key: cord-0276768-4y3auoa3 authors: Gaude, E.; Nogueira, B.; Graham, S.; Smith, S.; Shaw, L.; Graziadio, S.; Ladreda, G.; Sloan, P.; Bernstock, J. D.; Shekhar, S.; Gropen, T. I.; Price, C. I. title: Combination of blood biomarkers and stroke scales improves identification of large vessel occlusions date: 2021-01-08 journal: nan DOI: 10.1101/2021.01.06.21249344 sha: f7e39d8eb6d305380c33c76b6db2910b6e5befab doc_id: 276768 cord_uid: 4y3auoa3 Background and Purpose. Acute ischemic stroke caused by large vessel occlusions (LVO) is a major contributor to stroke deaths and disabilities; however, identification for emergency treatment is challenging. Aims. To evaluate the diagnostic accuracy of a panel of biomarkers for LVO prediction. Methods. 170 patients with suspected stroke were recruited retrospectively at one hospital. We analysed the plasma levels of D-dimer, OPN, OPG, GFAP, vWF, and ADAMTS13 in LVO vs non-LVO. Diagnostic performance was estimated by using blood biomarkers alone or in combination with NIHSS-derived stroke severity scales. Results. Our patient cohort comprised 20% stroke mimics, 11% transient ischemic attack, 11% hemorrhagic stroke, 15% LVO ischemic stroke, 28% non-LVO ischemic stroke, and 15% ischemic stroke with unknown LVO status. Multivariable analysis found that the optimal set of blood biomarkers for LVO prediction was D-dimer (OR 15.4, 95% CI 4.9 to 57.6; p-value<0.001) and GFAP (OR 0.83, 95% CI 0.90 to 0.99; p-value=0.03). The combination of D-dimer and GFAP with stroke scales significantly improved LVO prediction, compared to the stroke scales alone (p-value<0.001). The combination of biomarkers with constructed FAST-ED or EMSA scales achieved an AUC of 95% (95% CI 91-100%) or 93% (CI 95% 89-97%), a sensitivity of 91% (95% CI 71-98%) or 86 (95% CI 66-97%), and a specificity of 95% (95% CI 89-98%) or 94% (95% CI 88-98%), for LVO prediction, respectively. Conclusions. The combination of D-dimer, GFAP, and stroke scales could provide a simple and highly accurate tool for identifying LVO patients. Stroke is among the leading causes of death and disability worldwide. Ischemic stroke caused by large vessel occlusion (LVO) contributes disproportionally to these figures and accounts for 62% of post-stroke disabilities and 96% of post-stroke mortality 1 . Acute ischemic stroke patients with LVO can be effectively treated with endovascular thrombectomy (EVT) 2 , but this treatment is only available in comprehensive stroke centers (CSC) or other EVT-capable centers. Inter-hospital transfer of LVO patients from primary stroke centers to EVT-capable centers significantly delays time to treatment and leads to higher disability rates 3 . Identification of LVO patients in the pre-hospital setting (e.g. within an ambulance) could enable the transfer of LVO patients to EVT-capable centers directly, reducing time to treatment, functional disabilities, and deaths. Multiple studies have investigated the ability of brief pre-hospital stroke assessment scales to identify LVO strokes in the field. Despite this, these severity scales lack the sensitivity and specificity required for triaging LVO patients with confidence, resulting in false negatives in patients with LVO and milder stroke as well as false positives in patients with stroke mimics or hemorrhagic strokes 4, 5 . A more accurate diagnostic test able to complement these assessment scores and direct LVO patients to EVT-capable centers is much needed. Several research studies have investigated the ability of blood biomarkers to differentiate stroke from non-stroke patients, or ischemic from hemorrhagic stroke patients 6 . In addition, a number of studies have been dedicated to blood biomarkers of stroke etiology 7 . Yet, fewer studies have investigated the association of blood biomarker levels with LVO strokes specifically (i.e. as a subtype of ischemic stroke) and comprehensively (i.e. as a combination of different etiologic subtypes) 8, 9 . Blood biomarkers for the identification of LVO strokes are yet to be identified and validated. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint The primary aim of this study was to identify the optimal combination of biomarkers that could differentiate between LVO and non-LVO. The screening panel of biomarkers comprised D-dimer, osteopontin (OPN) and osteoprotegerin (OPG), which have been associated with stroke etiologic subtypes 7,10 ; glial fibrillary acidic protein (GFAP), an astrocyte marker found increased in hemorrhagic stroke 6 ; von Willebrand factor (vWF), and a disintegrin and a metalloproteinase with a thrombospondin type I motif, member 13 (ADAMTS13), known markers of hemostasis that have been linked to ischemic stroke 11 . An additional analysis considered whether an optimal panel could increase the accuracy of stroke severity scales to differentiate between LVO and non-LVO stroke patients. