Risk factors for acute ischemic stroke caused by anterior large vessel occlusion

Abstract

Background and Purpose

Accurate prediction of acute ischemic stroke (AIS) caused by anterior large vessel occlusion (LVO) that is amendable to mechanical thrombectomy remains a challenge. We developed and validated a prediction model for anterior circulation LVO stroke using past medical history elements present on admission and neurologic examination.

Methods

We retrospectively reviewed AIS patients admitted between 2009 and 2017 to three hospitals within a large healthcare system in the United States. Patients with occlusions of the internal carotid artery or M1 or M2 segments of the middle cerebral artery were randomly split into 2/3 derivation and 1/3 validation cohorts for development of an anterior circulation LVO prediction model and score that was further curtailed for potential use in the prehospital setting.

Results

A total of 1,654 AIS were reviewed, including 248 (15%) with anterior circulation LVO AIS. In the derivation cohort, National Institute of Health Stroke Score Scale (NIHSS) at the time of cerebrovascular imaging, current smoking status, type 2 diabetes mellitus, extracranial carotid, and intracranial atherosclerotic stenosis were significantly associated with anterior circulation LVO stroke. The prehospital score was curtailed to NIHSS, current smoking status, and type 2 diabetes mellitus. The area under the curve for the prediction model, prehospital score, and NIHSS alone were 0.796, 0.757, and 0.725 for the derivation cohort, and 0.770, 0.689, and 0.665 for the validation cohort, respectively. The Youden index J was 0.46 for a score of > 6 with 84.7% sensitivity and 62.0% specificity.

Conclusions

Previously reported LVO stroke prediction scores focus solely on elements of the neurologic exam. In addition to stroke severity, smoking, diabetes mellitus, extracranial carotid, and intracranial atherosclerotic stenosis were associated with anterior circulation LVO AIS. While atherosclerotic stenosis may not be known until imaging is obtained, smoking and diabetes mellitus history can be readily obtained in the field and represent important elements of the prehospital score supplementing NIHSS.

Introduction

Mechanical thrombectomy provides significantly better functional outcomes than the best medical therapy in acute ischemic stroke (AIS) caused by large vessel occlusion (LVO) of the anterior circulation.^1–5 Early recognition of LVO is critical to timely allocate patients to comprehensive stroke centers and offer this treatment.⁶ Commonly, prehospital scores contain several NIHSS items that are used by emergency medical service providers to quickly identify patients at risk for LVO.^7,8 Exploration of additional variables is required to further improve the prediction of LVO, particularly in the prehospital setting. The purpose of this study was to develop and validate a prediction model for anterior circulation LVO stroke consisting of past medical history elements present on admission and neurologic examination.

Materials and Methods

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Study design

Acute ischemic stroke patients admitted between October 2009 and December 2017 at three hospitals within a large healthcare system in the northeastern United States were retrospectively reviewed. We focused on patients identified through the American Heart Association “Get With The Guidelines® Stroke” center initiative⁹ who had also been enrolled in a healthcare system-wide bio-banking exome sequencing program aimed at linking genetic samples and electronic health records data as part of a precision health initiative. The database was created to provide a framework for genetic investigation of diseases including stroke. The present study employed a subgroup analysis of this database. The institutional phenomic analytics and clinical data core performed electronic health record (EHR) data extraction from the clinical documentation improvement specialist (CDIS) and various disparate data sources, followed by subsequent manual chart review (N.S., M.D.A., S.K.) to verify the accuracy of diagnosis of AIS and document subsequent clinical management. Recruitment for the exome sequencing program was random and unrelated to the stroke diagnosis. Still, to assess for selection bias, patient demographics and cardio- and cerebrovascular risk factors were also obtained for all acute ischemic strokes admitted during the study period. Ethics approval was obtained from the local institutional review board and informed consent was waived.

