Predictive Value of the Prehospital RACE Scale for Large Vessel Occlusion in Acute Stroke Patients

Endovascular thrombectomy (EVT) has been proven to be more effective than medical therapy alone in patients with acute ischemic stroke (AIS) due to anterior large vessel occlusion (LVO) up to 24 h from symptom onset or last known well (LKW) [1‒4]. The promising benefits of EVT are also seen in selected LVO patients using perfusion imaging or more recently noncontrast head CT in the very late window [5‒7]. Currently, EVT is commonly offered by Comprehensive Stroke Centers (CSCs), while Primary Stroke Centers (PSCs) or Community Hospitals can only administer intravenous thrombolysis (IVT) and stroke care. Because of the limited availability of EVT and its time-dependent efficacy, there is a critical need to rapidly identify AIS patients with LVO and direct transport to the EVT-capable stroke centers [8].

Several prehospital stroke scales have been designed and validated to predict the presence of LVO in patients with acute stroke. In a comparative analysis of eight prehospital stroke scales (RACE, LAMS, C-STAT, G-FAST, PASS, CPSS, CG-FAST, FAST-PLUS), the results showed that the Rapid Arterial oCclusion Evaluation (RACE) scale and G-FAST scale were the two highest accuracy scales for LVO detection [9]. The RACE scale is a simplification of the NIHSS scale with a strong correlation and similar capacity to predict LVO compared with the NIHSS [10]. Although the RACE scale is a valid tool which is widely used in many countries, the performance of this scale in the Vietnamese population remains unknown. Moreover, previous studies of the RACE scales have primarily focused on patients assessed within 6 or 8 h of symptom onset [10, 11]. There are limited publications examining this scale in the 6-to-24-h window, or comparing its performance across different time periods. Such comparisons may be essential since optimal NIHSS cutoff values for LVO prediction tend to decrease over time, and may differ between the early- and late-time windows [12]. Therefore, the aim of this study was to evaluate the predictive value of the RACE scale for LVO detection in patients with AIS presenting within 24 h in Vietnam.

We conducted a prospective study of AIS patients who were admitted to the emergency department (ED) within the first 24 h of symptom onset or LKW at People’s 115 Hospital between May 2022 and October 2022. All patients were assessed with five items on the RACE scale: facial palsy (scored 0–2), arm motor function (0–2), leg motor function (0–2), gaze (0–1), and aphasia or agnosia (0–2) at hospital admission. The RACE scale assessment was performed in the ED by neurology physicians from the stroke team as part of the routine work-up, with the results confirmed by consensus of at least two physicians. LVO was confirmed on admission using computed tomography angiography (CTA) or magnetic resonance angiography (MRA). LVO was defined as occlusion of the internal carotid artery or middle cerebral artery (MCA) (M1 or M2 segment). Cerebrovascular imaging studies (CTA/MRA) were reviewed by a blinded, independent neuroradiologist to confirm LVO diagnosis. Patients were excluded if there was evidence of intracranial hemorrhage (ICH) on brain imaging or if there was no information about cerebrovascular status. The decision for EVT was made by a treating physician according to the hospital protocol. Baseline characteristics, risk factors, time from onset, RACE score, imaging data, and revascularization treatment were registered prospectively. The study was approved by the Ethics Committee of the People’s 115 Hospital and Pham Ngoc Thach University of Medicine, and followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline [13].

Receiver operating characteristic (ROC) curves and areas under the curves (AUCs) with the respective 95% confidence intervals (95% CIs) were calculated to evaluate the predictive value of the RACE scale for LVO detection. Cross tables for different thresholds of the RACE scale were used to evaluate the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and overall accuracy in identifying LVO patients. The optimal cutoff for the RACE score was determined at the maximal Youden index. We also calculated AUC, specificity, sensitivity, and accuracy of the RACE scale for detecting LVO in early-time (0–6 h) and late-time window (6–24 h) groups. AUC values were compared using the method of DeLong et al. [17]. Significance testing was done by the χ² test, Fisher’s exact test (for categorical variables), and Mann-Whitney U test or t test (for continuous variables) where appropriate. Data were analyzed using SPSS (version 25.0) or MedCalc (version 22), with a p value of 0.05 considered statistically significant.

This study included 318 patients (Fig. 1). Baseline characteristics of patients are summarized in Table 1. LVO was detected in 121 of 318 patients (38.1%). Diagnosis of LVO was performed mostly by CTA in 82.7% of patients and less frequently by MRA (17.3%). LVO was found in 36 (11.3%) patients in the internal carotid artery, 69 (21.7%) in the MCA-M1, and 16 (5.1%) in the MCA-M2. The median RACE score was significantly higher in AIS patients with LVO (6.0 [3.0–8.0]) than in those without LVO (3.0 [3.0–4.0]) (p < 0.001). As the RACE score increased, there was a higher proportion of patients with LVO (Fig. 2).

