Prehospital-Stroke-Scale Parameterized Hospital Selection Protocol for Suspected Stroke Patients Considering Door-to-Treatment Durations.

Background To mitigate uncertainty that may arise in the judgment of emergency medical technicians when relying on a prehospital stroke scale at the scene, we propose a hospital selection protocol that considers the uncertainty of a prehospital stroke scale and the actual door-to-treatment durations, and we have developed a web-based system to be used with mobile devices. Methods and Results This hospital selection protocol incorporates real-time, estimated transport time obtained from Google Maps, historical median door-to-treatment duration at hospitals that only provide the standard intravenous thrombolysis treatment, and at hospitals with endovascular thrombectomy for probable large-vessel occlusion cases. We have validated the efficiency of the proposed protocol and compared it with other strategies used by emergency medical technicians when deciding on a receiving hospital. Using the proposed protocol for the triage reduces the time from onset to receiving definitive treatment by nearly 11 minutes. We found that the nearest endovascular thrombectomy-capable hospital from the scene may not be the most ideal if the door-to-treatment durations are discriminative. The results show that, when the tolerable bypass transport threshold and administration time are reduced to 9 minutes and 30.5 minutes, respectively, 228 patients out of 7678 cases, whose receiving hospitals were changed to endovascular thrombectomy-capable hospitals, received definitive treatment in a shorter time. The results of our analysis give recommendations for appropriate allowable bypass transport time for regional planning. Conclusions By applying almost-real value parameters, we have validated a web-based model, which can be universally adapted for optimal, time-saving hospital selection for patients with stroke.

acute ischemic stroke

Cincinnati Prehospital Stroke Scale

endovascular thrombectomy

intravenous thrombolysis

large vessel occlusion

recombinant tissue plasminogen activator

Clinical Perspective

What Is New?

Incorporating the Mathematical Programming and Geographic Information System, we propose a protocol to decide which hospital a patient suspected of experiencing stroke should be sent to.

The protocol, a web‐based system, accessible via the mobile devices of prehospital personnel, has been developed. The prehospital personnel will be able to use this system at the scene and make timely decisions.

What Are the Clinical Implications?

Aided by the web‐based system, the prehospital personnel can make more appropriate decisions for patients.

A reasonable bypass strategy can allow patients to receive treatment faster for better prognosis with this system.

Patients experiencing acute ischemic stroke (AIS) have better outcomes if the time is reduced between onset and receiving definitive treatment, such as intravenous thrombolysis (IVT) or endovascular thrombectomy (EVT), to reperfuse the brain tissues. , , , , , , There are already strategies designed to ensure that patients with AIS receive definitive treatment as quickly as possible. However, it is sometimes difficult for emergency medical technicians (EMTs) to determine the best approach when evaluating a patient because procedural uncertainties (such as transport time, door‐to‐treatment duration, and testing, etc.) have to be considered. EMTs commonly reference prehospital stroke scales to identify patients with large vessel occlusion (LVO) and determine the receiving hospital accordingly. However, these scales are not 100% accurate in identifying LVO. Some patients with LVO may be sent to a hospital that only provides IVT. They then have to be transferred to an EVT‐capable hospital, after tests, which delay their receiving treatment.

Every time interval that a patient has to undergo before receiving definitive treatment must be carefully calculated. Real‐time transport time is often discussed in the literature. For time intervals, after a patient has arrived at the first receiving hospital, Schlemm et al considered the door‐to‐treatment duration based on the American Heart Association , , , , , guidelines, while other researchers used the data of door‐to‐treatment duration in clinical trials, or based on systems of care recommendations. Actual door‐to‐treatment duration in hospitals is rarely discussed in the literature.

The aim of developing hospital selection protocol is to provide advice and to help EMTs make a reasoned decision. Before the introduction and implementation of the prehospital‐stroke‐scale parameterized hospital selection protocol, EMTs would send a patient to the closest hospital in time or distance from the scene. However, existing models do not factor in the differences in the procedures needed by patients with stroke because of the uncertainties that arise when assessing the severity of the stroke using the prehospital stroke scales.

We propose a hospital selection protocol with a probability measure to identify patients with LVO according to the number of the prehospital stroke scale indicators presented, which other mathematical models have not considered. Furthermore, the method is guaranteed to minimize the expected time for a patient to receive definitive treatment. The protocol, a web‐based system, accessible via the EMTs’ mobile devices, has been developed for Taipei City. EMTs will be able to use this system at the scene and make timely decisions.