Patient characteristics, clinical variables and outcome measures

Diagnosis of AIS was based on the neurologic examination and corresponding neuroimaging. Ischemic strokes were classified according to published TOAST criteria¹⁰ by individual chart review. The NIHSS was obtained at the time of cerebrovascular imaging. The diagnosis of LVO was based on non-invasive imaging (CT or MR angiography of head and neck) and/or digital subtraction angiography (DSA). There was no upper time limit from stroke onset to cerebrovascular imaging. Proximal anterior circulation LVO was defined as occlusion of the internal carotid artery (ICA) or the M1 and/or M2 segments of the middle cerebral artery (MCA), as those have been extensively studied in the context of mechanical thrombectomy.^1–5 Anterior cerebral artery LVO, albeit in the anterior circulation, was not included in the anterior circulation LVO prediction model since mechanical thrombectomy is less established in this territory. A number of clinical and imaging variables characterizing the AIS were collected (blinding during collection was not possible). Extracranial carotid or intracranial atherosclerotic stenosis was defined as narrowing of the lumen of 50% or greater on carotid ultrasound, CT or MR angiography, or digital subtraction angiography. Intracranial atherosclerosis present in any cerebral vessel was included. Anemia was defined as hemoglobin levels below 12 g/dl in women and 13 g/dl in men. Functional outcome was assessed using the modified Rankin Scale (mRS) with mRS 0 – 2 representing a favorable and mRS 3 – 6 representing an unfavorable functional outcome (mRS 6 death).

Derivation and validation cohorts

For development of an anterior circulation LVO prediction model, only LVOs of the ICA, M1, or M2 segments of the MCA were included. Next, the cohort was randomly split into 2/3 derivation and 1/3 validation cohorts. From the derivation cohort, a prediction score for anterior circulation LVO was obtained using a multivariable binary logistic regression model. The ß coefficients were used to create proportional scores. Consequently, this score was tested in the validation cohort.

Statistical analysis

Continuous variables are presented as mean ± standard deviation and categorical variables are presented as frequency and percent. Univariable analyses were carried out using binary logistic regression, Chi-square and Fisher’s exact tests, as appropriate. Post hoc testing for crosstabs exceeding 2×2 dimension was performed by calculating adjusted standardized residuals (z-scores) and associated p-values. Significance was evaluated after Bonferroni correction. Multivariable analysis was performed by integrating variables with a possible association (p < 0.15) with LVO. Stepwise backward elimination was performed and p-values of < 0.05 were considered statistically significant. Collinearity diagnostics were performed assessing correlation coefficients and variance inflation factors. Discrimination of predictive models was performed using the area under the receiver operating characteristic curve (AUC). Analysis of AUC, sensitivity/specificity criteria were calculated with MedCalc. Calibration of scores was tested using Hosmer and Lemeshow test. Comparison of receiver operating characteristic curves was performed using DeLong test. Statistical analyses, creation of figures and charts were performed using IBM SPSS version 22, MedCalc, MS Excel, and Adobe Photoshop CS5.