ROC curves indicated that the RACE scale effectively identified LVO in AIS patients up to 24 h with an AUC of 0.767 (95% CI, 0.71–0.82) (Fig. 3a). Different thresholds of the RACE scale for predicting LVO were assessed (Table 2). The optimal predictive value of RACE was achieved for a score of ≥5, with a sensitivity 0.68, specificity 0.79, PPV 0.67, and NPV 0.80 for detecting LVO. Two-thirds of patients with a RACE ≥5 had LVO (67%), compared with 19.6% of those with a RACE <5 (p < 0.001) (Fig. 4). Moreover, a cutoff RACE ≥5 showed a sensitivity of 0.802, specificity 0.75, PPV 0.52, and NPV 0.92 for EVT treatment. There were 65 of 124 (52.4%) patients with a RACE ≥5 who received EVT compared with 16 of 194 (8.2%) of those with RACE scale <5 (p < 0.001).

Of the 318 patients, 129 (40.6%) presented within 6 h from symptom onset, while 189 (59.4%) presented in the late window (6–24 h). There were no significant differences between two groups in age, sex, RACE score, and the prevalence of LVO (Table 3). In prespecified subgroup analyses, the accuracy of the RACE scale for LVO was slightly higher in patients assessed within 0–6 h than those in 6–24 h (AUC, 0.79 vs. 0.75, p = 0.41; Fig. 3b). The RACE score cutoff at ≥5 showed good binary performance in identifying LVO patients in both early- and late-window groups (Table 4). This cutoff point showed slightly lower sensitivity, but same specificity in the late-presenting patients as compared with early-presenting patients (sensitivity = 0.65 vs. 0.74, p = 0.28; specificity = 0.80 vs. 0.79, p = 0.84).

This study indicates that the RACE scale is a simple and valuable tool in predicting LVO among ischemic stroke patients up to the first 24 h of symptom onset or LKW. The RACE scale in our study showed fair performance with an AUC of 0.767 (95% CI, 0.71–0.82) in identifying LVO patients. In general, the predictive ability of the RACE scale for LVO in prospectively tested Vietnamese patients was comparable to the results of prior studies conducted in other populations. In a recent study of the RACE scale based on data from 132 centers in the SITS-ISTR program in Berlin, Scheitz et al. [18] reported that the RACE scale predicted anterior LVO stroke with an AUC of 0.755 (95% CI, 0.740–0.769). Similar predictive capability of the RACE scale for LVO was observed in real-world experiences reported by the emergency medical service (EMS) personnel in Spain and in the USA, with the AUC of the RACE score to be 0.77 and 0.72, respectively [11, 19]. Importantly, a RACE score ≥5 in our study detected 80.2% of the patients who eventually underwent EVT, similar to the results of the validation study accessed by EMS in suspected stroke patients within the first 8 h (75%) [11]. These findings suggest that the RACE scale, besides predicting patients with LVO, may also serve as a useful tool to identify patients who could receive EVT up to 24 h. The results of our study strengthen the value of the RACE scale for LVO detection in AIS patients presenting in the late-time window, especially in developing countries, in which prehospital triage system is associated with workflow delays. A study of over 6,000 stroke patients in Vietnam showed that there was a considerable delay in the time from stroke onset to hospital arrival, with the median time reaching 15.7 h, and more than 70% of patients were admitted to hospital more than 3.5 h after stroke onset [20]. These patients are often beyond the window for IVT therapy, making EVT crucial and potentially the only treatment option for those with LVO. Several studies have indicated that the shorter the time from stroke onset to arterial recanalization, the greater the clinical benefit of revascularization treatments [21]. Therefore, early identification of LVO stroke patients would enable prompt transfer to CSCs capable of administering both IVT and EVT, thereby minimizing unnecessary delays and improving clinical outcomes [22].