Methods

The data that support the findings of this study are available from the corresponding author upon reasonable request by email.

Study Setting

To carry out the study, we considered a capital city where the average stroke incidence rate is 330 per 100 000 people, of which 74% are ischemic stroke cases. The city has a metropolitan area of 272 km2. It has a population of 2.65 million with an inflow working population of 3 million. The 2‐tier fire‐based emergency medical service (EMS) system contains 41 basic life support units and 4 advanced life support units. In the city, the EMS helps to transport ≈30% of patients with stroke to a hospital. There are currently 1206 EMTs in the city, who at the scene, use the Cincinnati Prehospital Stroke Scale (CPSS) to identify patients with acute stroke. The symptoms of CPSS included the following presentations: facial palsy, arm weakness, and speech abnormalities. In addition, they do the pinprick test to check blood glucose levels. There are ten 24/7 hospitals in the city that provide recombinant tissue plasminogen activator (rt‐PA) 24/7, of which 6 are also EVT‐capable.

One‐Stage Stochastic Optimization Model

In the proposed hospital selection protocol, we use a 1‐stage stochastic optimization model, where the decision variable at the scene is the receiving hospital for a patient, while the random variable is the time taken for the patient to receive definitive treatment. The probability measure to identify patients with LVO, according to the number of CPSS symptoms presented, is used to calculate the minimized expected time for the patient to receive definitive treatment because we cannot know exactly the patient’s stroke level before obtaining the results of computed tomography angiography of the brain.

We can obtain 2 meaningful quantities from the model output: the expected time in which a patient will receive definitive treatment, and whether a patient should be sent to a hospital only providing IVT followed by possible transfer, or sent directly to an EVT‐capable hospital to receive definitive treatment. The sequential process before a patient receives definitive treatment, and the 2 treatment or transfer scenarios are all taken into account (Figure 1).

Figure 1

Processes for a patient experiencing acute ischemic stroke to receive definitive treatment.

EVT‐capable hospital, providing both intravenous thrombolysis and endovascular thrombectomy; rt‐PA hospital, providing only intravenous thrombolysis. EMTs indicates emergency medical technicians; EVT, endovascular thrombectomy; and rt‐PA, recombinant tissue plasminogen activator.

Let be the probability that patient is experiencing an AIS without LVO according to the number of CPSS symptoms they have when tested by the EMTs on the scene, and let be the probability that patient has LVO. The probabilities of a patient having LVO, and conditional on 3, 2, or 1 of the 3 CPSS symptoms, are 0.310, 0.265, and 0.239, respectively. There is also an alternative probability measure related to the number of CPSS symptoms, according to Richards et al. Results related to those of Richards et al are shown in Data S1.

Therefore, when a patient is sent to an EVT‐capable hospital, the expected time to receive the definitive treatment is with the following definitions:

The expected time to receive definitive treatment when a patient is initially sent to a hospital only providing IVT (rt‐PA hospital) but who may have to be transferred to an EVT‐capable hospital is with the following further definitions:

Transfer time includes the driving time from the rt‐PA hospital to the nearest EVT‐capable hospital and the door‐to‐treatment duration in the EVT‐capable hospital. From the data given, the transport driving time from the patient’s address was calculated as off‐peak according to Google Maps. Administration time was defined as the time interval from the first image of computed tomography angiography of brain shown on the computer screen to an rt‐PA hospital departure. The administration time based on Ng et al was initially set at 46.5 minutes. The door‐to‐test duration was defined as the time interval from rt‐PA hospital arrival to the first image of computed tomography angiography of the brain shown on the computer screen. The door‐to‐test duration and test‐to‐treatment duration were set by the medians of the historical data from each hospital in Taipei City, which varied among hospitals.

When the EMT inputs into the web‐based system the patient’s location and number of CPSS symptoms, whether they have LVO or not, the EMT will only get 1 suggested receiving hospital, which is considered to be the most appropriate. In addition, when the time difference between the scene to any rt‐PA hospital and the scene to the nearest EVT‐capable hospital is less than seconds, the model always sends the patient directly to the EVT‐capable hospital.

was initially set to 15 minutes because the American Heart Association guidelines suggest that good outcomes deteriorate with every 15‐minute delay. The mathematical model is described in Data S1.