Results

Patient characteristics

Between October 1, 2009 and December 31, 2017, a total of 7,503 acute ischemic strokes were admitted. Gender and cardio- and cerebrovascular risk factors such as prior TIA or stroke, coronary artery disease, and COPD were equally frequent between exome-sequenced and non-sequenced cohorts (for more details please see Table I and Figure I in supplementary materials at http://stroke.ahajournals.org). Of those, 1,654 strokes were enrolled in the exome sequencing program and included in this analysis. The cohort was comprised of Caucasian patients only. Mean age of patients at the time of stroke was 67.7 ± 13.5 years. According to TOAST classification, stroke was attributed to cardioembolism (30.5%), large-artery atherosclerosis (24.1%), small vessel occlusion (16.0%), stroke of undetermined etiology (25.9%), and stroke of other determined etiology (3.5%). In patients with strokes classified as cardioembolic, an ECG and echocardiography were obtained in 99.0% and 80.6% of the cases, respectively. Median NIHSS was 3; the majority of strokes (78.9%) had scores below 7 (for more details please see Table II in supplementary materials at http://stroke.ahajournals.org). Large vessel occlusion was detected in 362 strokes (21.9%). For development of an anterior LVO prediction model, 114 (6.9%) LVOs not located at the ICA, M1 or M2 MCA segments were excluded, resulting in 248 (15.0%) anterior circulation LVOs (Table 1). LVO was diagnosed using CT angiography (CTA), MR angiography (MRA), and digital subtraction angiography (DSA) in 74.6%, 27.7%, and 22.6% of cases, respectively. Some patients received more than one cerebrovascular imaging modality: CTA only in 73.3%, CTA and DSA in 18.1%, CTA and MRA in 5.1%, MRA only in 2.8%, DSA only 0.6%, MRA and DSA 0.6%, CTA and MRA and DSA 0.6%. Median time from stroke onset to cerebrovascular imaging was 5.5 hours (range 0 – 180.5 hours). Intravenous tPA was administered in 158 strokes (9.6%), equally frequent in LVO and non-LVO strokes. Mechanical thrombectomy was performed in 40 (2.4%) strokes with 31 (77.5%) performed since 2015. Successful recanalization (mTICI 2B and 3) was achieved in 32 (80%) of strokes. Functional outcome at discharge was favorable (mRS 0 – 2) in 661/1644 strokes (40.2%). In-hospital mortality was 2.1% (35/1644). At 90 days follow-up, 1074/1601 (67.1%) had a favorable functional outcome, whereas a total of 99/1601 (6.2%) were deceased (for more details please see Table III in supplementary materials at http://stroke.ahajournals.org).

Table 1.

Sites of large vessel occlusion (N = 362)

Anterior circulation
Internal carotid artery	112 (30.9%)
Middle cerebral artery
M1	77 (21.3%)
M2	59 (16.3%)
M3	4 (1.1%)
Anterior cerebral artery
A1	4 (1.1%)
A2	3 (0.8%)
A3	1 (0.3%)
Posterior circulation
Vertebral artery	54 (14.9%)
Basilar artery	10 (2.8%)
Posterior cerebral artery
P1	23 (6.4%)
P2	15 (4.1%)

Prediction of anterior large vessel occlusion from the derivation cohort

The derivation cohort was comprised of 165 LVOs out of 1,026 strokes (16.1%). Univariate analysis demonstrated association of anterior LVO stroke with NIHSS, body mass index (BMI) ≥ 25, extracranial carotid atherosclerosis, intracranial atherosclerotic stenosis, atrial fibrillation, current smoking status, and alcohol consumption. Subjects with diabetes mellitus type 2 (DM2) and intake of oral anti-diabetic medication demonstrated a significant inverse relationship with anterior LVO (Table 2). In multivariate analysis, NIHSS (OR = 1.152, 95% CI 1.118 – 1.187, p < 0.001), extracranial carotid atherosclerotic stenosis (OR = 2.558, 95% CI 1.699 – 3.852, p < 0.001), intracranial atherosclerotic stenosis (OR = 1.772, 95% CI 1.167 – 2.692, p = 0.007), current smoking status (OR = 1.599, 95% CI 1.044 – 2.449, p = 0.031), and DM2 (OR = 0.0438, 95% CI 0.282 – 0.681, p < 0.001) were independently associated with anterior LVO stroke (Table 3). The AUC for this model was 0.796 (95% CI 0.769 – 0.821; Hosmer and Lemeshow test p = 0.637). Two weighted prediction scores were developed from the multivariable regression model (Table 3), assigning points for total NIHSS (weight 1+ per point), current smoking status (+3 points), history of extracranial carotid atherosclerotic stenosis (+7 points) and intracranial atherosclerotic stenosis (+4 points) and subtracting points for DM2 (−6 points). In the derivation cohort, the prediction model score yielded an AUC of 0.796 (95% CI 0.769 – 0.821). The prediction model predicted LVO more accurately than NIHSS alone (AUC_d-NIHSS = 0.725, (95% CI 0.696 – 0.753), DeLong test p = 0.0001) (Figure 1A). The curtailed prehospital score with an AUC_d-pre of 0.757 (95% CI 0.729 – 0.784; Hosmer and Lemeshow test p = 0.180) was inferior to the prediction model (DeLong test p = 0.0114), but superior to NIHSS alone (DeLong test p = 0.0125). The proportion of patients with anterior LVO in relation to prediction model score is presented in the supplement (for more details please see Figure II in supplementary materials at http://stroke.ahajournals.org). Using the 75% percentile of the NIHSS score (NIHSS 6), an additional prediction model was created by dichotomizing into NIHSS < 5 and NIHSS ≥ 6. This model yielded an AUC of 0.767 (0.739 – 0.793; Hosmer and Lemeshow test p = 0.091), which was inferior to the prediction model using each NIHSS point (DeLong test p = 0.0012) (for more details please see Table IV in supplementary materials at http://stroke.ahajournals.org).

