The Finnish Prehospital Stroke Scale Detects Thrombectomy and Thrombolysis Candidates—A Propensity Score-Matched Study Jyrki P. Ollikainen, MD,* Heikki V. Janhunen, MD,† Juho A. Tynkkynen, MD,‡ Kalle M. Mattila, MD, PhD,† Minna M. Hälinen, MD,† Niku K. Oksala, MD, PhD, DSc,§,1 and Satu-Liisa K. Pauniaho, MD, PhD‖,1
Background: Prehospital stroke triage is challenged by endovascular treatment for large vessel occlusion (LVO) being available only in major stroke centers. Conjugate eye deviation (CED) is closely related to LVO, whereas common stroke signs (face-arm-leg-speech-visual) screen stroke. We hypothesized that combining CED with common stroke signs would yield a prehospital stroke scale for identifying both LVO and stroke in general. Methods and Results: We retrospectively analyzed consecutive patients (n = 856) with prehospital Code Stroke (recanalization candidate). The National Institutes of Health Stroke Scale (NIHSS) and computed tomography were administered to patients on arrival. Computed tomography angiography was performed on patients with NIHSS score of 8 or greater and considered to benefit from endovascular treatment. With random forest analysis and deviance analysis of the general linear model we confirmed the superiority of the NIHSS “Best Gaze” over other NIHSS items in detecting LVO. Based on this and commonly used stroke signs we presented the Finnish Prehospital Stroke Scale (FPSS) including dichotomized face drooping, extremity weakness, speech difficulty, visual disturbance, and CED. FPSS detected LVO with a sensitivity of 54%, specificity of 91%, positive predictive value of 48%, negative predictive value of 93%, and likelihood ratio of 6.2. Conclusions: Based on CED and universally used stroke signs, FPSS recognizes stroke in general and additionally, LVO as a stroke subtype comparably to other scales intended to detect LVO only. As the FPSS items are dichotomized, it is likely to be easy for emergency medical services to implement. Key Words: Conjugate eye deviation—ischemic stroke—large vessel occlusion—endovascular treatment—prehospital stroke scale. © 2018 National Stroke Association. Published by Elsevier Inc. All rights reserved.
From the *Department of Neurosciences and Rehabilitation, Tampere University Hospital, Tampere, Finland; †Emergency Department, Central Finland Central Hospital, Jyväskylä, Finland; ‡Faculty of Medicine and Life Sciences, University of Tampere, Tampere, Finland; §Faculty of Medicine and Life Sciences, Surgery, University of Tampere, Tampere, Finland and Division of Vascular Surgery, Department of Surgery, Tampere University Hospital, Tampere, Finland; and ‖Emergency Division, Tampere University and Tampere University Hospital. Received June 10, 2017; revision received September 29, 2017; accepted October 11, 2017. Grant support: This study was financially supported by the Competitive State Research Financing of the Expert Responsibility Area of Tampere University Hospital, the Finnish Association of Emergency Medicine, and the Finnish Cultural Foundation Pirkanmaa Regional Fund. Address correspondence to Niku K. Oksala, MD, PhD, DSc, Department of Surgery, Tampere University Hospital, PL 2000, 33521 Tampere, Finland. E-mail:
[email protected]. 1 These authors contributed equally. 1052-3057/$ - see front matter © 2018 National Stroke Association. Published by Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.jstrokecerebrovasdis.2017.10.015
Journal of Stroke and Cerebrovascular Diseases, Vol. 27, No. 3 (March), 2018: pp 771–777
771
J.P. OLLIKAINEN ET AL.