The parameters inputted into the model are almost actual data. The transport time is calculated according to off‐peak driving time in Google Maps, and the processing time in each hospital is based on the 4‐year median data from 2016 to 2019. To test the model’s accuracy, we used the 6‐year historical data of 7678 patients who had a suspected stroke and who exhibited at least 1 of the 3 CPSS symptoms between January 1, 2010, and December 31, 2015. The model was implemented with A Mathematical Programming Language, which is an intuitive algebraic modeling system, and IBM ILOG CPLEX Optimization Studio, which was used to solve the underlying mathematical programming model. This study and stroke registry were approved by the Institutional Review Board of National Taiwan University Hospital.

Primary Approach of Critical Parameters

We tested the performance of the proposed protocol with a 6‐year data set of 7678 histories of patients who have had a suspected stroke in Taipei City. Among the 7678 patients, 4037 had 3 CPSS symptoms, 1319 had 2 symptoms, and 2322 had 1 symptom.

Using the probability measure given by Scheitz et al, we conducted a primary approach of the administration time needed for hospital transfer, and the tolerable bypass transport threshold below which a patient bypasses the nearer hospital providing IVT to go straight to an EVT‐capable hospital.

We then decreased the initial parameters of the tolerable bypass transport threshold of 15 minutes and administration time of 46.5 minutes by 1 minute at a time. From the results of the analysis, we selected 3 critical combinations of values. The first is when patients are sent to rt‐PA hospitals to begin with (as opposed to all patients being sent directly to EVT‐capable hospitals). The second and third critical combinations, when decreasing the 2 parameters, show significant changes in the total expected time for a patient to receive definitive treatment. Potentially, setting 1 of these 2 critical combinations of values of U and A as a new practical standard could be more appropriate to Taipei City than the initial (current) values.

Comparisons With Other Strategies for Deciding on a Receiving Hospital

We compared the time to receive definitive treatment when using the proposed hospital selection model with the other 4 strategies. We also generated plots to validate the efficiency of the proposed protocol under different situational parameters. We can thus suggest future applications of the proposed strategy. The following are the 5 strategies we compared for sending patients with AIS to a hospital:

A patient with a suspected stroke with at least 1 CPSS symptom is sent to the nearest hospital, whether it is EVT‐capable or rt‐PA‐capable. If a patient with LVO is sent to an rt‐PA‐capable hospital, the patient should be transferred to the nearest EVT‐capable hospital.

A patient with a suspected stroke with at least 1 CPSS symptom is sent directly to the nearest EVT‐capable hospital.

A patient with a suspected stroke with at least 1 CPSS symptom is sent to a hospital according to the result of the proposed hospital selection model (proposed strategy).

A patient with a suspected stroke is sent to a hospital based on the number of their CPSS symptoms. If a patient has 3 CPSS symptoms, they are sent directly to the nearest EVT‐capable hospital. A patient with 1 or 2 CPSS symptoms is sent to the nearest hospital, whether it is EVT‐capable or rt‐PA‐capable.

A patient with a suspected stroke is sent to a hospital based on the number of their CPSS symptoms. If a patient has 2 or 3 CPSS symptoms, they are sent directly to an EVT‐capable hospital. If a patient has 1 CPSS symptom they are sent to the nearest hospital, whether EVT‐capable or rt‐PA‐capable.

The information in our historical data only gives each patient’s number of CPSS symptoms. It does not include whether or not a patient had confirmed LVO. To evaluate the performances of the above 5 strategies, we simulate the distributions of the 2 classes of patients with stroke, AIS with LVO and AIS without LVO, using the following sampling method where we adopt the probability that a patient is LVO, conditional on their number of CPSS symptoms; and we randomly extract patient data and assume their confirmed diagnosis:

Random sampling method: In our 2010 to 2015 historical data, there were 4037 people with 3 CPSS symptoms, and the probability of LVO in this group was 0.31 based on estimations by Scheitz et al ; that is, 31%, or 1251 out of 4037 patients were estimated to have had LVO. There were 1319 people with 2 CPSS symptoms, and the probability of those patients having LVO was estimated at 0.265: that is, 26.5%, or 350 out of 1319 people. There were 2322 people with 1 CPSS symptom, and the probability of patients having LVO in this group was estimated at 0.239. That is, 23.9% or 555 out of 2322 people. We coded in R to randomly extract patients’ data and assumed that these patients had confirmed LVO. Following the Scheitz et al. probability measure, we extracted 1251 patients from those with 3 CPSS symptoms, 350 patients with 2 CPSS symptoms, and 555 patients with 1 CPSS symptom. Thus, 2156 patients were assumed to have LVO, while the other 5522 patients were assumed to be without LVO.