Table 2.

Characteristics of derivation and validation cohort used for prediction model

	Derivation cohort (n=1026)			Validation cohort (=514)
	Anterior circulation LVO (n=165)	Ø LVO (n = 861)	OR (95% CI) / p-value	Anterior circulation LVO (n=83)	Ø LVO (n= 431)
Age	67.7 ± 14.0	67.1 ± 13.3	0.582	66.7 ± 14.0	68.2 ± 12.6
Females	91 (55.2%)	441 (51.2%)	1.171 (0.838 – 1.637)	43 (51.8%)	190 (44.1%)
BMI ≥ 25	121 (73.3%)	688 (79.9%)	0.691 (0.471 – 1.014)	64 (77.1%)	342 (79.4%)
Current smoker&	50 (33.1%)	196 (22.9%)	1.667 (1.146 – 2.424)	27 (36.0%)	82 (19.2%)
Alcohol consumption§	64 (45.1%)	311 (37.5%)	1.367 (0.954 – 1.957)	22 (29.7%)	138 (34.1%)
Carotid stenosis	100 (60.6%)	292 (33.9%)	2.998 (2.128 – 4.223)	50 (60.2%)	140 (32.5%)
Intracranial atherosclerotic stenosis	73 (44.2%)	222 (25.8%)	2.284 (1.621 – 3.219)	35 (42.2%)	114 (26.5%)
Atrial fibrillation	44 (26.7%)	184 (21.4%)	1.338 (0.914 – 1.960)	19 (22.9%)	98 (22.7%)
Diabetes mellitus type 2	43 (26.1%)	345 (40.1%)	0.527 (0.363 – 0.766)	28 (33.7%)	176 (40.8%)
NIHSS$	7.5 (10.0 ± 8.2)	2 (3.8 ± 4.5)	< 0.001	6 (9.0 ± 9.0)	2 (3.3 ± 4.0)
NIHSS$ ≥ 6	94 (58.5%)	174 (21.2%)	5.288 (3.702 – 7.553)	41 (51.2%)	71 (17.3%)

NIHSS presented with median (mean ± SD). Data missing in derivation and validation cohort:

^&19 and 11;

^§55 and 35,

^$46 and 23 strokes

Table 3.

Weighted prediction score components of anterior large vessel occlusion

Variable	OR (95% CI)	ß coefficients	Prediction model	Prehospital score
NIHSS (per point)	1.1552 (1.118 – 1.187)	0.141	+1	+1
Current smoker	1.599 (1.044 – 2.449)	0.469	+3	+3
Carotid stenosis	2.558 (1.699 – 3.852)	0.939	+7	N/A
Intracranial atherosclerotic stenosis	1.772 (1.167 – 2.692)	0.572	+4	N/A
Diabetes mellitus type 2	0.438 (0.282 – 0.681)	− 0.824	−6	−6

Figure 1.