772
Introduction The major predictor of recovery in cases of acute ischemic stroke is the time elapsing between onset of symptoms and recanalization of the artery.1 Along with public awareness, the first contacts with emergency response centers (ERCs) and emergency medical services (EMSs) are crucial. Stroke triage is based on the use of prehospital stroke scales enabling early identification of recanalization candidates (Code Stroke). Use of intravenous tissue plasminogen activator available in primary stroke centers is efficacious in occlusions of small- and medium-sized vessels (non-large vessel occlusion [non-LVO]), whereas endovascular treatment (EVT), more efficacious than tissue plasminogen activator alone in LVO, is only available in major stroke centers, capable of providing EVT.2 The different approaches in treating these 2 entities highlight the importance of the primary triage. Prehospital recognition of LVO is essential for direct transport to major stroke centers.3 Early stroke recognition scores like the Cincinnati Prehospital Stroke Scale and its modification, the Face Arm Speech Test (FAST), are highly sensitive and moderately specific.4,5 Containing dichotomized items (i.e., “yes” or “no” answers) for facial, upper extremity, and speech function, they are easy to remember and performed in a minute. Designed to detect strokes of the anterior circulation, they are not sensitive to strokes of the posterior circulation.6 A number of stroke scales have been proposed to detect LVO: the 3-Item Stroke Scale (3I-SS),7 the Los Angeles Motor Scale (LAMS),8 the Cincinnati Stroke Triage Assessment Tool (C-STAT) (first introduced as the Cincinnati Prehospital Stroke Severity Scale),9,10 the Prehospital Acute Stroke Severity Scale (PASS),11 the Rapid Arterial oCclusion Evaluation Scale (RACE)12 Stroke vision, aphasia, neglect assessment (VAN),13 the Field Assessment Stroke Triage for Emergency Destination (FAST-ED),14 the Face Arm Speech Test with Gaze Deviation (G-FAST),15 and shortened versions of the NIHSS16 like the National Institutions of Health Stroke Scale-8 (NIHSS-8).17 All except the LAMS include conjugated eye deviation (CED). A thrombus of the first segment (M1) of the middle cerebral artery is by far the most common target for EVT.18 Regions participating in visual horizontal spatial attention are cortical Brodmann 8 and 40 and subcortical nuclei putamen and caudate. Damage to any of the regions may cause CED (usually accompanied by head turning), but the likelihood is greatest if all the regions are involved.19 This is logically followed by an M1 thrombus, feeding all the regions, likely causing the characteristic sign, horizontal CED away from hemiparesis. It is regularly also seen in hemorrhagic stroke,20 but predicts a greater likelihood of ischemic stroke which is 10-fold more frequent.21 In epilepsy, CED or head turning is usually in the opposite direction, toward the symptomatic side of the body.22
In the recognition of stroke in general, the “Face Arm Speech Test”5,6 is improved when combined with the “leg” and “visual symptoms.”23 We hypothesized that CED together with 1 or more symptoms of contralateral hemiparesis is sufficient to predict the M1 thrombus, thereby releasing the rest of the score items to assist in nonLVO stroke recognition.
Methods Patients We analyzed a retrospective cohort of 856 consecutive patients with prehospital Code Stroke in 2 Finnish hospitals within a 4.5-hour time window (Tampere University Hospital and Central Finland Central Hospital from January 1, 2014 to March 31, 2015 (study cohort, Table 1). The Code Stroke was assigned during prenotification from EMS to emergency department (ED) physician in patients considered eligible for thrombolysis treatment, having any of the symptoms used by Finnish EMSs in stroke recognition: face (1-sided facial weakness), extremity (arm or leg weakness), speech (slurring, wrong words, or unable
