Abstract
Objectives:
Prehospital routing decisions for patients with suspected stroke are complex, and consideration of actual hospital performance data for door-to-needle and door-to-puncture times for thrombolysis and endovascular thrombectomy (EVT) may impact on-scene decision making. We sought to determine whether optimal routing destinations for patients with suspected stroke in the prehospital setting differ between three strategies: 1) American Heart Association (AHA) consensus recommendations, 2) personalized decision modeling without hospital-specific performance data (base model) and 3) with inclusion of actual hospital data for door-to-needle and door-to-puncture times for thrombolysis and EVT (enhanced model).
Methods:
Our previously published decision-analytic model (base model) incorporates geographic, patient, and hospital data to determine optimal transport destination, using a standard distribution of thrombolysis and EVT times by hospital type (primary vs comprehensive stroke center). We enhanced the model by incorporating real hospital performance data on thrombolysis and EVT treatment times from all Northeast United States Get with the Guidelines-Stroke hospitals (enhanced model). We generated 500 probable pick-up locations with 800 patient profiles of varying age, stroke severity, and last known well time, giving 400,000 patient-location scenarios. For each scenario we determined the optimal destination based on AHA consensus recommendations and from 1000 simulations each using base and enhanced models to determine the optimal destination for each strategy. We compared destination recommendations and examined conditions under which results changed.
Results:
Of the 400,000 patient-location scenarios, 56.1% and 63.1% of routing destinations were different from the consensus recommendations in the base and enhanced models, respectively, as patients were more often directed to hospitals with faster times to reperfusion. Higher stroke severity and greater odds of large vessel occlusion were associated with concordance of recommendations.
Conclusions:
Prehospital routing models for patients with suspected stroke may benefit from inclusion of actual hospital performance characteristics.
Introduction
Long-term disability from stroke can be reduced with timely access to reperfusion interventions (1–10). The benefits of reperfusion with thrombolytic and endovascular thrombectomy (EVT) are time-dependent, with faster reperfusion being critical (2,6, 11,12). As a result, configuring prehospital stroke systems to optimize rapid access to time-dependent intervention for stroke patients is complicated. Recent consensus recommendations from the American Heart Association (AHA) and several stakeholder societies for stroke routing in the prehospital setting included a proposed algorithm that suggests transport to an EVT-capable hospital for patients with suspected large vessel occlusion (LVO) if it does not add more than 30–60 min of additional travel time (depending on rurality of the location) (13). However, identification of stroke due to LVO in the prehospital setting remains imperfect (14–16), and some evidence suggests the current routing algorithm is suboptimal as a nationwide policy (17). Furthermore, the authors of the consensus recommendations noted a paucity of evidence to help determine appropriate time thresholds for when to bypass closer hospitals for more comprehensive centers. Their conclusions included a suggestion that prehospital simulation modeling may provide additional evidence to support real-time optimal destination decision-making on-scene.
Because prehospital routing decisions for patients with stroke depend on many factors, existing decision model analyses have incorporated patient and hospital characteristics such as age, stroke severity, and average door-to-reperfusion times (18–20). However, despite extensive evidence of variation between hospitals in timeliness of acute stroke care and reperfusion interventions (21–24), to date, these decision models have not included actual hospital performance data. For a hypothetical patient transported by emergency medical services (EMS) with more than one option for a destination hospital, it is conceivable that—when every minute counts—transport to a hospital with faster time-to-reperfusion performance may benefit the patient the most, thus becoming the optimal transport destination.
To address this question, we sought to determine whether the optimal routing destinations for patients with suspected stroke in the prehospital setting differ between three strategies: 1) AHA consensus recommendations; 2) personalized decision modeling using standard distributions of hospital reperfusion performance data (base model); and 3) personalized decision modeling with inclusion of actual hospital data for door-to-needle and door-to-puncture times for thrombolysis and EVT respectively (enhanced model).
Methods
Overview
This was an experimental simulation-based study design, examining three strategies for prehospital routing of patients with suspected stroke: 1) following AHA consensus recommendations; 2) using the base model of our personalized mathematical decision model; and 3) using the enhanced model of our personalized mathematical decision model which includes actual hospital performance data. We generated 400,000 simulated patient-location scenarios in the Northeast United States (U.S.) and applied the three strategies to each of these scenarios to examine recommended transport destination. The study was considered exempt by our institutional review board.
