Use of deep artificial neural networks to identify stroke during triage via subtle changes in circulating cell counts.

BACKGROUND: The development of tools that could help emergency department clinicians recognize stroke during triage could reduce treatment delays and improve patient outcomes. Growing evidence suggests that stroke is associated with several changes in circulating cell counts. The aim of this study was to determine whether machine-learning can be used to identify stroke in the emergency department using data available from a routine complete blood count with differential. METHODS: Red blood cell, platelet, neutrophil, lymphocyte, monocyte, eosinophil, and basophil counts were assessed in admission blood samples collected from 160 stroke patients and 116 stroke mimics recruited from three geographically distinct clinical sites, and an ensemble artificial neural network model was developed and tested for its ability to discriminate between groups. RESULTS: Several modest but statistically significant differences were observed in cell counts between stroke patients and stroke mimics. The counts of no single cell population alone were adequate to discriminate between groups with high levels of accuracy; however, combined classification using the neural network model resulted in a dramatic and statistically significant improvement in diagnostic performance according to receiver-operating characteristic analysis. Furthermore, the neural network model displayed superior performance as a triage decision making tool compared to symptom-based tools such as the Cincinnati Prehospital Stroke Scale (CPSS) and the National Institutes of Health Stroke Scale (NIHSS) when assessed using decision curve analysis. CONCLUSIONS: Our results suggest that algorithmic analysis of commonly collected hematology data using machine-learning could potentially be used to help emergency department clinicians make better-informed triage decisions in situations where advanced imaging techniques or neurological expertise are not immediately available, or even to electronically flag patients in which stroke should be considered as a diagnosis as part of an automated stroke alert system.

Stroke Scales as Assessment Tools in Emergency Settings: A Narrative Review.

In the last 20 years, substantial improvements have been made in stroke recanalization treatment. Good outcomes after modern reperfusion treatment require the rapid and accurate identification of stroke patients. Several stroke rating scales are available or have been proposed for the early recognition of stroke and the evaluation of stroke severity and outcome. This review aims to provide an overview of commonly used stroke scales in emergency and clinical settings. The most commonly used scale in a prehospital setting for stroke recognition is the Face, Arms, Speech, Time (FAST) test. Among many prehospital stroke scales, the Los Angeles Prehospital Stroke Screen has the highest sensitivity and specificity for confirming stroke diagnosis. The National Institutes of Health Stroke Scale (NIHSS) is the most recommended tool for the evaluation of stroke patients in hospital settings and research, and it has two variants: the shortened NIHSS for Emergency Medical Service and the modified NIHSS. The evaluation of comatose patients usually involves assessment with the Glasgow Coma Scale, which is very useful in patients with hemorrhagic stroke or traumatic brain injury. In patients with subarachnoid hemorrhage, the outcome is usually accessed with the Hunt and Hess scale. A commonly used tool for stroke outcome evaluation in clinical/hospital settings and research is the modified Rankin scale. The tools for disability evaluation are the Barthel Index and Functional Independence Measure.

Detailed severity assessment of Cincinnati Prehospital Stroke Scale to detect large vessel occlusion in acute ischemic stroke.

BACKGROUND: Selecting stroke patients with large vessel occlusion (LVO) based on prehospital stroke scales could provide a faster triage and transportation to a comprehensive stroke centre resulting a favourable outcome. We aimed here to explore the detailed severity assessment of Cincinnati Prehospital Stroke Scale (CPSS) to improve its ability to detect LVO in acute ischemic stroke (AIS) patients. METHODS: A cross-sectional analysis was performed in a prospectively collected registry of consecutive patients with first ever AIS admitted within 6 h after symptom onset. On admission stroke severity was assessed using the National Institutes of Health Stroke Scale (NIHSS) and the presence of LVO was confirmed by computed tomography angiography (CTA) as an endpoint. A detailed version of CPSS (d-CPSS) was designed based on the severity assessment of CPSS items derived from NIHSS. The ability of this scale to confirm an LVO was compared to CPSS and NIHSS respectively. RESULTS: Using a ROC analysis, the AUC value of d-CPSS was significantly higher compared to the AUC value of CPSS itself (0.788 vs. 0.633, p < 0.001) and very similar to the AUC of NIHSS (0.795, p = 0.510). An optimal cut-off score was found as d-CPSS≥5 to discriminate the presence of LVO (sensitivity: 69.9%, specificity: 75.2%). CONCLUSION: A detailed severity assessment of CPSS items (upper extremity weakness, facial palsy and speech disturbance) could significantly increase the ability of CPSS to discriminate the presence of LVO in AIS patients.