Table 1. Study cohort of consecutive patients arriving to hospital with Code Stroke (n = 856), Tampere University Hospital and Central Hospital of Finland (January 1, 2014 to March 31, 2015) Stroke Age (standard deviation) Gender (% male) Discharge diagnosis Ischemic stroke Hemorrhagic stroke Seizure Migraine Syncope, presyncope, or hypotension Intoxication Dissociative disorder or simulation Old deficit Muskuloskeletal symptom Hypoglycemia Otogenic vertigo Infection Metabolic, other than hypoglycemia Chronic subdural hematoma Acute brain trauma Subarachnoidal hemorrhage Peripheral neuropathy Brain tumor without seizure Other or unknown Total
Stroke mimics 65 (16.0) 52
462 108 38 36 25 24 22 18 18 16 13 13 10 9 8 7
577
7 4 18 279
SIMPLE STROKE SCALE FOR THROMBECTOMY CANDIDATES
to speak), visual (loss of vision in 1 area of the visual field). The items are commonly used in EMS and ERC protocols for stroke recognition including US Card 28 protocol for ERCs.23-25 The NIHS Score was first examined by a specialist or a specializing physician (neurology, emergency medicine) when patients arrived at the ED. Thereafter computed tomography (CT) of the brain was performed within 20 minutes of arrival at the ED on all patients with clinical suspicion of stroke in neurologic examination. CT was not performed on 77 patients whose symptoms were considered clinically clearly other than stroke (migraine n = 29, seizure n = 12, dissociative disorder n = 10, otogenic vertigo n = 9, musculoskeletal symptom n = 7, peripheral neuropathy [e.g., radial mononeuropathy] n = 6, syncope n = 4). Computed tomography angiography (CTA) was performed on patients with ischemic stroke and NIHSS score of 8 or greater and who on the basis of their CT findings and previous medical histories were considered to benefit from EVT. CTA was not necessarily performed, for example, due to detection of a large irreversible infarction in noncontrast CT, prior morbidities like renal insufficiency, severe congestive heart failure, or mental impairment due to a neurodegenerative disorder. LVO included a CTA-confirmed thrombi at following locations: (1) internal carotid artery (ICA), (2) continuing from ICA to M1 and A1-branches (ICA-T), (3) involving more than two thirds of M1-branch (entire M1), (4) involving not more than first half of M1branch when M1 was divided into 2 parts of equal length (proximal M1), (5) involving not more than second half of M1 branch (distal M1), (6) involving a large M2 branch simulating M1 trunk or located in more than 1 M2 branches (dominant M2), (7) single M2 branch (M2), and (8) basilar artery (BA). A thrombus in the only (functioning) vertebral artery is comparable to a BA thrombus. No such patients were observed in our cohort. This study was approved by the Head of Tampere University Science Centre (R15529).
Statistical Methods Classification-based random forest analysis using a forest of 5000 trees was used to test whether “Best Gaze” was the best predictor of LVO, and, additionally, to evaluate the hierarchy of other NIHSS items in predicting LVO. NIHSS items were used as such and the classification forest was built with all the study subjects. Median decrease of accuracy (MDA) was used to assess the importance of separate NIHSS items. Even though we expected to see some level of collinearity in the regression model, we calculated the deviance and odds ratios (ORs) of each NIHSS item in a binomial regression model to evaluate the weight of each individual NIHSS item in predicting LVO.