Consensus Recommendation Routing
To determine the optimal destination based on AHA consensus recommendations, we first used RACE score to determine the probability of LVO, and if the probability was 0.5 or greater, we considered this a scenario in which LVO was suspected. When LVO was not suspected, the recommended destination was the closest stroke center hospital. When LVO was suspected, we used patient pick-up location to determine rurality. For rural locations (those with rural urban commuting area (RUCA) codes of 4–10) the recommendation was to transport to a CSC if it was within 60 min transport time. When a CSC was not within 60 min, the recommendation was to transport to the nearest PSC. For suburban locations (RUCA 2–3), the recommendation was to transport to a CSC if it was within 45 min transport time. When a CSC was not within 45 min, the recommendation was to transport to the nearest PSC. Finally for urban locations (RUCA 1), the recommendation was to transport to a CSC if within 30 min, otherwise the nearest PSC. This was a modest simplification of the consensus recommendations because our hospital performance data from 2015 to 2016 predated the establishment of thrombectomy-capable stroke center (TSC) certification.
Details of the Mathematical Decision Model
For our base model, we use an existing prehospital mathematical decision model that identifies a personalized ideal transport scenario for patients with potential stroke in the prehospital setting. This model incorporates patient characteristics (e.g., stroke severity, last known well time), characteristics of potential hospital options (e.g., stroke related capabilities), and variation in transport times (20). It is designed for use in real-time at the point of EMS initiation of care to recommend an optimal transport destination by ground for individually optimized patient outcomes. We compare simulated patient outcomes based on two versions of the model. The base model uses patient reperfusion treatment times from a standard distribution of times based on the hospital type, using one distribution of times for primary stroke centers (PSCs) and an alternative distribution of times for comprehensive stroke centers (CSCs). The enhanced model draws patient reperfusion treatment times from hospital-specific time-distributions based on each hospital’s actual past performance times. The model was written in Python 3.6.1 and is publicly available in the base model format (hospital performance data are not publicly available) (25).
The personalized decision model has two components: 1) projecting future outcomes based on patient characteristics, and 2) determining the best first destination hospital for patient transport by ground based on patient characteristics, location characteristics, and current traffic patterns. Full details of the model and how estimations are performed have been previously published (20). Key inputs, all of which are available at the time of initial contact with EMS, are: patient characteristics measurable by EMS providers (age, sex, stroke severity, and time from last known well) and geographical location data (proximity to all hospital options and any potential transfer scenarios, as well as hospital capabilities and time-to-treatment performance data). The Rapid Arterial oCclusion Evaluation (RACE) score is used for stroke severity measurement because it is strongly correlated with National Institutes of Health Stroke Scale and probability of large vessel occlusion (26). A RACE score of five or above has the optimal predictive value for identifying LVO, with test characteristics from the original validation study incorporated into our model predicted probability of LVO (26). Patient age and sex are used for lifetime all-cause mortality estimates, and time from last known well and prehospital stroke severity scores are used for estimates of treatment efficacy and post-stroke disability. After the initial event, outcomes are based on the likelihood of the patient to be in each of the following states: independent (modified Rankin score of 0–2), disabled (modified Rankin of 3–5), and death (modified Rankin of 6). The model interfaces with Google Maps for real-time routing and traffic conditions. The output is a recommendation for initial patient transport destination by ground EMS.
The recommendation is derived by incorporating the inputs described above, using parameters obtained from clinical trials and meta-analyses (11,27–31), and generating anticipated patient outcomes based on lifetime cost and quality-adjusted life-years to give the destination that will result in the greatest quality-adjusted life-years at a cost that is not prohibitive. Alternative transport scenarios are compared based on cost-effectiveness, with a maximum threshold of $100,000 per quality-adjusted life-years gained. Because of the uncertainty in the model results, a stochastic approach is employed with each scenario run 1000 times. For each of the 1000 simulations, treatment times for alteplase and EVT are drawn randomly from prespecified time distributions. Of those 1000 simulations, the most frequently occurring optimal destination is given as the final recommendation for the scenario, with a level of confidence based on the proportion of simulations for which the recommended transport scenario was optimal.
Data for hospital performance on reperfusion times are drawn from two sources. The base model uses one distribution of times for all CSCs and another distribution of times for all PSCs based on unpublished data from the Get with the Guidelines (GWTG)-Stroke registry. The enhanced version uses hospital-specific distributions based on each hospital’s actual hospital performance data for median time-to-intervention. These data were provided by the AHA for Northeast hospitals participating in the GWTG–Stroke registry. Times were based on all patients with ischemic stroke receiving thrombolysis or EVT at a hospital participating in GWTG–Stroke registry during a two-year period (January 1, 2015 through December 31, 2016) in the Northeast US. For hospitals without data in the GWTG-Stroke registry, times were drawn from the standard distribution of times for PSC and CSCs (i.e., no change from base model). Of all hospitals in the region, 16.9% were GWTG. TSCs were grouped with CSCs.