Initial National Institute of Health Stroke Scale to Early Predict the Improvement of Swallowing in Patients with Acute Ischemic Stroke.

OBJECTIVES: To study the applicability of National Institutes of Health Stroke Scale (NIHSS) in early predicting the prognosis of poststroke dysphagia in an acute ward. METHODS: This is an observational retrospective cohort study including adult patients with ischemic stroke. Patients with various factors affecting swallowing were excluded to obtain a representative sample of 165 patients. The main outcome measure was the improvements of oral intake function. RESULTS: The scores of facial palsy (NIHSS item 4) (odds ratio [OR]: 0.484, 95% confidence interval [CI]: 0.279-0.838, P = .0096] and language/aphasia (NIHSS item 9) (OR: 0.562, 95% CI: 0.321-0.982, P = .0430) demonstrated significantly negative effects on the early improvement of dysphagia. Moreover, the improved patients had a 4.14-fold (95% CI: 2.53-11.23, P = .005) increased odds of returning home compared with nonimproved patients. CONCLUSIONS: Our findings provide evidence that early improvement of poststroke dysphagia was significantly associated with a favorable discharge destination and NIHSS items of facial palsy and language/aphasia can be used at the onset of stroke to identify dysphagic patients at risk of achieving limited improvement. These findings provide valuable prognostic indicators for clinicians to make a precise outcome prediction at very early stage.

DEFENSIVE Stroke Scale: Novel Diagnostic Tool for Predicting Posterior Circulation Infarction in the Emergency Department.

BACKGROUND: Dizziness is the most common posterior circulation symptom; however, diagnosing a posterior circulation infarction is difficult due to a lack of typical symptoms. We aimed to investigate the frequency of misdiagnosis of a posterior circulation infarction in patients who presented with dizziness and to develop a new stroke scale that increased the diagnostic accuracy for stroke among these subjects. METHODS: We retrospectively analyzed consecutive data from subjects hospitalized with ischemic stroke who presented with dizziness (the developmental phase). Based on these results, we created a novel stroke scale, which was used as a diagnostic procedure in the prospective validation phase. We compared the rate of misdiagnosis of ischemic stroke between phases. RESULTS: During the development phase, 115 subjects were hospitalized for ischemic stroke accompanied by dizziness. Six ischemic stroke subjects were not properly diagnosed (6/115, 5.2%). We created the new DisEquilibrium, Floating sEnsation, Non-Specific dizziness, Imbalance, and VErtigo (DEFENSIVE) stroke scale to prevent underdiagnosis of a posterior circulation infarction. During the validation phase, 949 subjects with dizziness were examined with the DEFENSIVE stroke scale; among these subjects, 100 were hospitalized for ischemic stroke accompanied by dizziness. No subject with ischemic stroke was overlooked. The new DEFENSIVE stroke scale had a sensitivity of 100% and decreased the rate of improper diagnosis of stroke (5.2% versus 0%; P = .022). CONCLUSIONS: Our new stroke recognition instrument for a posterior circulation infarction presenting with dizziness and related symptoms (the DEFENSIVE stroke scale) is easy to administer and has good diagnostic accuracy.

Derivation and Validation of the Emergency Medical Stroke Assessment and Comparison of Large Vessel Occlusion Scales.