773
The negative predictive value (NPV), positive predictive value (PPV), sensitivity, specificity, positive likelihood ratio, and areas under the receiver operating characteristic curve (AUC) and comparison of AUCs were calculated using 1000 bootstrap replicates to compensate the overfitting of the generated score. Below we present median bootstrapped results with basic 95% confidence intervals (95% CIs). We categorized all ischemic stroke patients without CTA (“No CTA”) to the non-LVO group. This could lead to increased numbers of false positive findings (Finnish Prehospital Stroke Scale [FPSS] ≥5 with LVO categorized as non-LVO ischemic stroke due to lack of CTA) and increased numbers of true negative findings. This possible misclassification leads to decreased PPV and increased NPV. Sensitivity and specificity will also change due to this misclassification, but the direction cannot be determined directly. To estimate this bias we conducted the same analyses among the patients with the CTA propensity score matched to those of non-CTA patients. “No CTA” was used as a “treatment” indicator and propensity score was estimated using NIHSS items No. 4 “Facial Palsy,” No. 5 a-b “Motor Arm,” No. 9 “Best Language,” and No. 2 “Best Gaze.” These NIHSS items were included in the FPSS score but raw numbers from the NIHSS were used. To achieve a better match the nonCTA group was randomly restricted to half (n = 247) using propensity score tentile-specific selection. All statistical analyses were based on the Study cohort (Table 1). All analyses were performed using “survival,” “boot,” “MatchIt,” “Proc,” and “Random Forest” packages of R version 3.2.3 software (R Foundation for Statistical Computing, Vienna, Austria).26-31
Results Out of 856 Code Strokes activated during the study period, 462 were ischemic strokes, 115 were hemorrhagic strokes, and 279 were stroke mimics (Table 1). A total of 107 LVOs was detected (23% of all ischemic strokes). In the random forest analysis the greatest decreases in mean accuracy (MDA) were seen in the NIHSS item “Best Gaze” (75.3%). The other NIHSS items achieved MDA as follows: “Extinction and Inattention (neglect)” (66.7%), “Motor Arm” (43.1%), “Motor Leg” (40.5%), “Level of Consciousness (LOC)” (39.3%), “Best Language” (37.5%), “Sensory” (30.8%), “Facial Palsy” (29.7%), “Level of Consciousness Questions (LOCq)” (29.0%), “Dysarthria” (27.0%), “LOC Commands (LOCc)” (14.2%), “Visual” (8.0%), and “Limb Ataxia” (−12.9%). As a positive control for the key NIHSS items in LVO prediction, binomial regression model was performed: “Best Gaze” (OR 2.2, 95% CI 1.6-3.1, P < .0001), “Sensory” (OR 1.7, 1.2-2.3, P = .002), “Facial Palsy” (OR 1.6, 1.2-2.2, P = .003), “LOC” (OR .5, .3-0.8, P = .01), and “Dysarthria” (OR 1.5, 1.1-2.1, P = .01) were associated with LVO. “Best Gaze”
J.P. OLLIKAINEN ET AL.
774
Table 2. Finnish Prehospital Stroke Scale (FPSS) Points
NIHSS correspondence Item No. 4 “Facial Palsy” scored over 0 Item No. 5 “Motor Arm” or No. 6 “Motor Leg” scored over 0 Item No. 9 “Best Language” or No. 10 “Dysarthria” scored over 0 Item No. 3 “Visual” scored over 0 Item No. 2 “Best Gaze” scored over 0
Face Extremity
Facial droop Weakness of 1 or more extremities
1 1
Speech
Difficulty of understanding or producing speech, including slurring Field cut or blindness Partial or fixed gaze or head deviation away from the paretic side 1-4 predicts non-LVO, ≥5 predicts LVO
1
Vision Gaze Total points
1 4 0-8
Abbreviations: LVO, large vessel occlusion; endovascular treatment candidate; Non-LVO, small- or medium-sized vessel occlusion; ivthrombolysis candidate.