Simulated Sample
We sampled 500 sites in the Northeast U.S. based on population density to provide probabilistic hypothetical scenarios. At each of these 500 sites we considered the same set of 800 patient profiles of age, stroke severity, and time from last known well based on the literature (32–34). This led to 400,000 patient-location scenarios. Simulated patients had a mean age of 74 years and 53% were female. Median RACE score was 3 (interquartile range [IQR] 0–6), and median NIHSS score was 6 (IQR 1–13). Median time from symptom onset was 159 min (IQR 70–472). Of the 500 pickup locations, 387 were urban (RUCA 1), 32 were suburban (RUCA 2–3) and 40 were rural (RUCA 4–10); 41 had a missing RUCA code.
Simulation and Concordance Analysis
To determine the optimal destination based on the personalized decision model, we ran the base version of the model to derive an optimal transport destination based on the model using generic hospital reperfusion times, and then ran the enhanced model to derive an optimal transport destination based on the model using actual hospital-specific reperfusion times.
Among all 400,000 patient-location scenarios, we determined the frequency of agreement between the three strategies and identified the characteristics of patient-location scenarios with concordant versus discordant destination recommendations. We also examined the frequency of the recommendation changing from a PSC to a CSC, the frequency of the recommendation changing from a CSC to a PSC, and the frequency of changes within the same category (PSC or CSC). We generated 95% confidence intervals around these numbers.
Results
Comparison of Modeled Destination Recommendations
Of the 400,000 patient-location scenarios, the personalized decision model led to different destination recommendations compared to AHA consensus recommendations in over half of the cases; 56.1% of scenarios in the base version of the model, and 63.1% of scenarios in the enhanced version of the model incorporating actual hospital performance were different from the AHA consensus recommendations (Table 1).
Table 1.
Concordant versus discordant destination recommendations between AHA consensus recommendations, a personalized decision model, and an enhanced personalized decision model incorporating hospital performance (n = 400,000 simulated scenarios).
| Comparison | Concordance | Discordance | Missing | 95% Confidence interval for proportion of Concordant recommendations |
|---|---|---|---|---|
|
| ||||
| Consensus recommendations vs. base model | 167,476 (41.9%) | 224,239 (56.1%) | 8,285 (2.0%) | 0.417, 0.420 |
| Consensus recommendations vs. enhanced model | 139,138 (34.8%) | 252,577 (63.1%) | 8,285 (2.0%) | 0.346, 0.349 |
| Base model vs. enhanced model | 304,169 (76.0%) | 95,825 (24.0%) | 6 (0.002%) | 0.759, 0.762 |
Legend. Missingness based on locations with unavailable RUCA codes, on which the consensus recommendations relied.
When comparing stroke center status of destination recommendations with the consensus recommendations, the consensus recommendations more often favored PSC destinations whereas the base and enhanced personalized decision models more often favored CSCs (Figure 1).
Figure 1.
Frequency of primary versus comprehensive stroke center destination recommendation by source recommendation. (a) consensus recommendations versus base personalized decision model, (b) consensus recommendations versus enhanced personalized decision model, (c) base versus enhanced personalized decision model.
Legend. PSC Primary Stroke Center; CSC Comprehensive Stroke Center; Table 2 gives counts and percentages for each category
Consensus Recommendations versus Base Personalized Decision Model
When the consensus recommendations and base personalized decision model were concordant, the recommended destination was a PSC in 32.6% and a CSC in 67.4% of patient location scenarios. When the consensus recommendations and base personalized decision model were discordant, the consensus recommendations were for PSC in 93.6% and CSC in 6.4%. Among the scenarios for which the consensus recommendations were for a PSC, the base model recommended a different PSC in 1.3% and a CSC in 78.1%. Among the scenarios for which the consensus recommendations were for a CSC, the base model recommended a PSC in 0.9% and a different CSC in 10.3%. These results are illustrated in Figure 1 and Table 2.
Table 2.
Destination recommendations by stroke center type.
| Same PSC | Same CSC | Different PSCs | Different CSCs | PSC from consensus CSC from model | CSC from consensus PSC from model | |
|---|---|---|---|---|---|---|
|
| ||||||
| Consensus Recommendations vs. Base Personalized Decision Model, n (%) | 54,551 (13.9) | 112,925 (28.8) | 3,362 (0.9) | 13,099 (3.3) | 206,614 (52.7) | 1,164 (0.3) |
| Consensus Recommendations vs. Enhanced Personalized Decision Model, n (%) | 47,186 (12.0) | 91,952 (23.5) | 12,857 (3.3) | 31,822 (8.1) | 204,484 (52.2) | 3,414 (0.9) |
| Base vs. Enhanced Personalized Decision Model, n (%) | 41,856 (10.5) | 262,313 (65.6) | 4,275 (1.1) | 60,472 (15.1) | 13,295 (3.3) | 17,783 (4.4) |
Legend. PSC primary stroke center; CSC comprehensive stroke center.