BACKGROUND: This study aims to develop a simple scale to identify patients with prehospital stroke with large vessel occlusion (LVO), without losing sensitivity for other stroke types. METHODS: The Emergency Medical Stroke Assessment (EMSA) was derived from the National Institutes of Health Stroke Scale (NIHSS) items and validated for prediction of LVO in a separate cohort. We compared the EMSA with the 3-item stroke scale (3I-SS), Cincinnati Prehospital Stroke Severity Scale (C-STAT), Rapid Arterial oCclusion Evaluation (RACE) scale, and Field Assessment Stroke Triage for Emergency Destination (FAST-ED) for prediction of LVO and stroke. We surveyed paramedics to assess ease of use and interpretation of scales. RESULTS: The combination of gaze preference, facial asymmetry, asymmetrical arm and leg drift, and abnormal speech or language yielded the EMSA. An EMSA less than 3, 75% sensitivity, and 50% specificity significantly reduced the likelihood of LVO (LR- = .489, 95% confidence interval .366-0.637) versus 3I-SS less than 4 (.866, .798-0.926). A normal EMSA, 93% sensitivity, and 47% specificity significantly reduced the likelihood of stroke (LR- = .142, .068-0.299) versus 3I-SS (.476, .330-0.688) and C-STAT (.858, .717-1.028). EMSA was rated easy to perform by 72% (13 of 18) of paramedics versus 67% (12 of 18) for FAST-ED and 6% (1 of 18) for RACE (χ2 = 27.25, P < .0001), and easy to interpret by 94% (17 of 18) versus 56% (10 of 18) for FAST-ED and 11% (2 of 18) for RACE (χ2 = 21.13, P < .0001). CONCLUSIONS: The EMSA has superior abilities to identify LVO versus 3I-SS and stroke versus 3I-SS and C-STAT. The EMSA has similar ability to triage patients with stroke compared with the FAST-ED and RACE, but is simpler to perform and interpret.

Validating Existing Scales for Identification of Acute Stroke in an Inpatient Setting.

Background and Purpose: A significant proportion of strokes occur while patients are hospitalized for other reasons. Numerous stroke scales have been developed and validated for use in pre-hospital and emergency department settings, and there is growing interest to adapt these scales for use in the inpatient setting. We aimed to validate existing stroke scales for inpatient stroke codes. Methods: We retrospectively reviewed charts from inpatient stroke code activations at an urban academic medical center from January 2016 through December 2018. Receiver operating characteristic analysis was performed for each specified stroke scale including NIHSS, FAST, BE-FAST, 2CAN, FABS, TeleStroke Mimic, and LAMS. We also used logistic regression to identify independent predictors of stroke and to derive a novel scale. Results: Of the 958 stroke code activations reviewed, 151 (15.8%) had a final diagnosis of ischemic or hemorrhagic stroke. The area under the curve (AUC) of existing scales varied from .465 (FABS score) to .563 (2CAN score). Four risk factors independently predicted stroke: (1) recent cardiovascular procedure, (2) platelet count less than 50 × 109 per liter, (3) gaze deviation, and (4) presence of unilateral leg weakness. Combining these 4 factors into a new score yielded an AUC of .653 (95% confidence interval [CI] .604-.702). Conclusion: This study suggests that currently available stroke scales may not be sufficient to differentiate strokes from mimics in the inpatient setting. Our data suggest that novel approaches may be required to help with diagnosis in this unique population.

Optimization of Large Vessel Occlusion Detection in Acute Ischemic Stroke Using Machine Learning Methods.