In the current study, the best predictive value of RACE was determined as ≥5, with a high sensitivity and specificity in detecting LVO in AIS patients within 24 h, and even higher sensitivity with the same specificity in patients within the first 6 h. The best cutoff of RACE was also found to be ≥5 in previous cohorts conducted in both the pre- and in-hospital settings [10, 11, 18, 19, 23, 24]. In the pooled studies involving ischemic stroke conducted in the prehospital settings or in the ED, a RACE scale ≥5 showed similar sensitivity (67%) and higher specificity (85%) for detecting LVO compared to our study [15]. Generally, this cutoff value performed with moderate to high sensitivity, but low PPV in identifying LVO among patients with suspected stroke, with PPV ranging from 28.8% to 40.8% accordingly to different publications [15, 18, 19]. The low PPV leads to a significant number of false-positive cases, where patients without LVO are incorrectly identified as candidates for EVT and excessively transported to CSCs. Our study, on the other hand, identified a higher PPV (67%) than previous studies and similar to the findings from the Screening Technology and Outcomes Project in Stroke (STOPStroke) cohort in AIS patients within 24 h (68%) [23]. This is mainly due to the exclusion of patients with ICH, which constituted the majority of false-positive cases in RACE ≥5 patients [11]. Furthermore, based on a recent analysis of the RACECAT Randomized Clinical Trial, direct transfer of hemorrhagic stroke patients with severe symptoms (RACE score ≥5) to CSCs may result in worse functional outcomes than transfer to the nearest PSCs for initial treatment [25]. Therefore, these findings suggest that combining the clinical RACE scale with technological diagnostic methods such as noncontrast CT to differentiate AIS and ICH might be promising strategies to reduce the number of false-positive cases and improve the accuracy of prehospital triage.

In practice terms, determining the RACE threshold value should be based on various factors. A high-sensitive threshold will reduce the risk of missing patients with LVO (low false negatives) but may lead to more transfers of patients without LVO and potentially overwhelm EVT-capable hospitals. Thus, if patients are assessed early after symptom onset or in close proximity to the CSCs, a RACE cutoff with high sensitivity is preferable to avoid missing possible LVO, as observed in our study with RACE ≥3 (sensitivity 93%). On the other hand, if there is a large distance between patients and the nearest CSC, a RACE cutoff with high specificity (low false positives) would be better to minimize potential delays in receiving IVT and futile transfers among AIS patients without LVO, which can be seen in our study with RACE ≥6 (specificity 88%).

Our study has some limitations. First, the study was conducted at a single, thrombectomy-capable stroke center, which may limit the generalizability of the present study to prehospital settings. Second, the data did not include patients with posterior circulation stroke, limiting the scope of our findings regarding the performance of the RACE scale in this group of patients. Third, due to the exclusion of patients with ICH stroke, the rate of LVO in our study could be higher than in the general population with suspected stroke seen in the prehospital setting. Nevertheless, our findings remain valid because we focused on sensitivity and specificity, which are considered to be less influenced by disease prevalence. Fourth, we lack data on the impact of the RACE score on time to treatment and clinical outcomes following EVT in early- and late-presenting patients. Future validation cohorts of the RACE scale, evaluated by EMS personnel or with clinical outcome assessments, are needed to strengthen the reliability of our study among patients with suspected stroke in prehospital setting.

The RACE scale is a simple and valuable tool in predicting LVO among AIS patients within the first 24 h. Its performance in Vietnamese patients was comparable to previous studies conducted in other populations. This scale may serve as a useful tool to early detection of LVO patients and should be validated in the prehospital setting in Vietnam.

We would like to express our gratitude to all the clinicians, neurologists, interventionists, imaging and laboratory technicians, and statisticians for their contributions to the data collection and analysis for this study.

This study protocol was reviewed and approved by the Ethics Committee of the People’s 115 Hospital and Pham Ngoc Thach University of Medicine, Approval No. 629/TDHYKPNT-HDDD. Written informed consent to participate in the study was obtained from all adult participants and all vulnerable participants’ next of kin if they were unable to provide consent.

The authors disclosed no potential conflicts of interest regarding the research, authorship, or publication of this article. Thanh N. Nguyen reports advisory board for Idorsia, Brainomix, Associate Editor of Stroke.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Dr. Anh Tuan Le Truong and Dr. Hang Thi Minh Tran contributed to data acquisition, analysis, or interpretation, contributed equally to this work, and wrote the manuscript. Dr. Loc Dang Phan, Huong Bich Thi Nguyen, Trung Quoc Nguyen, Tra Vu Son Le, and Huy Nguyen reviewed and drafted the manuscript. Dr. Duc Nguyen Chiem contributed to data acquisition, statistical analysis, and interpretation of data. Dr. Thanh N. Nguyen and Dr. Thang Huy Nguyen critically revised the manuscript. All authors gave final approval and agreed to be accountable for all aspects of the work, ensuring its integrity and accuracy.

Anh Tuan Le Truong and Hang Thi Minh Tran have made equal contributions to this paper as first authors.

All data generated or analyzed during this study have been included in this article. For further inquiries, contact the corresponding author.