Consensus Recommendations versus Enhanced Personalized Decision Model
When the consensus recommendations and enhanced personalized decision model were concordant, the recommended destination was a PSC in 33.9% and a CSC in 66.1% of patient location scenarios. When the consensus recommendations and enhanced personalized decision model were discordant, the consensus recommendations were for PSC in 86.0% and CSC in 14.0%. Among the scenarios for which the consensus recommendations were for a PSC, the enhanced model recommended a different PSC in 4.9% and a CSC in 77.3%. Among the scenarios for which the consensus recommendations were for a CSC, the enhanced model recommended a PSC in 2.7% and a different CSC in 25.0%. These results are illustrated in Figure 1 and Table 2.
Base versus Enhanced Personalized Decision Model
When the base and enhanced personalized decision models were concordant, the recommended destination was a PSC in 13.8% and a CSC in 86.2% of patient location scenarios. When the base enhanced personalized decision model were discordant, the base model recommendations were for PSC in 18.3% and CSC in 81.7%. Among the scenarios for which the base model recommendations were for a PSC, the enhanced model recommended a different PSC in 7.2% and a CSC in 22.4%. Among the scenarios for which the base model recommendations were for a CSC, the enhanced model recommended a PSC in 5.2% and a different CSC in 17.8%. These results are illustrated in Figure 1 and Table 2.
Characteristics of Scenarios with Changed Recommendations
Table 3 presents characteristics of simulated patient-location scenarios stratified by type of recommendation change between the AHA consensus recommendations and the enhanced personalized decision model. Scenarios with concordant recommendations had higher stroke severity and probability of LVO than discordant scenarios. Among the discordant scenarios, destinations recommended by the enhanced model had shorter door-to-needle times than destinations based on consensus recommendations.
Table 3.
Comparison of simulated patient-location scenario characteristics, by concordance versus discordance between AHA consensus recommendations and enhanced personalized decision tool.
| Concordant recommendations n = 139,138 | Discordant recommendations n = 252,577 | Standard mean difference (SMD) | |
|---|---|---|---|
|
| |||
| Age, mean (SD) | 73.8 (14.8) | 74.5 (14.9) | 0.05 |
| Female gender, % | 52.4 | 52.9 | 0.01 |
| Time since last known well, mean (SD) | 274.38 (298.9) | 343.19 (314.4) | 0.22 |
| NIHSS, mean (SD) | 8.99 (8.6) | 7.46 (7.5) | 0.19 |
| Patients with suspected LVO (%) | 31.6 | 19.3 | 0.29 |
| Rurality of pick-up location, n (%) | 12,870 (9.8) | 19,130 (8.1) | 0.06 |
| Rural | 9,346 (7.1) | 16,253 (6.9) | |
| Suburban Urban | 108,751 (83.0) | 200,847 (85.0) | |
| Median door-to-needle time for destination, Median (IQR) | 56.0 (17.0) | Consensus Recommendation: 64.0 (24.0) Enhanced model recommendation: (17.0) |
p < 0.001* |
Legend. SD standard deviation; NIHSS National Institute of Health Stroke Scale; LVO large vessel occlusion; IQR interquartile range.
Mann-Whitney-U test used to compare medians between the DTN of the hospitals with concordant recommendations versus the DTN of the consensus-recommended hospitals when discordant and the DTN of the enhanced model-recommended hospital when discordant. The median times were significantly different with p < 0.001 for both comparisons.
Figure 2 illustrates the difference in destination sites’ median door-to-needle times for thrombolysis. Among scenarios with concordant recommendations between the AHA consensus recommendations and enhanced personalized decision model, median of hospitals’ median door-to-needle time was 56 min, IQR (48–65). in contrast, among scenarios with discordant destination recommendations, the enhanced personalized model recommended transport to sites with faster DTNs (median 50 min, IQR 41–58), versus consensus recommended sites which had median DTN 64 min, IQR 56–80 (SMD between groups 0.11).
Figure 2.
Boxplot of scenarios’ simulated door-to-needle times by concordance vs. discordance between AHA consensus recommendations and enhanced personalized decision model.
Legend. AHA American Heart Association; DTN: Door-to-needle time
Figure 3 illustrates the difference in destination sites’ median door-to-puncture times. Among scenarios with concordant recommendations between the AHA consensus guidelines and enhanced personalized decision model, median time was 139 min, IQR (104–167). In contrast, among scenarios with discordant destination recommendations, the enhanced personalized model recommended transport to sites with faster DTPs (median 104 min, IQR 70–139), versus consensus recommended sites which had median DTP 165 min, IQR 105–178 (SMD between groups 0.21).
Figure 3.