Area under the curve for prediction model, prehospital score, and National Institutes of Health Stroke Scale (NIHSS) alone. Receiver operating curves in derivation (A) and validation (B) cohorts for prediction model, prehospital score, and NIHSS alone in anterior circulation large vessel occlusion (LVO) acute ischemic stroke (AIS).

Validation of the LVO prediction score

The validation cohort was comprised of 83 LVOs out of 514 strokes (16.1%). The validation cohort was used to assess validity of the prediction model and the curtailed prehospital score. Using the prediction model, the AUC was 0.770 (95% CI 0.729 – 0.806). The Youden index J was 0.46 for a score of > 6 with 84.7% sensitivity and 62.0% specificity. Additional sensitivity, specificity, positive, and negative predictive values are given in Table 4. The prediction model predicted LVO in the validation cohort more accurately than the prehospital score (AUC_v-pre = 0.689 (95% CI 0.646 – 0.730), DeLong test p = 0.0015) and NIHSS alone AUC_v-NIHSS = 0.665 (95% CI 0.620 – 0.707), DeLong test p = 0.002). The curtailed prehospital score and NIHSS alone yielded similar AUCs (DeLong test p = 0.3625). (Figure 1B). The prediction model dichotomizing NIHSS < 5 and NIHSS ≥ 6 yielded an AUC of 0.747 (0.706 – 0.785, Hosmer and Lemeshow test p = 0.104), with this model showing a trend to be inferior to the prediction model using each NIHSS point (DeLong test p = 0.0817).

Table 4.

Sensitivity, specificity, PPV, and NPV of different cutoff values of the anterior LVO prediction model

Cut off	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)
> −1	94.4	17.2	16.7	94.6
> 2	91.7	41.7	21.7	96.6
> 5	84.7	58.3	26.4	95.6
> 8	63.9	72.3	28.9	91.9
> 11	45.89	86.3	37.1	90,0
> 14	30.6	92.2	40.7	88.3
> 17	22.26	97.1	57.1	87.6
> 20	15.3	98.5	64.7	86.8
> 23	9.7	99.8	87.5	86.2

PPV = positive predictive value, NPV = negative predictive value

Discussion

Early prediction of acute ischemic LVO stroke is critical for timely triage of patients amenable to mechanical thrombectomy, as earlier endovascular intervention provides better functional outcomes.⁶ Prior studies predicted LVO with various scores using a breakdown of the original NIHSS, since the NIHSS represents the most accurate predictor of LVO.^11–19 To simplify and shorten scoring in the prehospital setting, breakdown of the rather complex original NIHSS has repeatedly been investigated. The current study addresses a broad array of individual stroke-related variables to identify additional predictors of anterior circulation LVO stroke. We found NIHSS, current smoking status, prevalence of intracranial atherosclerotic disease and extracranial atherosclerotic carotid stenosis, as well as DM2 to independently associate with anterior circulation LVO stroke. Our model predicted LVO stroke more accurately than NIHSS alone, which highlights the importance and utility of past medical history items present on admission. Consideration of documented past medical history is particularly valuable when a single health care system provides longitudinal care to the patient. The present study was performed in a healthcare system in the northeastern United States with 29 counties of the service area designated as rural. The system also provides care to a remarkably stable population (with the exception of two counties, the out-migration rate is less than 1% per year), which offers the opportunity to review the past medical history that has been documented in the EHR while the patient is en route, and before non-invasive cerebrovascular imaging has been obtained. The non-NIHSS factors that we identified as significant may also be valuable in the decision to transfer the patient to a comprehensive instead of a primary stroke center.