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint This study was a retrospective observational study. Between August 2018 and February 2020, Cellular Pathology (CEPA) Biobank staff identified study eligible patients that had presented to the Emergency Department (ED) at the Royal Victoria Infirmary Hospital in Newcastle upon-Tyne (UK). Patients were identified based on the following criteria: 1) >18 years old; 2) evaluated in the ED for suspected stroke, as identified by ambulance paramedics, ED clinicians, and/or stroke specialist nurses; 3) <12 hours from their last known well or symptom onset time; 4) reperfusion therapy had not yet been received. As per clinical routine, whole venous blood was withdrawn at ED admission and stored at 4ºC (protected from light). Whole blood diagnostic remnants of eligible patients were identified by biobank staff and centrifuged at 2000 x g for 15 minutes at 4ºC and immediately frozen at -80ºC. Frozen samples were then transferred in one batch to Pockit Diagnostics Ltd for biomarkers measurement. Ethics approval for this study was part of CEPA Biobank's existing approval (NHS-HRA-North-East-Newcastle & North Tyneside 1 Research Ethics Committee, REC Reference: 17/NE/0070). All procedures were in accordance to institutional guidelines. The following routine clinical data were collected for eligible patients: demographics (age, gender), clinical characteristics (blood pressure, pulse, atrial fibrillation, hypertension), clinical laboratory results (hematology, biochemistry, blood lipids), admission NIHSS score, last known well or symptom onset time, blood withdrawal time, imaging findings within 1 hour, and final clinical diagnosis. Data were provided to Pockit diagnostics Ltd. . CC-BY-ND 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Routine clinical data collected as described above were used to assign patients to the following diagnostic categories: LVO, non-LVO ischemic stroke, hemorrhagic stroke, transient ischemic attack (TIA), stroke mimic condition. • Transient ischemic attack (TIA) and stroke mimic were assigned based on local specialist opinion • Hemorrhagic stroke was assigned based on the presence of hemorrhage on brain imaging • LVO required CT angiography confirmation by neuroradiologist report • Remaining patients where a stroke specialist assigned a diagnosis of ischemic stroke were categorized as either non-LVO or not classifiable as follows: o non-LVO if CTA had been undertaken and LVO was not present OR if a CTA had not been undertaken but the NIHSS score on admission was <5. The latter was a pragmatic threshold reflecting little likelihood of eligibility for thrombectomy treatment 12 . o not classifiable if CTA had not been undertaken and NIHSS score on admission was >4. Reference standard results were not available to the operators involved in blood biomarker measurements. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint Plasma biomarkers were measured using commercial enzyme-linked immunosorbent assays The proposed intended use of the blood biomarkers is to rule in patients with LVO to redirect patients to EVT-capable centres from a population of suspected stroke, thus we powered the study on specificity. Assuming a prevalence of 15%, a minimum specificity of 90%, a twotailed 5% type I error rate (α) and 90% power (β), a sample size of 161 suspected patients with 25 LVO would detect a specificity of 97% (95%CI: 90%-100%) for LVO. Symptom onset to blood collection time (OBT) was calculated from last known well/onset time and time of blood withdrawal. Where patients were reported to have woken up with symptoms, the time since last known well was assumed to be midnight of the previous day. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint and: i) for normally distributed variables, Welch's t-test and mean ± standard deviation (SD) were used or ii) for non-normally distributed variables Wilcoxon-Mann-Whitney U test, median and interquartile range (IQR) were used. Categorical variables were assessed by Pearson's chi-square test. To compare variables between subtypes of suspected stroke, ANOVA was used. When >10 variables were tested at the same time, multiple hypothesis correction was performed using Benjamini-Hochberg's method. Analysis of variance was used to assess the interaction between all blood biomarkers and subtypes of suspected stroke. If overall significance was confirmed, pairwise comparisons were performed with Tukey's test. To identify the optimal panel of