The early detection of large-vessel occlusion (LVO) strokes is increasingly important as these patients are potential candidates for endovascular therapy, the availability of which is limited. Prehospital LVO detection scales mainly contain symptom variables only; however, recent studies revealed that other types of variables could be useful as well. Our aim was to comprehensively assess the predictive ability of several clinical variables for LVO prediction and to develop an optimal combination of them using machine learning tools. We have retrospectively analysed data from a prospectively collected multi-centre stroke registry. Data on 41 variables were collected and divided into four groups (baseline vital parameters/demographic data, medical history, laboratory values, and symptoms). Following the univariate analysis, the LASSO method was used for feature selection to select an optimal combination of variables, and various machine learning methods (random forest (RF), logistic regression (LR), elastic net method (ENM), and simple neural network (SNN)) were applied to optimize the performance of the model. A total of 526 patients were included. Several neurological symptoms were more common and more severe in the group of LVO patients. Atrial fibrillation (AF) was more common, and serum white blood cell (WBC) counts were higher in the LVO group, while systolic blood pressure (SBP) was lower among LVO patients. Using the LASSO method, nine variables were selected for modelling (six symptom variables, AF, chronic heart failure, and WBC count). When applying machine learning methods and 10-fold cross validation using the selected variables, all models proved to have an AUC between 0.736 (RF) and 0.775 (LR), similar to the performance of National Institutes of Health Stroke Scale (AUC: 0.790). Our study highlights that, although certain neurological symptoms have the best ability to predict an LVO, other variables (such as AF and CHF in medical history and white blood cell counts) should also be included in multivariate models to optimize their efficiency.

Acute Middle Cerebral Artery Occlusion Detection Using Mobile Non-Imaging Brain Perfusion Ultrasound-First Case.

Mobile brain perfusion ultrasound (BPU) is a novel non-imaging technique creating only hemispheric perfusion curves following ultrasound contrast injection and has been specifically designed for early prehospital large vessel occlusion (LVO) stroke identification. We report on the first patient investigated with the SONAS® system, a portable point-of-care ultrasound system for BPU. This patient was admitted into our stroke unit about 12 h following onset of a fluctuating motor aphasia, dysarthria and facial weakness resulting in an NIHSS of 3 to 8. Occlusion of the left middle cerebral artery occlusion was diagnosed by computed tomography angiography. BPU was performed in conjunction with injection of echo-contrast agent to generate hemispheric perfusion curves and in parallel, conventional color-coded sonography (TCCS) assessing MCAO. Both assessments confirmed the results of angiography. Emergency mechanical thrombectomy (MT) achieved complete recanalization (TICI 3) and post-interventional NIHSS of 2 the next day. Telephone follow-up after 2 years found the patient fully active in professional life. Point-of-care BPU is a non-invasive technique especially suitable for prehospital stroke diagnosis for LVO. BPU in conjunction with prehospital stroke scales may enable goal-directed stroke patient placement, i.e., directly to comprehensive stroke centers aiming for MT. Further results of the ongoing phase II study are needed to confirm this finding.

Sensitivity of prehospital stroke scales for different intracranial large vessel occlusion locations.

INTRODUCTION: Prehospital stroke scales have been proposed to identify stroke patients with a large vessel occlusion to allow direct transport to an intervention centre capable of endovascular treatment (EVT). It is unclear whether these scales are able to detect not only proximal, but also more distal treatable occlusions. Our aim was to assess the sensitivity of prehospital stroke scales for different EVT-eligible occlusion locations in the anterior circulation. PATIENTS AND METHODS: The MR CLEAN Registry is a prospective, observational study in all centres that perform EVT in the Netherlands. We included adult patients with an anterior circulation stroke treated between March 2014 and November 2017. We used National Institutes of Health Stroke Scale scores at admission to reconstruct previously published prehospital stroke scales. We compared the sensitivity of each scale for different occlusion locations. Occlusions were assessed with CT angiography by an imaging core laboratory blinded to clinical findings. RESULTS: We included 3021 patients for the analysis of 14 scales. All scales had the highest sensitivity to detect internal carotid artery terminus occlusions (ranging from 0.21 to 0.97) and lowest for occlusions of the M2 segment (0.08 to 0.84, p-values < 0.001).Discussion and conclusion: Although prehospital stroke scales are generally sensitive for proximal large vessel occlusions, they are less sensitive to detect more distal occlusions.