showed the greatest deviance drop among the NIHSS items when each item was introduced separately into the binomial regression model. According to our hypothesis and the results presented above, the dichotomized FPSS items were weighted so that face asymmetry, extremity weakness, speech, or visual disturbance in any combination with CED will predict LVO, whereas any combination of the items excluding CED will predict stroke in general. FPSS and its correlation with NIHSS items are described in Table 2. With NIHSS score 1 or over from the “Best Gaze” the mean summed NIHSS score was 10.2 (SD 5.1) (“Best Gaze” not summed to this) and with NIHSS score 0 from “Best Gaze” the mean NIHSS score was 4.3 (SD 4.7). The corresponding motor only (arm + leg) NIHSS sums were 5.8 (SD 2.6) and 1.6 (SD2.4), respectively. Nearly 15% (n = 124) of the 856 patients had FPSS scores of 5 or greater, suggesting LVO. Of these patients, 58 (47%) were cases of CTA-confirmed LVO, 31
(25%) were cases of ischemic stroke with no CTA (patients ineligible for EVT or active treatment), 27 (22%) were cases of hemorrhagic stroke, 6 (5%) were cases of ischemic stroke with CTA negative to LVO, and 2 (2%) were patients with epileptic status. PPV and NPV were 46.7% (95% CI 38.6-55.2) and 93% (95% CI 91.5-95.2) to detect LAO using FPSS with predetermined cutoff point of 5 or greater, and the overall performances of LVO scales in comparison (containing items derivable from NIHSS) are presented in Table 4. The baseline distribution of vascular territories is shown in Table 3. When comparing proximal and more distal LAOs, FPSS was able to detect 70% of cases of proximal LVO (ICA-M1 + BA), and 24% of cases of distal LVO (M2). As all ischemic stroke patients with no CTA were categorized as non-LVO strokes, we conducted a propensity score-matched analysis to estimate this bias. In this matched set of patients with CTA performed, the results remained almost the same (n = 247). There were
Table 3. Sensitivity of the Finnish Prehospital Stroke Scale (FPSS) according to artery and level of large vessel occlusion (LVO) in the FPSS study cohort of consecutive patients admitted to hospital with Code Stroke (n = 856)
All
ICA
ICA-T
Entire M1
Prox. M1
Distal M1
Dom. M2
M2
BA
No. of cases 107 29 4 9 8 11 7 30 9 with LVO No. of cases 58 21 3 8 8 7 5 4 2 with FPSS ≥5 54 72 75 89 100 64 71 13 22 Sensitivity of FPSS for (44.3-63.9) (52.8-87.3) (19.4-99.4) (51.8-99.7) (63.0-100) (30.8-89.1) (29.0-96.3) (9.09-61.4) (2.81-60.0) LVO (%) (95% CI, %) Abbreviations: BA, thrombus of basilar artery; distal M1, M1-thrombus containing not more than second half of M1-branch; dom. M2, thrombus of a large M2 segment simulating M1 trunk or thrombi in more than 1 M2 branches; entire M1, M1-thrombus containing more than two thirds of M1-branch; ICA, thrombus of internal carotid artery; ICA-T, thrombus continuing from ICA to M1 and A1 branch; M2, thrombus of single M2 branch; prox. (proximal) M1, M1-thrombus containing not more than first half of M1-branch.
SIMPLE STROKE SCALE FOR THROMBECTOMY CANDIDATES
775
Table 4. Diagnostic characteristics of the Finnish Prehospital Stroke Scale (FPSS) compared with scores designed for LVO detection in the FPSS study cohort
Score FPSS NIHSS-8 G-FAST ≥3 G-FAST = 4 3I-SS C-STAT PASS FAST-ED
Sensitivity (%)
Specificity (%)
PPV (%)
NPV (%)
LR+
AUC
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
54.3 (44.8-63.4) 62.6 (53.2-71.9) 74.5 (66.9-83.2) 35.5 (26.4-44.4) 41.8 (33.0-51.5) 54.0 (45.1-63.9) 68.5 (59.5-77.4) 61.6 (31.9-47.2)
91.2 (89.3-93.5) 88.7 (86.6-91.19) 83.0 (80.3-85.7) 96.5 (95.4-97.9) 92.7 (90.9-94.5) 90,5 (88.3-92.5) 85.4 (82.9-87.7) 84.2 (52.8-72.0)
46.8 (38.6-55.2) 44.3 (36.4-51-9) 38.6 (32.1-45.7) 59.3 (47.9-71.3) 44.8 (35.0-55.1) 44.9 (36.5-53.4) 40.0 (32.7-47.0) 39.5 (40.0-47.2)