Boxplot of scenarios’ simulated door-to-puncture times by concordance versus discordance between AHA consensus recommendations and enhanced personalized decision model.
Legend. AHA American Heart Association; DTP: Door-to-puncture time
Discussion
In this regional analysis of prehospital triage destination choice in the Northeast U.S., we found that consideration of hospitals’ past performance on reperfusion times could lead to an alternative optimal transport destination in nearly a quarter of simulated patient-location scenarios. Notably, when the enhanced decision model produced a different recommendation from the AHA consensus recommendations, it typically was to a hospital with faster reperfusion times. Thus, incorporation of hospital performance on reperfusion interventions may have a meaningful impact on patient care delivery and outcomes by facilitating more rapid time to intervention for patients.
These results build on prior work modeling prehospital stroke transport destinations (18–20) by demonstrating the importance of considering actual hospital performance in routing recommendations. Given the critical timeliness of reperfusion interventions and the significant variation between hospitals in performance on timeliness metrics, it is not surprising that hospital performance influenced model recommendations in scenarios where multiple destination options were present. These changes were more frequently found in areas with higher density of population and hospitals. In future work it will be important to consider more patient-centered outcomes beyond hospital destination (e.g., disability-adjusted life years and further quantify the extent of expected benefit to patients with optimal prehospital routing.
Our results are similar to prior work by Holodinsky et al., who reported that transport decisions for ischemic stroke patients were highly sensitive to transportation times between centers and to door-to-needle times (18). However, our modeling strategy is distinct from the prior work in important ways. First, our model does not assume that ischemic stroke is identified with complete accuracy in the field and incorporates diagnostic uncertainty at the time of EMS transport decision-making, including consideration of stroke mimics and hemorrhagic stroke diagnoses. Second, we now include actual historical hospital-specific performance data on door-to-needle and door-to-puncture times. Though hospital performance data must remain confidential, use of an honest broker or software program to assess for real-time decision-making could be a future implementation focus for EMS systems attempting to incorporate the enhanced algorithm. In addition, incorporation of empirical performance variation (e.g., weekend effects) and real-time operational factors (e.g., ED and hospital crowding and system capacity) may further refine predictions and identification of optimal transport destination to enhance EMS decision support. Finally, a truly personalized approach to prehospital destination decisions would further take into account factors such as frailty and patient preferences. Any future implementation of an approach with a decision model such as this one will be most patient-centered and personalized with this in mind.
Limitations
Limitations of this study include that it was confined to a specific geographic region and may not be generalizable to other locations within the U.S. which likely have a different distribution of hospitals and resources. We were only able to include reperfusion performance times from hospitals participating in the GWTG-Stroke registry. This may have biased our results if nonparticipating hospitals had significantly different performance characteristics. Further, hospital performance data used in the modeling were from 2015 to 2016. Despite the age of the data, there is persistent evidence of variation in hospitals’ performance on timeliness of stroke care delivery (35). Thus even if the absolute values of the time metrics have changed with time, the persistence of variation continues to suggest the value of considering this variation in routing destination choice. Our model assumptions include the relationship between timeless of treatment and patient outcomes. While the parameters we used were derived from the literature, we did not model hospital-level heterogeneity in the effect of time-to-treatment on outcomes. Thus, if there is hospital-level variation in the relationship, this is not captured by our approach. We also did not consider the role for acute stroke ready hospitals as we limited potential destination options to PSC and CSC hospitals. This may have led to a more limited hospital choice set for some scenario locations and thus may have biased toward more agreement. Moreover, the landscape of hospital certification has evolved with increasing numbers of hospitals gaining certification status. We also were unable to incorporate performance variation within hospitals (e.g., by time or day or day of week). In settings where such data are available, future models should prioritize their inclusion to account for the contribution to care delivery and patient outcomes. In addition, our model does not incorporate emergency department overcrowding and other dynamic characteristics of the health care system that may influence patient outcomes. Finally, patient-centeredness is a critical component of medical decision-making and future strategies should incorporate patient destination preferences.
Conclusions
We present results from a decision model for optimized routing of patients with suspected stroke in the prehospital setting. Our findings suggest that hospital-specific performance data such as time-to-reperfusion metrics should be considered in transport destination decision making.
Acknowledgments
The authors acknowledge Ayman Ali, MD, and Patrick Eschenfeldt, PhD, for their early work on development of the decision model.
Funding
This work was supported by the MGH Executive Committee on Research and the NIH/NIA (R01 AG079887).
Footnotes
Disclosure Statement
KSZ reports grant funding from NIH/NIA, the Society for Academic Emergency Medicine Foundation and the American College of Emergency Physicians not related to this work. LHS reports consulting activities with Medtronic, LifeImage, Genentech, Penumbra. YC, ZY, FL, MAI and CH have no disclosures. CR reports grants funding from NIH/NINDS. REC reports grant funding from NIH/NICHD and the Society for Academic Emergency Medicine Foundation.