blood biomarkers for LVO prediction, we used a multivariate logistic regression with Diagnosis (LVO vs non-LVO) as outcome variable and plasma levels of D-dimer, GFAP, OPN, OPG, vWF, and ADAMTS13 as exploratory variables. Bidirectional stepwise elimination based on Akaike information criterion (AIC) levels was used for model selection. Linearity between predictors and outcome measure was assessed through logarithmic and quadratic transformation. Transformations were selected based on the AIC. To investigate whether the addition of blood biomarkers improved the accuracy of stroke severity scales for LVO identification, we used a second multivariate logistic with Diagnosis as the outcome variable and the optimal panel and one of the stroke severity scales (FAST, FAST-ED, RACE, C-STAT, or EMSA) as exploratory variables. We used this approach because the scales were highly correlated, and, since a comparison of different severity scales was outside the scope of this work, this approach reduced the level of collinearity in the model. To assess the goodness of fit of the blood biomarker panel and the stroke scales, the likelihood ratio test (LR) and AIC were used. The area under the receiver operating is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint characteristic curve (AUC) with 95% CIs was used as a measure of discrimination. At selected cut-off points sensitivity, specificity, positive likelihood ratio (LR+), and negative likelihood ratio (LR-) were also evaluated. For each model, the cut-off point was selected by maximising specificity for LVO prediction while maintaining a minimum specificity level of 90%, in line with our power analysis. Correction for optimistic predictions was performed through bootstrapping with 2000 resamples, and presented with confidence intervals (CI). All analyses were performed with R version 3.6.2 with the help of RStudio version 1.2.5033 by using the packages nnet, ROCR, caret, tidyverse, oddsratio, lmtest, and OptimalCutpoints. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Stroke mimic diagnoses were as follows: anaemia (3%), anxiety (10%), Bell's palsy (7%), delirium (3%), dementia (3%), depression (3%), dysphasia (3%), metastatic cancer (3%), migraine (16%), seizure (13%), syncope (13%), vertigo (13%), and undetermined (10%). Clinical characteristics for LVO and all non-LVO patients are reported in Table 1 . We found significant differences between LVO and all non-LVO in NIHSS score (18±9 and 3±5, p-value<0.001), presence of atrial fibrillation (52% and 10%, p-value<0.001), and systolic blood pressure (140±22 and 157±29 mmHg, p-value=0.03). Of note, age, gender, OBT, or is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint OPN, and GFAP were significantly associated with specific patient subgroups (Supplementary Table 2 ). Levels of D-dimer, OPN, OPG, and ADAMTS13 were log transformed, while levels of vWF underwent quadratic transformation; no transformation was applied to GFAP levels. Potentially useful predictors with univariate analysis for LVO accuracy were D-dimer (OR 1.13, 95% CI 1.07-1.20; p-value<0.001), and OPG (OR 1.10, 95% CI 1.01 to 1.22; p-value=0.04; Table 2 ). The optimal set of blood biomarkers for LVO prediction identified by the logistic regression was D-dimer (OR 15.4, 95% CI 4.9-57.6; p-value<0.0011) and GFAP (OR 0.83, 95% CI 0.50-0.99; p-value=0.03; Table 3 ). The AUC of the model with D-dimer and GFAP for LVO prediction was 81% (95% CI 74-88%; Table 4 ). The cut-off that maximized specificity was 0.33 and gave a sensitivity of 56% (95% CI 34-77%) at a specificity of 92% (95% CI 85-96%). LR+ was 7 (95% CI 3-15), and LR-0.47 (95% CI 0.3-0.76; Table 5 ). We report stroke severity scales in LVO and non-LVO patients in Supplementary Table 3 . The AUCs obtained from the univariate logistic regressions of each stroke scale for LVO discrimination were (Table 4) : C-STAT 79% (95% CI 72-86%), EMSA 84% (95% CI 79-89%), FAST 83% (95% CI 78-88%), FAST-ED 91% (95% CI 86-95%), RACE 87% (95% CI 82-93%). The addition of the optimal biomarker panel previously identified (D-dimer and GFAP) to the stroke severity models resulted in models with a lower AIC, a higher AUC and a significant LR test (Table 4 and Figure 1 ). In line with this, D-dimer and GFAP substantially improved sensitivity without adversely affecting specificity of clinical scales, such that all achieved high performance for LVO prediction (Table 5) ; of note, addition of D-dimer and GFAP to is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint FAST-ED or EMSA resulted in the highest LR+ of 23 (95% CI 9-60) or 17 (95% CI 7-41), LR-of 0.09 (95% CI 0.02-0.34) or 0.14 (95% CI 0.05-0.39), sensitivity of 