93.3 (91.5-95.2) 94.4 (92.6-96.0) 95.9 (94.4-97.4) 91.3 (89.4-93.3) 91.8 (89.9-93.9) 93.3 (91.6-95.2) 95.0 (93.3-96.6) 94.0 (92.4-95.9)
6.2 (3.9-7.7) 5.5 (4.0-6.8) 4.4 (3.5-5.2) 10.1 (4.7-13.9) 5.7 (3.4-7.4) 5.7 (3.9-7.1) 5.0 (3.6-5.6) 4.6 (3.2-5.9)
0.848 (0.808-0.887) 0.843 (0.80-0.89) 0.848 (0.808-0.886)
0.782 (0.730-0.830) 0.794 (0.747-0.841) 0.805 (0.763-0.847) 0.841 (0.800-0.879)
P value for AUC difference between AUC compared with FPSS
NA 0.697
0.981 0.000 0.000 0.002 0.645
Abbreviations: 3I-SS, 3-Item Stroke Scale; AUC, area under the receiver operating characteristic curve; CI, confidence interval; C-STAT, Cincinnati Stroke Triage Assessment Tool; FAST-ED, Field Assessment Stroke Triage for Emergency Destination; G-FAST, Best Gaze added to Face Arm Speech Test items; LR+, positive likelihood ratio (sensitivity/1 − specificity, higher value means higher likelihood that disease exists after positive scoring result); LVO, large vessel occlusion; NA, not available; NIHSS-8, shortened NIHSS; NPV, negative predictive value; PASS, Prehospital Acute Stroke Severity Scale; PPV, positive predictive value. Median bootstrap results with 95% CI of basic bootstrap method are presented.
altogether 45 cases of LVO and 26 patients had FPSS of 5 or greater. As expected, NPV decreased (87.9, 95% CI 83.7-92.2) and PPV increased (69.2, 95% CI 51.5-88.5) slightly. Sensitivity 40 (95% CI 25.5-53.3) and Specificity 96.1 (95% CI 93.5-99.0) were also changed but only modestly.
Discussion Vulpian’s law or Prévost sign, conjugate eye deviation (CED) toward an acute cerebral hemispheric lesion, was first published in 1865.32 The FPSS incorporates Prévost sign to predict the M1 thrombus, with universally recognized stroke signs (face, extremity, speech, vision) to predict stroke in general.4,5,23,24 FPSS recognized LVO in our cohort with at least at the same accuracy as the scales intended to predict LVO alone. FPSS detected 70% of cases of proximal LVO even if basilar thrombus was included (ICA-M1, BA). The sensitivity was best for thrombus of the proximal or entire M1 segment, whereas it seemed to decrease in more distal vascular regions. Given the neurofunctional explanation of CED this is logical.19 As FPSS was developed to detect LVO of the anterior circulation, it is no surprise that basilar thrombosis was mostly missed.
LVO scores 3I-SS, LAMS with 3 items scored, C-STAT and PASS with 3, and G-FAST with 4 dichotomized items are short and structured. RACE and FAST-ED with 6 and NIHSS-8 with 8 items scored are more complex. VAN seeks arm drift with 1 additional cortical symptom: visual disturbance, aphasia, or neglect (extinction). However, abstract symptoms like extinction (VAN, NIHSS-8, FAST-ED) and agnosia (RACE) as well as subtle symptoms like visual quadrant field cut or receptive aphasia (VAN) may be difficult to teach for EMS. The accuracies of several LVO scores reported in original publications were higher than that detected in the present study. This is likely due to the validation cohorts. For example, all the scores in comparison (Table 4) include CED, which is regularly present in hemorrhagic stroke and in epileptic status as well.20,22 Thus, the absence of hemorrhages and stroke mimics in evaluation cohorts is likely to result in incorrectly high specificity for LVO. Accuracy is not just about high sensitivity or high specificity, but also about the balance between them. For example, G-FAST of 3 or greater had the highest sensitivity of the scores in the comparison (Table 4), but the lowest specificity, whereas G-FAST of 4 had the lowest sensitivity and the highest specificity. To balance this, the
776