Declaration of Generative AI in Scientific Writing
The authors did not use a generative artificial intelligence (AI) tool or service to assist with preparation or editing of this work. The author(s) take full responsibility for the content of this publication.
Data Sharing Statement
We are not able to share the hospital-specific performance data due to data use agreements, however model code is publicly available on github at the citation included in the manuscript.
References
- 1.Saver JL, Goyal M, Bonafe A, Diener H, Levy EI, Pereira VM, Albers GW, Cognard C, Cohen DJ, Hacke W, et al. Stent-retriever thrombectomy after intravenous t-PA vs. t-PA alone in stroke. N Engl J Med. 2015;372(24):2285–95. doi: 10.1056/NEJMoa1415061. [DOI] [PubMed] [Google Scholar]
- 2.Emberson J, Lees KR, Lyden P, Blackwell L, Albers G, Bluhmki E, Brott T, Cohen G, Davis S, Donnan G, et al. Effect of treatment delay, age, and stroke severity on the effects of intravenous thrombolysis with alteplase for acute ischaemic stroke: a meta-analysis of individual patient data from randomised trials. Lancet. 2014;384(9958):1929–35. doi: 10.1016/S0140-6736(14)60584-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Hacke W, Kaste M, Bluhmki E, Brozman M, Dávalos A, Guidetti D, Larrue V, Lees KR, Medeghri Z, Machnig T, et al. Thrombolysis with alteplase 3 to 4.5 hours after acute ischemic stroke. N Engl J Med. 2008;359(13):1317–29. doi: 10.1056/NEJMoa0804656. [DOI] [PubMed] [Google Scholar]
- 4.Berkhemer OA, Fransen PSS, Beumer D, van den Berg LA, Lingsma HF, Yoo AJ, Schonewille WJ, Vos JA, Nederkoorn PJ, Wermer MJH, et al. A randomized trial of intraarterial treatment for acute ischemic stroke. N Engl J Med. 2015;372(1):11–20. doi: 10.1056/NEJMoa1411587. [DOI] [PubMed] [Google Scholar]
- 5.National Institute of Neurological Disorders and Stroke rt-PA Stroke Study Group. Tissue plasminogen activator for acute ischemic stroke. New Eng J Med. 1995;333(24):1581–8. [DOI] [PubMed] [Google Scholar]
- 6.Lees KR, Bluhmki E, von Kummer R, Brott TJ, Toni D, Grotta JC, Albers GW, Kaste M, Marler JR, Hamilton SA, et al. Time to treatment with intravenous alteplase and outcome in stroke: an updated pooled analysis of ECASS, ATLANTIS, NINDS, and EPITHET trials. Lancet. 2010;375(9727):1695–703. doi: 10.1016/S0140-6736(10)60491-6. [DOI] [PubMed] [Google Scholar]
- 7.Goyal M, Demchuk AM, Menon BK, Eesa M, Rempel JL, Thornton J, Roy D, Jovin TG, Willinsky RA, Sapkota BL, et al. Randomized assessment of rapid endovascular treatment of ischemic stroke. N Engl J Med. 2015;372(11):1019–30. doi: 10.1056/NEJMoa1414905. [DOI] [PubMed] [Google Scholar]
- 8.Albers GW, Marks MP, Kemp S, Christensen S, Tsai JP, Ortega-Gutierrez S, McTaggart RA, Torbey MT, Kim-Tenser M, Leslie-Mazwi T, et al. Thrombectomy for stroke at 6 to 16 hours with selection by perfusion imaging. N Engl J Med. 2018;378(8):708–18. doi: 10.1056/NEJMoa1713973. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Nogueira RG, Jadhav AP, Haussen DC, Bonafe A, Budzik RF, Bhuva P, Yavagal DR, Ribo M, Cognard C, Hanel RA, et al. Thrombectomy 6 to 24 hours after stroke with a mismatch between deficit and infarct. N Engl J Med. 2018;378(1):11–21. doi: 10.1056/NEJMoa1706442. [DOI] [PubMed] [Google Scholar]
- 10.Mazya MV, Cooray C, Lees KR, Toni D, Ford GA, Bar M, Frol S, Moreira T, Sekaran L, Švigelj V, et al. Minor stroke due to large artery occlusion. When is intravenous thrombolysis not enough? Results from the SITS International Stroke Thrombolysis Register. Eur Stroke J. 2018;3(1):29–38. doi: 10.1177/2396987317746003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Saver JL, Goyal M, van der Lugt A, Menon BK, Majoie CBL, Dippel DW, Campbell BC, Nogueira RG, Demchuk AM, Tomasello A, et al. Time to treatment with endovascular thrombectomy and outcomes from ischemic stroke: a meta-analysis. JAMA. 2016;316(12):1279–88. doi: 10.1001/jama.2016.13647. [DOI] [PubMed] [Google Scholar]
- 12.Kim JT, Goyal M, Levy EI, Liebeskind D, Jahan R, Pereira VM, Gralla J, Bonafe A, Saver JL. Onset to reperfusion time as a determinant of outcomes across a wide range of ASPECTS in endovascular thrombectomy: pooled analysis of the SWIFT, SWIFT PRIME, and STAR studies. J Neurointerv Surg. 2020;12(3):240–5. doi: 10.1136/neurintsurg-2019-014906. [DOI] [PubMed] [Google Scholar]