Validation of Stroke Network of Wisconsin Scale at Aurora Health Care System.

BACKGROUND: The Stroke Network of Wisconsin (SNOW) scale, previously called the Pomona scale, was developed to predict large-vessel occlusions (LVOs) in patients with acute ischemic stroke (AIS). The original study showed a high accuracy of this scale. We sought to externally validate the SNOW scale in an independent cohort. METHODS: We retrospectively reviewed and calculated the SNOW scale, the Vision Aphasia and Neglect Scale (VAN), the Cincinnati Prehospital Stroke Severity (CPSS), the Los Angeles Motor Scale (LAMS), and the Prehospital Acute Stroke Severity Scale (PASS) for all patients who were presented within 24 hours after onset at AHCS (14 hospitals) between January 2015 and December 2016. The predictive performance of all scales and several National Institute of Health Stroke Scale cutoffs (≥6) were determined and compared. LVO was defined by total occlusions involving the intracranial internal carotid artery, middle cerebral artery (MCA; M1), or basilar arteries. RESULTS: Among 2183 AIS patients, 1381 had vascular imaging and were included in the analysis. LVO was detected in 169 (12%). A positive SNOW scale had comparable accuracy to predict LVO and showed a sensitivity of 0.80, specificity of 0.76, the positive predictive value (PPV) of 0.31, and negative predictive value of 0.96 for the detection of LVO versus CPSS ≥ 2 of 0.64, 0.87, 0.41, and 0.95. A positive SNOW scale had higher accuracy than VAN, LAMS, and PASS. CONCLUSION: In our large stroke network cohort, the SNOW scale has promising sensitivity, specificity and accuracy to predict LVO. Future prospective studies in both prehospital and emergency room settings are warranted.

Prehospital Advanced Diagnostics and Treatment of Acute Stroke: Protocol for a Controlled Intervention Study.

BACKGROUND: Acute ischemic stroke (AIS) is a medical emergency. The outcome is closely linked to the time elapsing from symptom onset to treatment, and seemingly small delays can mean the difference between full recovery and physical and cognitive dysfunction. Recanalization to allow blood to reenter the affected area is most efficient immediately after symptoms occur, and intravenous thrombolysis must be initiated no later than 4.5 hours after the symptom onset. A liable diagnosis is mandatory to administer the appropriate treatment. Prehospital diagnosis and, in cases where contraindications are ruled out, prehospital initiation of intravenous thrombolysis have been shown to significantly decrease the time from alarm to the treatment. OBJECTIVE: The objective of this paper is to investigate the effectiveness of prehospital thrombolysis as measured by (1) time spent from symptom onset to treatment and (2) the number of patients treated within 4.5 hours. In addition, we want to conduct explorative studies. These will include (1) the use of biomarkers for diagnostic and prognostic use where we will collect blood samples from various time points, including the hyperacute phase and (2) the study of magnetic resonance imaging (MRI) images at day 1 to determine the infarct volume and if the time to thrombolysis has an influence on this. METHODS: This is a prospective controlled intervention study. The intervention will involve a computed tomography (CT) and thrombolysis in a physician-manned ambulance called a mobile stroke unit (MSU). The control will be the conventional pathway where the patient is transported to the hospital for CT, and thrombolysis as per current procedure. RESULTS: Patient inclusion has started and a total of 37 patients are enrolled (control and intervention combined). The estimated time to completed inclusion is 36 months, starting from May 2017. The results of this study will be analyzed and published at the end of the trial. CONCLUSIONS: This trial aims to document the feasibility of saving time for all stroke patients by providing prehospital diagnostics and treatment, as well as transport to appropriate level of care, in a safe environment provided by anesthesiologists trained in prehospital critical care. TRIAL REGISTRATION: ClinicalTrials.gov NCT03158259; https://clinicaltrials.gov/show/NCT03158259 (Archived by WebCite at http://www.webcitation.org/6wxNEUMUD).

Validation of simplified FABS scale to predict stroke mimics in a Chinese population undergoing intravenous thrombolysis.