Selection bias in study cohort

Recruitment for the exome sequencing program was random and unrelated to the stroke diagnosis.^20,21 Comparison of the study cohort (exome-sequenced) of acute ischemic strokes to non-sequenced strokes admitted during the study period demonstrated comparable demographics. Whereas distinct cardio- and cerebrovascular risk factors such as prior TIA or strokes, coronary artery disease, and COPD were evenly distributed, others were more frequent in either the study cohort or the non-sequenced cohort. Overall, absolute risk factor frequency differences were moderate at best, ranging from 0.4 to 9.8%. Importantly, cardio- and cerebrovascular risk factors in both groups counterbalanced quite evenly. Thus, an unintended selection bias can be neglected. Distribution of TOAST stroke subtypes and NIHSS were comparable to other large stroke studies.²²

Proposed large vessel occlusion stroke scales based on NIHSS

Whether emergency medical services should bypass primary stroke centers to comprehensive stroke centers with endovascular treatment on the basis of suspected LVO is under debate. Early identification of LVO patients is crucial for appropriate transfer, thus various LVO prediction scores have been proposed although with notable variability in sensitivity (0.61 in PASS¹¹ to 1.00 in VAN¹²), specificity (0.40 in CPSSS¹³ to 0.90 in VAN¹²), positive predictive value (0.31 in G-FAST¹⁴ to 0.86 in NIHSS-9negative predictive value (0.82 in FAST-ED¹⁵ to 0.94 in RACE¹⁶) and area under the curve (0.75 in 3I-SS¹⁷ to 0.84 in ASTRAL¹⁸). Some scores have primarily been designed to assess risk stratification of LVO in a prehospital setting by non-neurologists and emergency medical service providers (RACE¹⁶, LAMS¹⁹, VAN¹²). In the present study, the AUC values for the derivation and validation cohorts were 0.796 and 0.770, respectively. Presence of DM2, absence of extracranial carotid and intracranial atherosclerotic stenosis, and a low NIHSS score in non-smokers yielded a prediction score of ≤ 5, resulting in a negative predictive value of > 95%. In contrast, a NIHSS of ≥ 10 in the smoking, non-diabetic stroke patients also suffering from extracranial carotid and intracranial atherosclerotic stenosis yielded a score of ≥ 24, resulting in a predictive value of > 87%. These findings, when compared with the aforementioned scoring systems, raise the possibility that they might be augmented by inclusion of the past medical history items identified in the present study. While atherosclerotic stenosis may not be known until cerebrovascular imaging is obtained, smoking and diabetes mellitus history can be readily obtained in the field and represent important elements of the prehospital score supplementing NIHSS. There was no upper time limit from stroke onset to cerebrovascular imaging in acute stroke patients in the current study. The purpose of the study was to detect LVO and not whether LVO patients were eligible for mechanical thrombectomy. While randomized trial data found mechanical thrombectomy beneficial up to 24 hours from stroke onset²³, patients with exceptionally favorable penumbra to core ratio may benefit from thrombectomy even beyond 24 hours with successful cases reported up to 120 hours from onset.²⁴ A recent DEFUSE 3 post hoc analysis showed a persistent favorable target mismatch profile > 38 hours after last known well time in 20% of the medical arm patients. Still, only 10% of those patients had a favorable outcome at 90 days implying a potential role for mechanical thrombectomy even beyond 24 hours.²⁵

Non-NIHSS factors associated with AIS caused by anterior circulation LVO

Diabetes mellitus type 2 was significantly more frequent in patients with non-LVO ischemic stroke in our cohort. Therefore, 6 points were subtracted for presence of DM2. Association of DM2 with lacunar ischemic stroke has also been observed by Tanizaki et al. who found a relative risk of 2 in their community-based prospective cohort study in Japanese.²⁶ Others also reported DM2 as a risk factor for lacunar infarction²⁷, likely owing to microangiopathic small vessel disease.²⁸ Ntaios et al. compared risk factors for small vessel occlusion and large-artery atherosclerotic strokes in diabetic patients. A total of 1,069 patients from four European prospective stroke registries were analyzed. They found peripheral atherosclerotic disease and smoking to be predictive for diabetic patients suffering from large-artery atherosclerotic stroke rather than small vessel occlusive stroke.²⁹ Hence, DM2 is involved in both macro- and microvascular pathologies that result in stroke, whereas generalized atherosclerosis and smoking appear to increase the risk for large-artery atherosclerotic stroke. These observations match our findings, where patients with extracranial carotid and intracranial atherosclerotic stenosis and current smoking were more likely to suffer from LVO, with large-artery atherosclerosis representing a major source of LVO. We also found intake of oral anti-diabetic medication to be associated (p = 0.001) with non-LVO ischemic strokes compared to LVO-strokes. However, it remains to be determined whether intake of oral antidiabetic-medication is a mere indicator of DM2 or whether it is integrated in the pathophysiology of LVO.