- 13.Jauch EC, Schwamm LH, Panagos PD, Barbazzeni J, Dickson R, Dunne R, Foley J, Fraser JF, Lassers G, Martin-Gill C, et al. Recommendations for regional stroke destination plans in rural, suburban, and urban communities from the prehospital stroke system of care consensus conference: a consensus statement from the American Academy of Neurology. Stroke. 2021;52(5):e133–e152. doi: 10.1161/STROKEAHA.120.033228. [DOI] [PubMed] [Google Scholar]
- 14.Nguyen TTM, Van Den Wijngaard IR, Bosch J, van Belle E, van Zwet EW, Dofferhoff-Vermeulen T, Duijndam D, Koster GT, de Schryver ELLM, Kloos LMH, et al. Comparison of prehospital scales for predicting large anterior vessel occlusion in the ambulance setting. JAMA Neurol. 2021;78(2):157–64. doi: 10.1001/jamaneurol.2020.4418. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Schwamm LH. Admitting the patient with acute stroke to the right house-lessons from the sorting hat of hogwarts. JAMA Intern Med. 2016;176(9):1368–70. doi: 10.1001/jamainternmed.2016.3930. [DOI] [PubMed] [Google Scholar]
- 16.Zachrison KS, Khatri P. Self-driven prehospital triage decisions for suspected stroke - another step closer. JAMA Neurol. 2021;78(2):146–8. doi: 10.1001/jamaneurol.2020.4425. [DOI] [PubMed] [Google Scholar]
- 17.Venema E, Burke JF, Roozenbeek B, Nelson J, Lingsma HF, Dippel DWJ, Kent DM. Prehospital triage strategies for the transportation of suspected stroke patients in the United States. Stroke. 2020;51(11):3310–9. doi: 10.1161/STROKEAHA.120.031144. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Holodinsky JK, Williamson TS, Kamal N, Mayank D, Hill MD, Goyal M. Drip and ship versus direct to comprehensive stroke center: conditional probability modeling. Stroke. 2017;48(1):233–8. doi: 10.1161/STROKEAHA.116.014306. [DOI] [PubMed] [Google Scholar]
- 19.Holodinsky JK, Williamson TS, Demchuk AM, Zhao H, Zhu L, Francis M, Goyal M, Hill MD, Kamal N. Modeling stroke patient transport for all patients with suspected large-vessel occlusion. JAMA Neurol. 2018;75(12):1477–86. doi: 10.1001/jamaneurol.2018.2424. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Ali *A, Zachrison *KS, Eschenfeldt PC, Schwamm LH, Hur C. Optimization of prehospital triage of patients with suspected ischemic stroke: results of a mathematical model. Stroke. 2018;49(10):2532–5. doi: 10.1161/STROKEAHA.118.022041. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Sauser K, Levine DA, Nickles AV, Reeves MJ. Hospital variation in thrombolysis times among patients with acute ischemic stroke: the contributions of door-to-imaging time and imaging-to-needle time. JAMA Neurol. 2014;71(9):1155–61. doi: 10.1001/jamaneurol.2014.1528. [DOI] [PubMed] [Google Scholar]
- 22.Den Hartog SJ, Lingsma HF, van Doormaal PJ, Hofmeijer J, Yo LSF, Majoie CBLM, Dippel DWJ, van der Lugt A, Roozenbeek B. Hospital variation in time to endovascular treatment for ischemic stroke: what is the optimal target for improvement? J Am Heart Assoc. 2022;11(1):e022192. doi: 10.1161/JAHA.121.022192. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Fonarow GC, Smith EE, Saver JL, Reeves MJ, Bhatt DL, Grau-Sepulveda MV, Olson DM, Hernandez AF, Peterson ED, Schwamm LH. Timeliness of tissue-type plasminogen activator therapy in acute ischemic stroke: patient characteristics, hospital factors, and outcomes associated with door-to-needle times within 60 minutes. Circulation. 2011;123(7):750–8. doi: 10.1161/CIRCULATIONAHA.110.974675. [DOI] [PubMed] [Google Scholar]
- 24.Sauser Zachrison K, Levine DA, Fonarow GC, Bhatt DL, Cox M, Schulte P, Smith EE, Suter RE, Xian Y, Schwamm LH. Timely reperfusion in stroke and myocardial infarction is not correlated: an opportunity for better coordination of acute care. Circ Cardiovasc Qual Outcomes. 2017;10(3):e003148. doi: 10.1161/CIRCOUTCOMES.116.003148. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.GitHub - aymannnn/stroke: Optimization of prehospital triage of patients with suspected ischemic stroke: results of a mathematical model. [accessed 2019 Dec 3]. https://github.com/aymannnn/stroke.