We used the sampling method 5 times to randomly generate 5 different patient profiles for each probability measure. We then simulated the prehospital process under 5 strategies to determine the patients’ first receiving hospitals and computed the time for each patient to receive treatment over the 5 profiles. We referred to running 5 strategies on 1 profile as a trial. We ran 5 trials for each probability measure.

Results

With the parameters set at the aforementioned initial values, the simulation of the proposed protocol sends all patients with a suspected stroke directly to an EVT‐capable hospital, and 2643 (34.42%) of those bypass the nearest rt‐PA hospital. The results of the primary approach show that, when the tolerable bypass transport threshold and administration time are reduced to 14 minutes and 41.5 minutes,, respectively, a few patients are sent to the rt‐PA hospitals. When the tolerable bypass transport threshold is 9 minutes and the administration time is 30.5 minutes, the number of patients sent to rt‐PA hospitals substantially increases (Figure 2). (The comprehensive results of the numbers of patients sent to an EVT‐capable hospital at different tolerable bypass transport thresholds and administration times are shown in Table S1. The comprehensive results related to those of Richards et al are shown in Figure S1 and Table S2.) To decrease the time needed for a patient to get definitive treatment, we consider the parameter combination in Taipei City of the tolerable bypass transport threshold set to 9 minutes, and administration time set to 30.5 minutes. With these settings, 228 patients are initially sent to the rt‐PA hospitals, and the overall time reduction for the 7678 patients is 767.8 minutes. That is, the 228 patients sent to rt‐PA hospitals can receive definitive treatment an average of 3.3 minutes faster, although they may need more time, such as transfer time, administration time, and 1 more door‐to‐test duration, than a patient sent directly to an EVT‐capable hospital.

Figure 2

Number of patients sent directly to an EVT‐capable hospital at different values of U for A=46.5 and 30.5 minutes.

U (minute): the time difference between the scene to any rt‐PA hospital and the scene to the nearest EVT‐capable hospital. A (minute): the time interval from the first image of CT angiography of brain shown on the computer screen to an rt‐PA hospital departure. CT indicates computed tomography; EVT, endovascular thrombectomy; and rt‐PA, recombinant tissue plasminogen activator.

To balance the provision of medical resources in Taipei City, we consider the parameter combination of set to 6 minutes and set to 30.5 minutes. With these settings, 378 patients are sent to rt‐PA hospitals according to the proposed model. Compared with the results where is 9 minutes with the same , these 378 patients can only reduce their time by 3 minutes before receiving definitive treatment, as an additional 150 patients are initially sent to the rt‐PA hospitals to mitigate congestion in EVT‐capable hospitals (Table 1). The results related to those of Richards et al are shown in Table S3.

Table 1

Primary Approach for Adjusting Threshold U and Administration Time A when the Probabilities of a Patient With Large Vessel Occlusion Showing 1, 2, or 3 Symptoms of the Cincinnati Prehospital Stroke Scale are 0.239, 0.265, and 0.310, Respectively

Tolerable bypass transport threshold U (min)	Administration time A (min)	Number of patients sent to rt‐PA hospitals first	Number of patients sent directly to EVT‐capable hospitals	Expected time that patients receive definitive treatment (min)
15	46.5	0	7678	101.78
9	30.5	228	7450	101.68
6	30.5	378	7300	101.63

EVT‐capable hospital, providing intravenous thrombolysis and endovascular thrombectomy; rt‐PA hospital, providing only intravenous thrombolysis. EVT indicates endovascular thrombectomy; and rt‐PA, recombinant tissue plasminogen activator.

According to the results shown in Table 2, when we used strategy c to determine the receiving hospital, the patients received definitive treatment in the shortest time. Although strategy c sends all patients to EVT‐capable hospitals, which has the same outcome as strategy b with the initial parameters, strategy c saves each patient ≈12 minutes before receiving definitive treatment. This difference is because, in strategy b, some patients are sent to an EVT‐capable hospital that is not the nearest one and strategy c benefits from a shorter door‐to‐treatment duration. In Table 3, all 6 EVT‐capable hospitals can receive patients with strategy b. With strategy c, however, patients are sent only to 3 EVT‐capable hospitals: B1, B2, and B3 (see Figure S2 for the map of the hospital distribution in Taipei City), because strategy c takes into account not only the transport time, but also the discriminative door‐to‐treatment duration in each hospital. The results related to those of Richards et al are shown in Table S4 and S5.