OBJECTIVES: A large number of suspected stroke patients undergoing intravenous thrombolysis are stroke mimics (SMs). In this study, we sought to revise the FABS scale for screening and stratifying SMs from acute ischemic stroke (AIS) in a Chinese stroke population receiving fibrinolytic therapy. PATIENTS AND METHODS: The simplified FABS (sFABS) scale includes 4 items with 1 point for each item present: absence of facial droop, negative history of atrial fibrillation, age <50years, systolic blood pressure <150mm Hg at presentation. We evaluated consecutive suspected stroke patients undergoing intravenous thrombolysis in our stroke center for validation of sFABS scale. Diagnosis of SMs was based on absence of acute ischemic lesions on first and second diffusion weight imaging sequence in addition to an alternate diagnosis at discharge. RESULTS: A total of 190 AIS patients and 28 SMs were included in this study from December 2015 to February 2017. The sFABS scale showed excellent discrimination (C statistic: 0.928, 95% CI: 0.887-0.969, P<0.001). The Hosmer and Lemeshow goodness of fit test showed that the sFABS scale also had a good calibration (Cox and Snell R2=0.294, Nagelkerke R2=0.549). The plot of observed versus predicted risk of SMs showed high correlation (Pearson correlation coefficient: 0.983) between observed and predicted risk in our registered stroke population. CONCLUSION: The sFABS scale had excellent discrimination and good calibration abilities to predict SMs among a Chinese stroke population receiving tPA therapy. Further imaging evaluation may be necessary before the use of tPA if the sFABS score is higher.

Diagnosis of Acute Ischemic Stoke.

Acute ischemic stroke is a challenging and time-sensitive diagnosis. Diagnosis begins with rapid detection of acute stroke symptoms by the patient, their family or caregivers, or bystanders. If acute stroke is suspected, EMS providers should be called for rapid assessment. EMS providers will utilize prehospital stroke tools to diagnose and determine potential stroke severity. Once at the hospital, the stroke team works rapidly to solidify the patient history, perform a focused neurologic examination and obtain necessary laboratory tests and brain imaging to accurately diagnose acute ischemic stroke and properly treat the patient.

Derivation and Validation of a Modified Short Form of the Stroke Impact Scale.

BACKGROUND: The Stroke Impact Scale (SIS) is a stroke-specific, quality of life measure recommended for research and clinical practice. Completion rates are suboptimal and could relate to test burden. We derived and validated a short form SIS (SF-SIS). METHODS AND RESULTS: We examined data from the Virtual International Stroke Trial Archive, generating derivation and validation populations. We derived an SF-SIS by selecting 1 item per domain of SIS, choosing items most highly correlated with total domain score. Our validation described agreement of SF-SIS with original SIS and the SIS-16 and correlation with Barthel Index, modified Rankin Scale, National Institutes of Health Stroke Scale, and Euro-QoL 5 dimensions visual analog scales. We assessed discriminative validity (associations between SF-SIS and factors known to influence outcome [age, physiological parameters, and comorbidity]). We assessed face validity and acceptability by sharing the SF-SIS with a focus group of stroke survivors and multidisciplinary stroke healthcare staff. From 5549 acute study patients (mean age 68.5 [SD 13] years, mean SIS 64 [SD 32]) and 332 rehabilitation patients (mean age 65.7 [SD 11] years, mean SIS 61 [SD 11]), we derived an 8-item SF-SIS that demonstrated good agreement with original SIS and good correlation with our chosen functional and quality of life measures (all ρ>0.70, P<0.0001). Significant associations were seen with our chosen predictors of stroke outcome in the acute group (P<0.0001). The focus group agreed with the choice of items for SF-SIS across 7 of 8 domains. CONCLUSIONS: Using multiple, complementary methods, we have derived an SF-SIS and demonstrated content, convergent, and discriminant validity. This shortened SIS should allow collection of robust quality of life data with less associated test burden.