- 26.Pérez de la Ossa N, Carrera D, Gorchs M, Querol M, Millán M, Gomis M, Dorado L, López-Cancio E, Hernández-Pérez M, Chicharro V, et al. Design and validation of a prehospital stroke scale to predict large arterial occlusion: the rapid arterial occlusion evaluation scale. Stroke. 2014;45(1):87–91. doi: 10.1161/STROKEAHA.113.003071. [DOI] [PubMed] [Google Scholar]
- 27.Tsivgoulis G, Zand R, Katsanos AH, Goyal N, Uchino K, Chang J, Dardiotis E, Putaala J, Alexandrov AW, Malkoff MD, et al. Safety of intravenous thrombolysis in stroke mimics. Stroke. 2015;46(5):1281–7. doi: 10.1161/STROKEAHA.115.009012. [DOI] [PubMed] [Google Scholar]
- 28.Adams RJ, Fisher M, Furlan AJ, del Zoppo G. Acute stroke treatment trials in the United States. Rethinking strategies for success. Stroke. 1995;26(12):2216–8. doi: 10.1161/01.str.26.12.2216. [DOI] [PubMed] [Google Scholar]
- 29.Hong KS, Saver JL. Years of disability-adjusted life gained as a result of thrombolytic therapy for acute ischemic stroke. Stroke. 2010;41(3):471–7. doi: 10.1161/STROKEAHA.109.571083. [DOI] [PubMed] [Google Scholar]
- 30.Achit H, Soudant M, Hosseini K, Bannay A, Epstein J, Bracard S, Guillemin F. Cost-effectiveness of thrombectomy in patients with acute ischemic stroke: the THRACE randomized controlled trial. Stroke. 2017;48(10):2843–7. doi: 10.1161/STROKEAHA.117.017856. [DOI] [PubMed] [Google Scholar]
- 31.Dachs RJ, Burton JH, Joslin J. A user’s guide to the NINDS rt-PA stroke trial database. PLoS Med. 2008;5(5):e113. doi: 10.1371/journal.pmed.0050113. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Fonarow GC, Pan W, Saver JL, Smith EE, Reeves MJ, Broderick JP, Kleindorfer DO, Sacco RL, Olson DM, Hernandez AF, et al. Comparison of 30-day mortality models for profiling hospital performance in acute ischemic stroke with vs without adjustment for stroke severity. JAMA. 2012;308(3):257–64. doi: 10.1001/jama.2012.7870. [DOI] [PubMed] [Google Scholar]
- 33.Schwamm LH, Fonarow GC, Reeves MJ, Pan W, Frankel MR, Smith EE, Ellrodt G, Cannon CP, Liang L, Peterson E, et al. Get with the guidelines-stroke is associated with sustained improvement in care for patients hospitalized with acute stroke or transient ischemic attack. Circulation. 2009;119(1):107–15. doi: 10.1161/CIRCULATIONAHA.108.783688. [DOI] [PubMed] [Google Scholar]
- 34.Zachrison KS, Onnela JP, Hernandez A, Reeves MJ, Camargo CA, Cox M, Matsouaka RA, Metlay JP, Goldstein JN, Schwamm LH. Ischemic stroke transfer patterns in the Northeast United States. J Stroke Cerebrovasc Dis. 2019;28(2):295–304. doi: 10.1016/j.jstrokecerebrovasdis.2018.09.048. [DOI] [PubMed] [Google Scholar]
- 35.Mac Grory B, Asif KS, Xu H, Alhanti B, Lusk JB, Hasan D, Park S, Boehme A, Zachrison KS, Goyal M, et al. Association of component strategies of the target stroke phase 3 nationwide quality improvement program with accelerated door-to-puncture and door-in-door-out times for ischemic stroke endovascular thrombectomy in the United States. Circ Cardiovasc Qual Outcomes. 2025;18(11):e012456. doi: 10.1161/CIRCOUTCOMES.125.012456. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
We are not able to share the hospital-specific performance data due to data use agreements, however model code is publicly available on github at the citation included in the manuscript.