Table 2

Mean Time (in Minutes) for a Patient to Receive Definitive Treatment Under the 5 Strategies for Deciding the Receiving Hospital. (, . Probability measure, Scheitz et al )

	Strategy a	Strategy b	Strategy c	Strategy d	Strategy e
Trial 1	111.92	113.38	101.77	112.43	112.78
Trial 2	111.67	113.22	101.75	112.20	112.50
Trial 3	111.90	113.38	101.77	112.40	112.73
Trial 4	112.35	113.72	101.90	112.87	113.07
Trial 5	111.80	113.38	101.77	112.42	112.65
Average	111.93	113.42	101.79	112.46	112.75

Table 3

Number of Patients Sent to Each Receiving EVT‐Capable Hospitals for Strategies b and c. B1‐B6 Refer to the 6 EVT‐Capable Hospitals. (, . Probability Measure, Scheitz et al )

	B1	B2	B3	B4	B5	B6
Strategy b	983	2104	836	1234	1397	1124
Strategy c	80	5277	2321	0	0	0

EVT indicates endovascular thrombectomy.

According to Table 4, when the parameters of the tolerable bypass transport threshold and administration time are 15 minutes and 46.5 minutes, respectively, the average time for a patient to receive definitive treatment taken over 5 trials is 101.7 minutes, which is very close to the expected time of 101.7 minutes estimated by the model. With 2 other sets of parameters, the average times taken over 5 trials are also close to the expected time estimated by the model. This phenomenon occurs because we used the same probability measure to simulate the LVO patient distribution as we did for the model. The results related to those of Richards et al are shown in Table S6.

Table 4

Mean Time for a Patient to Receive Definitive Treatment for the 5 Trials. (Probability Measure, Scheitz et al )

Tolerable bypass transport threshold U (min)	Administration time A (min)	Expected time for a patient to receive definitive treatment (min)	Trial 1	Trial 2	Trial 3	Trial 4	Trial 5
15	46.5	101.78	101.77	101.75	101.77	101.90	101.78
9	30.5	101.68	101.65	101.72	101.68	101.88	101.68
6	30.5	101.63	101.68	101.72	101.73	102.00	101.63

In the web‐based triage system, EMTs must enter all the required information in the form, which includes the patient's background information, current location, and the number of CPSS symptoms, on the “Acute Stroke Patient Information” page. After clicking on the “submit” button, EMTs will see the “The Best Solutions” page. On the “The Best Solutions” page, there are respectively 3 recommended hospitals based on the proposed protocol and the nearest‐delivery strategy. The EMTs then decide what hospital the patient will be sent to and will submit the result to the database.

Discussion

Currently, EMTs choose the receiving hospital based on the result of the prehospital stroke scale and the time or distance from the scene to the hospitals. However, basing the decision only on the result of the prehospital stroke scale is insufficient because of the inaccuracy of the prehospital stroke scales. To improve the accuracy of the decision and to minimize the time for a patient to receive definitive treatment, and in addition to the variables used by the EMTs, discriminative door‐to‐treatment duration in each hospital and transfer time between hospitals should be considered. The results of our model show that the optimality of a receiving hospital could be significantly affected because the door‐to‐treatment duration in the 6 EVT‐capable hospitals in Taipei City are quite varied, with the difference between the shortest door‐to‐treatment duration and the longest door‐to‐treatment being ≈1 hour. As a result, if a patient is sent to an EVT‐capable hospital with a longer door‐to‐treatment duration, it may take more time for them to receive definitive treatment than being sent to an EVT‐capable hospital further away, but with a shorter door‐to‐treatment duration.

There have been many studies that discuss prehospital triage for patients with acute stroke; however, few of them detail the in‐hospital time. Since the question of whether or not a patient has LVO can only be determined after a hospital test, we take into consideration the probability of a patient having LVO, discuss the possible hospital treatment needed, and calculate the total expected time, which is an important complementary factor in the triage strategy, and not fully addressed in previous research. This additional information gives EMTs a more comprehensive model to work with when making decisions.

The simulation results show that the hospitals providing stroke treatments in Taipei City are sufficient in number and are geographically close to each other. So the difference in transport times between the scene to the nearest rt‐PA hospital and the scene to any EVT‐capable hospital is rarely >15 minutes, which coincidentally makes these results seem to recommend sending a patient directly to an EVT‐capable hospital. If the proposed model is used in different regions, there will be no such results because of the special circumstances of Taipei City. Administration time for hospital transfer also impacts the results. The shorter the transfer time, the more patients with suspected LVO can tolerate being sent initially to an rt‐PA hospital and then transferred before receiving definitive treatment.