authors suggest using 2 different cutoff points depending on time and distance; patients with G-FAST of 4 with a long transfer distance should be triaged directly to a stroke center capable of providing EVT. However, even with a high specificity of G-FAST of 4, a sensitivity of about 35% is low. When comparing LVO scores, FPSS with a single cutoff point is ranked second highest in both PPV and positive likelihood ratio. In comparison of AUC values, discrimination abilities of FPSS for LVO seemed slightly better than those of 3I-SS, C-STAT, and PASS. However, when compared with NIHSS-8, G-FAST, and FAST-ED, no statistically significant difference was seen, and these scores seem to recognize LVO quite well. Yet, as discussed above, NIHSS-8 and FAST-ED containing several scored items and G-FAST with 2 cutoff points may be more challenging to implement to prehospital setting compared with FPSS. Furthermore, unlike LVO scores, FPSS is designed for both stroke screening and LVO detection, making concomitant use of 2 separate scores unnecessary. There were limitations to our study. Our cohort, consisting of normal clinical patient flow, was still retrospective and CTA was performed only on patients expected to benefit from EVT. However, in the propensity score matching, the results remained similar. Again, we collected the data from 2 independent stroke centers and performed statistical analyses using resampling methods to control for the overfit of the model. Furthermore, a strength of our study is that as it included consecutive Code Stroke patients, our cohort did not miss hemorrhagic strokes or stroke mimics as did most of earlier studies.7-9,11-15
Conclusion The FPSS is, to the best of our knowledge, the first scale to be designed for recognition of stroke in general and additionally, LVO as a stroke subtype. Despite being simpler than most of the scores designed to detect LVO alone, it appears to detect LVO with at least the same accuracy. A prospective comparison of the LVO scores to directly compare true and false positive detection rates is warranted. Whether FPSS can be implemented in both EMSs and ERCs remains to be evaluated.
References 1. Mazighi M, Meseguer E, Labreuche J, et al. Dramatic recovery in acute ischemic stroke is associated with arterial recanalization grade and speed. Stroke 2012;43:29983002. 2. Campbell BCV, Donnan GA, Lees KR, et al. Endovascular stent thrombectomy: the new standard of care for large vessel ischaemic stroke. Lancet 2015;14:846-854. 3. Zhao H, Coote S, Pesavento L, et al. Large vessel occlusion scales increase delivery to endovascular centers without excessive harm from misclassifications. Stroke 2017;48:568573.
J.P. OLLIKAINEN ET AL. 4. Kothari RU, Pancioli A, Liu T. Cincinnati Prehospital Stroke Scale: reproducibility and validity. Ann Emerg Med 1999;4:373-378. 5. Nor AM, McAllister C, Louw SJ, et al. Agreement between ambulance paramedic- and physician-recorded neurological signs with face arm speech test (FAST) in acute stroke patients. Stroke 2004;35:1355-1359. 6. Huwez FU. The FAST tool is effective but not for posterior fossa stroke. BMJ 2015;350:h994. 7. Singer OC, Dvorak F, du Mesnil de Rochemont R, et al. A simple 3-item stroke scale: comparison with the National Institutes of Health Stroke Scale and prediction of middle cerebral artery occlusion. Stroke 2005;36:773-776. 8. Nazliel B, Starkman S, Liebeskind D, et al. A brief prehospital stroke severity scale identifies ischemic stroke patients harboring persisting large arterial occlusions. Stroke 2008;39:2264-2267. 9. Katz BS, McMullan JT, Sucharew H. Design and validation of a prehospital scale to predict stroke severity: Cincinnati Prehospital Stroke Severity Scale. Stroke 2015;46:1508-1512. 10. McMullan JT, Katz B, Broderick J. Prospective prehospital evaluation of the Cincinnati Stroke Triage Assessment Tool. Prehosp Emerg Care 2017;21:481-488. 