Identification of embolic stroke in patients with large vessel occlusion: The Chinese embolic stroke score, CHESS

Abstract Aims The aim of the study was to develop a simple and objective score using clinical variables and quantified perfusion measures to identify embolic stroke with large vessel occlusions. Methods Eligible patients from five centers participating in the International Stroke Perfusion Imaging Registry were included in this study. Patients were split into a derivation cohort (n = 213) and a validation cohort (n = 116). A score was developed according to the coefficients of independent predictors of embolic stroke from stepwise logistic regression model in the derivation cohort. The performance of the score was validated by assessing its discrimination and calibration. Results The independent predictors of embolic stroke made up the Chinese Embolic Stroke Score (CHESS). There were: history of atrial fibrillation (3 points), non‐hypertension history (2 points), and delay time>6 s volume/delay time>3 s volume on perfusion imaging ≥0.23 (2 points). The AUC of CHESS in the derivation cohort and validation cohort were 0.87 and 0.79, respectively. Patients with a CHESS of 0 could be identified as low‐risk of embolic stroke, with a CHESS of 2–4 could be identified as medium‐risk and with a CHESS of 5–7 could be regarded as high‐risk. The observed rate of embolic stroke of each risk group was well‐calibrated with the predicted rate. Conclusion CHESS could reliably and independently identify embolic stroke as the cause of large vessel occlusion.


| INTRODUC TI ON
Endovascular treatment has been globally acknowledged as a standardof-care for acute ischemic stroke patients with large vessel occlusion (LVO), and prompt reperfusion is a key to achieving a satisfactory clinical outcome. However, Chinese LVO patients have been faced with the challenge of a longer reperfusion time compared with the Western population. [1][2][3][4][5] One of the main reasons may relate to the high prevalence of in-situ thrombosis due to intracranial atherosclerotic disease (ICAD) in Asian patients with acute LVO, 6 which is refractory to the current stent retrievers or aspiration systems and requires rescue therapy, like angioplasty and stenting. 7 Additionally, the mechanism of embolic stroke can also be very complicated due to cardiogenic embolism, artery to artery embolism from large vessel atherosclerosis and embolic stroke of undetermined source, which requires different recanalization techniques.
Thus, precisely identifying the etiology of the occlusion (embolic or nonembolic) at hyperacute stage before reperfusion therapy is crucial to the choice of the optimal recanalization strategy.
Several imaging markers have been developed to identify stroke etiology using thrombus location and clot formation. [8][9][10][11][12][13][14][15] Although they have a good predictive ability, the accuracy of these imaging markers depends on readers' experience, confining its application under emergent settings. Another marker of embolic versus ICAD as the cause of LVO is that ICAD patients have a better collateral flow, 16 which could quantitively assessed by perfusion imaging. 17,18 We aimed to develop and validate a concise and objective score combining baseline clinical data and a quantified measure of collateral flow using perfusion imaging to distinguish between embolic and non-embolic stroke preceding endovascular treatment. Additionally, though history of atrial fibrillation (AF) is a strong predictor of embolic stroke, it may not be easily known under emergent conditions. Moreover, acute LVO due to in-situ thrombosis may also occur in patients with AF. Therefore, the performance of the score that excluded history of AF was also assessed in the derivation and validation cohort.

| ME THODS
This study adheres to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement. An embolic stroke was determined using digital subtraction angiography imaging (DSA). Patients with a DSA imaging where there was no residual severe stenosis (>50% stenosis of the vessel caliber, using the diameter of the adjacent contact segment as the reference) after thrombectomy (without angioplasty or stenting) in the culprit occluded artery was considered embolic. DSA imaging was centrally and retrospectively analyzed by Dr. Lin and Dr. Hong blind to the CT Perfusion parameters and clinical information. Patients who underwent direct angioplasty were classified as non-embolic stroke.

| Patient selection and definition of embolic stroke
When discrepancy occurred between the two investigators, a third investigator (Dr. Cheng) was assigned to determine the cause of this index LVO.

| Acute multimodal imaging protocol and perfusion imaging analysis
Patients from these five Chinese sites were scanned using 64-, 256-, or 320-slice detector scanners for non-contrast CT, CT Angiography, and CT Perfusion. Details of different CT scanners are provided in Table S1.
All perfusion images were centrally post-analyzed using the commercial software MIStar (Apollo Medical Imaging Technology) with single value deconvolution with delay and dispersion correction.    was considered significant in the univariate analysis within the derivation cohort in order to acquire potential predictors. And in other steps of statistical analysis, a two-tailed p < 0.05 was considered significant.
The study cohort was divided into derivation and validation cohorts for analysis (65% and 35% respectively) using stratified sampling according to the rate of embolic stroke in the whole cohort using the "stratarand" command in Stata.