In the simulation, the proposed model (strategy c) has the shortest time for a patient to receive definitive treatment when compared with 4 typical strategies. Although no patients are sent to rt‐PA hospitals when using the model with the initial parameters, the time to get definitive treatment is shorter than the results of strategy b, which is to send patients directly to the nearest EVT‐capable hospital. Regarding whether patients can be assigned to rt‐PA hospitals to balance the use of medical resources and to mitigate the potential crowding in EVT‐capable hospitals, we found that shortening the administration time for hospital transfer can resolve the problem. Moreover, if the administration time for transfer is improved to the intended level according to our primary approach, patients in some locations can initially be sent to rt‐PA hospitals and still receive definitive treatment in a shorter expected time.

This model and the web‐based system can be applied to other regions and countries based on the preliminary experiments and validation in this work for Taipei City. The parameters of hospitals should be updated according to the historical data for hospitals in the target region. The tolerable bypass transport threshold and administration time should be adjusted according to a primary approach based on patients’ data in the target region. We believe that the model can help EMTs determine suitable receiving hospitals and that patients can receive definitive treatment in the shortest time. Obtaining an optimal solution to the underlying mathematical model can be done on Microsoft Excel, but using A Mathematical Programming Language and CPLEX, as we did here, ensures the shortest computation time.

Limitations

In our study, the model was tested using 2010 to 2015 historical patient data, and the parameters were set based on historical median durations. The period of patient data was before the major randomized control trials showing a benefit with EVT. , , , , However, these data and averages will gradually change. To ensure the method’s effectiveness, the model’s parameters should be adjusted periodically according to the latest information.

In addition, the model would output different optimal hospitals under different probability measures. Although we examined 2 probability measures, it requires further research to know whether these are close to the true probability measure for other regions, seasons, and races. Increasing the accuracy of the probability measure for the target region would improve the model and reduce the time for a patient to receive definitive treatment.

Finally, in our study, tolerable bypass transport threshold U was initially set to 15 minutes and then was shortened for primary approach. The tolerable bypass transport threshold was suggested to be 30 minutes in recent recommendations in 2021, and it seemed that the initial threshold in our study was shorter. However, since the tolerable bypass transport threshold U was initially set to 15 minutes, the simulation of the proposed protocol already sends all patients with a suspected stroke directly to an EVT‐capable hospital. It is believed that putting a longer threshold than 15 minutes into the model has the same results if the data in Taipei are used. A tolerable bypass transport threshold may be used up to 30 minutes in future models for different areas.

Conclusions

We propose an optimization model that considers not only the probability of a patient having LVO and the real‐time transport, but also the door‐to‐treatment duration in hospitals and the transfer time (secondary transport time), and administration time. Our web‐based system can help EMTs decide on the most suitable receiving hospital and enable patients with a suspected stroke to receive definitive treatment in the shortest time. The system has a generality that can be applied in other regions and countries.

Sources of Funding

The article was supported by the Taiwan Ministry of Science and Technology (MOST 107‐2221‐E‐007‐074‐MY3d, MOST 108‐2314‐B‐002‐131, and MOST 110‐2221‐E‐007‐108‐MY3) and National Taiwan University Hospital (108‐09). This funding source had no role in the design of this study, nor any role during its execution, analyses, interpretation of the data, or decision to submit results.

Disclosures

None.

Data S1–S2

Tables S1–S6

Figures S1–S2

Click here for additional data file.

S_i:	Response time for the ambulance to reach the site of patient i plus on‐scene time
T_a,i:	First transport time from getting patient i on the scene to hospital a
Q_a:	Door‐to‐test duration in hospital a
D_a:	Test‐to‐treatment duration in hospital a for a patient who has AIS without LVO
D¯_a:	Test‐to‐treatment duration in hospital a for a patient with LVO

A:	Administration time of hospital transfer
E_a:	The shortest possible time for a patient to be transferred from an rt‐PA hospital a to an EVT‐capable hospital and to receive definitive treatment, ie, minb∈setofCSCs(T_a,b¯+Q_b+D¯_b)
T_a,b¯:	The secondary transport time from an rt‐PA hospital a to an EVT‐capable hospital b

U:	Tolerable bypass transport threshold determined by the manager