11. Hastrup S, Damgaard D, Johnsen SP, et al. Prehospital Acute Stroke Severity Scale to predict large artery occlusion: design and comparison with other scales. Stroke 2016;47:1772-1776. 12. Pérez de la Ossa N, Carrera D, Gorchs M, et al. Design and validation of a prehospital stroke scale to predict large arterial occlusion: the Rapid Arterial oCclusion Evaluation Scale. Stroke 2014;45:87-91. 13. Teleb MS, Ver Hage A, Carter J, et al. Stroke vision, aphasia, neglect (VAN) assessment—a novel emergent large vessel occlusion screening tool: pilot study and comparison with current clinical severity indices. J Neurointerv Surg 2017;9:122-126. 14. Lima FO, Silva GS, Furie KL, et al. Field assessment stroke triage for emergency destination: a simple and accurate prehospital scale to detect large vessel occlusion strokes. Stroke 2016;47:1997-2002. 15. Scheitz JF, Abdul-Rahim AH, MacIsaac RL, et al. Clinical selection strategies to identify ischemic stroke patients with large anterior vessel occlusion: results from SITSISTR. Stroke 2017;48:290-297. 16. Lee CF, Venketasubramanian N, Wong KS, et al. Comparison between the original and shortened versions of the National Institutes of Health Stroke Scale in ischemic stroke patients of intermediate severity. Stroke 2016;47. 17. Demeestere J, Garcia-Esperon C, Lin L, et al. Validation of the National Institutes of Health Stroke Scale-8 to detect large vessel occlusion in ischemic stroke. J Stroke Cerebrovasc Dis 2017;pii: S1052-3057:30125-30128. 18. Goyal M, Menon BK, van Zwam WH, et al. Endovascular thrombectomy after large-vessel ischaemic stroke: a meta-analysis of individual patient data from five randomised trials. Lancet 2016;387:1723-1731. 19. Singer OC, Humpich MC, Laufs H, et al. Conjugate eye deviation in acute stroke: incidence, hemispheric asymmetry, and lesion pattern. Stroke 2006;37:27262732. 20. Sato S, Koga M, Yamagami H, et al. Conjugate eye deviation in acute intracerebral hemorrhage: stroke acute management with urgent risk-factor assessment and improvement—ICH (SAMURAI-ICH) study. Stroke 2012; 43:2898-2903.
SIMPLE STROKE SCALE FOR THROMBECTOMY CANDIDATES 21. Andersen KK, Olsen TS, Dehlendorff C, et al. Hemorrhagic and ischemic strokes compared: stroke severity, mortality, and risk factors. Stroke 2009;40:20682072. 22. Bauer G, Broessner G, Unterberger I, et al. Head turning as a prominent motor symptom in status epilepticus. Epileptic Disord 2008;10:119-129. 23. Aroor S, Singh R, Goldstein LB. BE-FAST (Balance, eyes, face, arm, speech, time): reducing the proportion of strokes missed using the fast mnemonic. Stroke 2017;48:479481. 24. Govindarajan P, Ghilarducci D, McCulloch C, et al. Comparative evaluation of stroke triage algorithms for emergency medical dispatchers (MeDS): prospective cohort study protocol. BMC Neuro.l 2011;11:14. 25. Oostema JA, Carle T, Talia N, et al. Dispatcher stroke recognition using a stroke screening tool: a systematic review. Cerebrovasc Dis 2016;42:370-377.
777
26. Canty A, Ripley B. 2015). boot: Bootstrap R (S-Plus) Functions. R package version 1.3-17. 27. Davison AC, Hinkley DV. Bootstrap methods and their applications. Cambridge.: Cambridge University Press, 1997 ISBN 0-521-57391-2. 28. Team RC. (2013). R: A language and environment for statistical computing. 29. Liaw A, Wiener M. Classification and regression by random forest. R News 2002;2:18-22. 30. Ho DE, Imai I, King G. MatchIt: nonparametric preprocessing for parametric causal inference. Journal of Statistical Software. 2011;42:1-28. 31. Robin X, Turck N, Hainard A, et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 2011;12. 32. Prévost J. Déviation des yeux et de la tête dans quelques cas d’hémiplegie. Gazette Hebdomadaire de Medecine et de Chirurgie 1865;41:649-650.