91% (95% CI 72-99%) or 87 (95% CI 66-97%), and specificity of 96% (95% CI 90-99%) or 95% (95% CI 89-98%), respectively. Overall, the addition of D-dimer and GFAP to stroke scales allowed to increase the number of true positives, while reducing the number of false positives due to hemorrhage or mimics (Supplementary Table 4 ), depending on the scale used. This study demonstrates that a biomarker panel composed of D-dimer and GFAP could provide a valuable tool for highly specific identification of stroke patients with LVO. Moreover, we have shown that combining D-dimer and GFAP with pre-hospital stroke scales can result in high predictive power (AUC=95%) for LVO in a hospital population of patients with suspected stroke. D-dimer has been previously associated with cardioembolic strokes 17,18 , but few reports have investigated its usefulness for LVO identification. Occlusions in the main stem of intracranial arteries or in the cervical arteries have been linked to increased levels of plasma D-dimer 18 . In a separate study, cut-off values of D-dimer at 445 ng/mL or 300 ng/mL (depending on detection technique) distinguished large artery from small artery strokes 17 . Nevertheless, relevant limitations of these studies are the lack of CTA results for confirmation of LVO and absence of stroke mimics within the study population. While previously published studies focussed on comparing D-dimer levels between subtypes of ischemic stroke or between ischemic stroke and control subjects, our study is the first to assess D-dimer levels in LVO patients compared to a heterogeneous group of suspected stroke patients encompassing hemorrhagic stroke, non-LVO ischemic stroke, stroke mimics, and TIAs. The population recruited in our study more accurately reflects the population that is present in an ambulance setting or in the ED. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Notably, prediction of cardioembolic stroke was poorer when BNP or D-dimer were used in isolation, while diagnostic performance increased when both biomarkers were combined to categorize stroke subtypes 7 . In line with this evidence, we showed that the combination of multiple biomarkers into a panel permits higher diagnostic performance for LVO identification, compared to the use of individual biomarkers. We estimated the diagnostic performance of validated stroke severity scales for LVO prediction and observed that their overall accuracy was higher, compared to previous studies [13] [14] [15] [16] 21, 22 . This could be due to derivation of stroke scales from the NIHSS score, which was taken by ED clinicians. This may be particularly relevant for the FAST-ED and RACE which were found to be more complex compared to the FAST or EMSA 16 . Other studies have showed that the combination of clinical variables with blood biomarkers can improve stroke diagnosis. Brouns and colleagues reported that the addition of the Oxford Community Stroke Project (OCSP) classification to D-dimer measurements could improve is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint to the biomarkers alone 7 . In line with this evidence, here we demonstrate that a model built on D-dimer and GFAP, in conjunction with pre-hospital stroke scales, can lead to higher predictive ability for LVO, compared to the use of blood biomarkers, or stroke scales, in isolation. Each of the NIHSS-derived scales had advantages and disadvantages; the combination of the biomarkers with FAST-ED resulted in the highest diagnostic performance for LVO overall. Nevertheless, the collection of FAST-ED is more complex compared to other stroke scales, such as FAST or EMSA 16 , which may be preferred for field applications. We showed that the combination of D-dimer and GFAP with FAST or EMSA achieved sufficient LVO prediction (AUC=93%), assuming these results are replicated in further studies. The combination of D-dimer and GFAP with FAST or EMSA may be easier to replicate in the prehospital setting. High levels of specificity (or positive predictive value) would be required to identify LVO patients in the field, in order to deviate the ambulance journey from transportation to the nearest stroke centre, towards the nearest EVT-capable centre 4 . Our findings indicate that combining our biomarkers with any of a number of pre-hospital stroke scales has the potential to provide the level of diagnostic performance needed to triage LVO patients with reasonable confidence. Our study had several practical limitations. We used blood samples derived from routine clinical tests that were stored in the dark at +4º before processing. Although previous studies have shown stability of some of the tested blood biomarkers up to 240h at +4ºC 