Large vessel occlusion may result from extracranial and intracranial large-artery atherosclerotic plaque rupture and thrombosis with or without associated embolization (the primary entity of LVO in the current study), cardioembolism, or unidentified causes.^30–32 Atrial fibrillation is the primary risk factor for cardioembolic stroke. Recently, Inoue et al. reported presence of atrial fibrillation and a systolic blood pressure of maximum 170 mmHg in ischemic stroke patients presenting with NIHSS > 7 to be predictive for LVO.³³ This cohort differed from ours as follows: the rate of 32/56 LVO patients (57.1%) suffering from atrial fibrillation in their cohort significantly exceeded our rates of 44/165 (26.7%) in the derivation and 19/83 (22.9%) in the validation cohorts. Second, among the LVO patients in the Japanese cohort, 21.4% had a prior stroke and 57.1% had atrial fibrillation. And anticoagulants and/or antiplatelet medications were only administered to 17.9% and 19.6%, respectively. Third, in our study NIHSS of 7 or higher was present in 21.1% of the entire cohort, i.e., the majority of patients presented with NIHSS below 7. In contrast, in the study of Inoue et al. the majority the patients presented with more severe AISs (i.e. NIHSS ≥ 8).³³

Limitations

Data collection and analysis were performed retrospectively and, as such, are subject to incomplete datasets. The individual elements of the NIHSS were not consistently documented, thus existing prehospital LVO prediction scores could not be validated. The study included mostly patients with AIS presenting prior to 2015, at which time the benefit of mechanical thrombectomy for LVO stroke had not yet been proven. Also, past medical history items may be more difficult to obtain if no immediate contact or family members are available or in areas where the fluctuation of patients between healthcare providers is high. Attribution of extracranial carotid and intracranial atherosclerotic stenosis was not assigned to a specific site. In this context, pathophysiological concepts cannot be evaluated. In contrast, inclusion of global presence of either of these past medical history items allows for integration into a scoring system in a clinical setting. Accounting for each NIHSS point is clinically less practical. Using the prediction model with a dichotomized NIHSS cut off value of 6 (the 75% percentile of NIHSS distribution in the present study cohort), however, was inferior to the prediction model using each NIHSS point. AUC values 0.6 to 0.7 are moderate at best. Thus, curtailing the prediction score mitigated predictability of LVO. Comparison to existing prediction scores was not feasible due to lack of NIHSS element documentation. The present model warrants validation in a larger cohort. Whether a prehospital score, such as the LAMS or FAST-ED, could be augmented by past medical history items remains to be determined.

Conclusions

Use of prediction scores may improve triage of suspected and subsequently confirmed LVO strokes to comprehensive stroke centers. Previously reported LVO stroke prediction scores focus solely on elements of the neurologic exam. In addition to the NIHSS, smoking, type 2 diabetes mellitus, extracranial carotid, and intracranial atherosclerotic stenosis were associated with anterior circulation LVO AIS in the current study. While atherosclerotic stenosis may not be known until cerebrovascular imaging is obtained, smoking and diabetes mellitus history can be readily obtained in the field and represent important elements of the prehospital score supplementing NIHSS. Our findings suggest that further investigation is needed to confirm the above risk factors and test for additional ones, in addition to NIHSS-related items, that optimally predict LVO stroke.