| Model Derivation
The model was then derived through the following steps: (1) Each statistically significant continuous variable was dichotomized using an optimal cut-point, which maximized the Youden index.
(2) Dichotomized variables were entered into the backward logistic regression model with a stepwise removal probability of p < 0.05 to select independent predictors of embolic stoke. (3) The coefficients of independent predictors of embolic stroke were rounded to the nearest integer to generate an integerbased scoring system: CHinese Embolic Stroke Score (CHESS).
The overall score of each patient was calculated as the sum of the variables' weighted scores if the patient presented any of the selected variables.

| Model Validation
The discrimination of the model was tested using the area under the receiver operating characteristic curve (AUC-ROC) in derivation and validation cohort. The 95% confidence intervals (95%CI) of AUC-ROCs were also provided. Calibration was evaluated using Hosmer-Lemeshow goodness-of-fit test and was plotted through calibration plots. The AUCs of CHESS in patients with large vessel occlusions of anterior circulation and posterior circulation were also compared. A two-tailed p < 0.05 was considered significant.

| Derivation and internal validation of the score
In the derivation cohort, compared with patients with non-embolic stroke, patients with embolic stroke were older, had lower baseline diastolic blood pressure, higher baseline NIHSS, a more severe perfusion imaging profile, higher rate of AF history and were more frequently female. Fewer patients with embolic stroke had a history of smoking or hypertension (    Figure 3A and Figure 4A).   Figure 3B and Figure 4B).

| Performance of CHESS with AF history excluded
When AF history was excluded, CHESS could also reliably pre-

| DISCUSS ION
We developed an integer-based score, CHESS, to identify embolic could be regarded as high-risk.
Since endovascular treatment has become a routine practice for acute ischemic stroke patients with LVO and the treatment window has been extended, 23 many clinical, imaging, and biochemical markers have been used to assess stroke severity, identify stroke etiology, optimize treatment effect and predict outcomes. [24][25][26] To our knowledge, this is the first model combining objective clinical and imaging factors to predict embolic stroke with LVO prior to thrombectomy.
Previous studies have proposed several imaging markers to predict embolic stroke, with the requirement of acute MRI and/or the need for highly experienced imaging readers. [8][9][10][11][12][13][14][15] Scores to predict presence of AF, like Score for the Targeting of Atrial Fibrillation (STAF) 27 and LADS, 28 may also be useful to detect cardiogenic embolism.
However In our derivation cohort, patients with embolic stroke were older, more frequently female, had lower baseline blood pressure and a more severely impaired collateral flow compared with non-embolic stroke, which was in accordance with the previous studies. 8,16,29 As for stroke severity, similar with other studies, 29 a more severe clinical manifestation (higher baseline NIHSS) was found in embolic stroke patients compared with the non-embolic ones. History of hypertension has also been recognized as a significant risk factor of ICADrelated occlusions in a meta-analysis of 1967 patients. 30  Additionally, CHESS is derived and validated from a Chinese-only population. Therefore, for patients in an extended time window (6-24 h) and with a younger age, and patients of other ethnic groups, CHESS should be applied with caution. Lastly, CHESS requires realtime advanced post-processing software of perfusion images to quantify collateral flow, which is not available in every stroke center and is not guideline-recommended for patients within 6 h of stroke onset. Therefore, a score that uses magnetic resonance angiography (MRA)/CTA to semi-quantify the level of collateral flow could probably be more practical. To be addressed, all the five centers of this research joined the INSPIRE study and acute multimodal CT scan has become a routine process after patients' arrival at the ER, so no selection bias was considered to exist during patient selection.
Since the concept of "tissue window" may replace the traditional "time window," CHESS puts forward another possible application of perfusion CT in the individualized treatment of acute stroke.

| CON CLUS IONS
CHESS, consists of history of AF, history of hypertension and quantified acute cerebral collateral flow, can reliably predict acute embolic stroke with LVO to assist reperfusion strategy in endovascular treatment.

AUTH O R S' CO NTR I B UTI O N S
All author participated in data collection, critical review and revision of this manuscript. XC, MP, QD designed the study. LH, LL analyzed the data. LH drafted this manuscript, prepared tables and figures.

D I SCLOS U R E
The authors have nothing to disclose.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from Dr. Xin Cheng (chengxin@fudan.edu.cn) upon reasonable request.