23,24 , we cannot exclude that some of our negative results could be due to protein degradation. Biomarker measurement was performed with standard laboratory immunoassays which are . CC-BY-ND 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint inherently variable and therefore would need comparison and validation in a future study. Symptom onset times were obtained from medical records by non specialists and there may be some inaccuracy but it is likely that patients were within 12 hours of onset. Categorisation into LVO and non-LVO patients was based on routinely available clinical and imaging findings, including a pragmatic decision for handling ischemic stroke patients where CTA was not available. This pragmatic method may have led to loss of LVO cases from the cohort (excluded as no CTA) or incorrect assignment of mild stroke as non-LVO (ischemic stroke with NIHSS < 5 assigned as non-LVO). A further study should include CTA for all ischemic stroke patients and independent adjudication of LVO. Stroke scales were derived based on hospital-completed NIHSS scores, which may not reflect the results that would be obtained by paramedics completion of these scales in the field.. We recruited less patients than the sample size calculation which may have reduced the confidence in estimates of diagnostic performance for LVO identification. In conclusion, our study suggests that combining the measurement of plasma D-dimer and GFAP with stroke severity scales may offer a valuable tool for early identification of LVO stroke patients. These findings need to be validated in a prospective study and if the tool is to be used in the pre-hospital setting, a point-of-care device is required. Evaluation of costeffectiveness and clinical utility of the tool would also be needed before introduction into clinical practice. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint The data that support the findings of this study, as well as the study protocol, are available from the corresponding author upon reasonable request. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.06.21249344 doi: medRxiv preprint Ischemic Strokes Due to Large-Vessel Occlusions Contribute Disproportionately to Stroke-Related Dependence and Death : A Review Endovascular thrombectomy after largevessel ischaemic stroke: A meta-analysis of individual patient data from five randomised trials Mothership versus drip-and-ship model for mechanical thrombectomy in acute stroke: A systematic review and meta-analysis for clinical and radiological outcomes Accuracy of Prediction Instruments for Diagnosing Large Vessel Occlusion in Individuals With Suspected Stroke: A Systematic Review for the 2018 Guidelines for the Early Management of Patients With Acute Ischemic Stroke Clinical Scales Do Not Reliably Identify Acute Ischemic Stroke Patients with Large-Artery Occlusion. Stroke. Epub ahead of print Blood protein biomarkers as diagnostic tool for ischemic stroke: a systematic review Etiologic diagnosis of ischemic stroke subtypes with plasma biomarkers Cardiac Biomarkers Predict Large Vessel Plasma thrombin-cleaved osteopontin as a potential biomarker of acute atherothrombotic ischemic stroke Assessment of the Diagnostic Value of Plasma Levels, Activities, and Their Ratios of von Willebrand Factor and ADAMTS13 in Patients with Cerebral Infarction Clinical Selection Strategies to Identify Ischemic Stroke Patients with Large Anterior Vessel Occlusion: Results from SITS-ISTR (Safe Implementation of Thrombolysis in Stroke International Stroke Thrombolysis Registry) Field Assessment Stroke Triage for Emergency Destination Design and Validation of a Prehospital Stroke Scale to Predict Large Arterial Occlusion Design and Validation of a Prehospital Scale to Predict Stroke Severity: Cincinnati Prehospital Stroke Severity Scale Derivation and Validation of the Emergency Medical Stroke Assessment and Comparison of Large Vessel Occlusion Scales Plasmatic retinol-binding protein 4 and glial fibrillary acidic protein as biomarkers to differentiate ischemic stroke and intracerebral hemorrhage Serum glial fibrillary acidic protein as a biomarker for intracerebral haemorrhage in patients with acute stroke ARTICLE IN PRESS Design , Application and In fi eld Validation of a Pre-Hospital Emergent Large Vessel Occlusion Screening Tool : Ventura Emergent Large Vessel Occlusion Score Shepherd Centrally Guided Identification of Patients With Large Vessel Occlusion: Lessons From Trauma Systems Evaluation of The Effect of Storage Temperature on D-dimer Stability, Using Two Different Techniques Stability of Von Willebrand Factor in Blood Stored at 4 o C corresponding model with the addition of biomarkers.