severity were estimated using the pooled effects of large randomized controlled trials.25, 26 In addition, we quantified the uncertainty of our results in probabilistic sensitivity analyses and present 95% CI for all outcome measures. On the other hand, we had to make certain assumptions in our model because of lack of availability of data that likely represent an oversimplification compared with reality. First, we were unable to model the correlation between demographic factors, especially age, stroke symptom severity, and the eligibility for IVT; and the correlation between the probability and timing of early recanalization of LVO after IVT, location of vessel occlusion, and stroke symptom severity. Second, data on the uncertainty of input parameters for probabilistic sensitivity analyses were not available from the literature for all parameters. Third, the spatial distribution of EMS calls was modeled using census data, which assume that strokes occur mostly close to home. Fourth, the possibility of MT in an extended time‐window up to 24 hours for selected patients32, 33 was not considered because for most of such patients, there would not be equipoise between transport to the nearest PSC or nearest CSC in the first place because of ineligibility for IVT (maximum time from symptom onset 4.5 hours). For the small number of patients who could be treated at a PSC within 4.5 hours but who could not be transferred to arrive at a CSC within 6 hours, advanced imaging protocols could help to select patients who are likely to derive benefit from MT beyond 6 hours and for whom transfer should be considered. Since the time‐decay of the treatment efficacy of MT in imaging‐selected patients is not yet well characterized and the impact on the overall results of our study is expected to be small because of the small number of patients, we did not include this scenario in our model. Concerning the eligibility for IVT, we assumed that lack of eligibility (eg, wake‐up stroke, anticoagulation, and recent major surgery) could be ascertained in the prehospital setting and excluded these patients from further analyses. The alternative assumption that eligibility for IVT can only be determined after transport to a stroke center would lead to more patients being affected by prehospital triage and larger effect sizes. Last, our model was based on a dichotomy of stroke center characteristics in terms of capability to perform MT; differences in procedural quality affecting outcome (eg, as a function of patient volume, or nonbinary quantification of the availability of MT) could not be considered because of lack of data.

Conclusions

In summary, we have applied a mathematical model based on conditional probabilities to highly granular real‐world geographic and demographic data to compare the impact of 10 prehospital triage strategy paradigms for patients with suspected AIS and unknown vessel status. In general, unconditional transport to the nearest CSC (mothership approach) yielded better outcome than did transport to the nearest PSC (drip‐‘n'‐ship) in urban and most rural scenarios. However, our results suggest that a stroke symptom severity‐based triage using variable cutoff scores that depend on estimated transport times is associated with the highest reduction in stroke‐related disability and mortality. Improvement of key performance measures at the PSC level has an important impact on the effect of the optimal triage strategy. Technologies that allow treating patients with IVT or MT on scene would be associated with significant additional reductions in stroke‐related disability. Last, prehospital triage strategies can have a significant impact on the distribution of patient volume between CSCs and PSCs that needs to be considered before making a decision to implement one of the available triage strategy paradigms.

Author Contributions

L. Schlemm and E. Schlemm conceived the study. L. Schlemm reviewed the literature, developed the model, performed the simulation, analyzed the data, created the figures, and wrote the first draft of the manuscript. E. Schlemm reviewed the model, performed geostatistical analyses for the application of the model to real‐world geographic data, and visualized spatial maps. Ernst contributed data about the stroke centers in Schleswig‐Holstein. All authors contributed to interpreting the data, revised the manuscript for intellectual content, and approved the final version of the manuscript.

Sources of Funding

L. Schlemm is a participant in the Berlin Institute of Health—Charité Clinical Scientist Program funded by the Charité—Universitätsmedizin Berlin and the Berlin Institute of Health.

Disclosures

Endres reports grants from Bayer and fees paid to the Charité from Bayer, Boehringer Ingelheim, BMS/Pfizer, Daiichi Sankyo, Amgen, GSK, Sanofi, Covidien, and Novartis, all outside the submitted work. Scheitz reports fees for lectures from Stryker GmbH & Co. KG and Bristol‐Myers Squibb, and grant support by the Corona Stiftung, all outside the submitted work. Nolte reports consulting and lecture fess from Boehringer Ingelheim, W.L. Gore and Associates, Bristol‐Myers Squibb, Pfizer, and Sanofi, all outside the submitted work. The remaining authors have no disclosures to report.

Data S1. Supplemental Methods and Results.

Table S1. Model Parameters—1

Table S2. Parameters of the Model—2

Table S3. Definitions

Table S4. Population and Geographic Parameters of the 5 Geographic Scenarios

Table S5. Custom Transport Profile Used With OSRM

Table S6. Patient‐Related Outcome Measures in Specific Real‐World Geographic Scenarios, Including Univariate and Probabilistic Sensitivity Analyses

Table S7. Health System–Related Outcome Measures in Specific Real‐World Geographic Scenarios, Including Univariate and Probabilistic Sensitivity Analyses

Table S8. Figure S1. Probability density functions of probabilities for final diagnoses according to RACE score.

Figure S2. Probability density functions of the relative frequencies of each RACE score category encountered by emergency medical services (EMS) personnel in the prehospital setting.

Figure S3. Probability density functions of National Institutes of Health Stroke Scale (NIHSS) scores according to RACE score.

Figure S4. Reduction of DALDs per minute faster treatment for acute ischemic stroke patients.

Figure S5. Fit of transport times vs Euclidean distances in specific real‐world geographic scenarios.

Figure S6. Prehospital stroke triage strategy paradigm‐associated transport destination decision rule maps in abstract urban and rural geographic scenarios.

Figure S7. Impact of prehospital triage strategy paradigms on patient‐centered outcome parameters in specific real‐world geographic scenarios.

Figure S8. Impact of prehospital triage strategy paradigms on health system–related outcome parameters in specific real‐world geographic scenarios

Figure S9. Impact of prehospital triage strategy paradigms on the reduction of stroke‐related disability in abstract geographic scenarios

Click here for additional data file.