If a scale existed that could accurately predict the presence of large vessel occlusion (LVO), it would be extremely useful in triaging patients to either primary or comprehensive stroke centers (CSCs). For patients with LVO who are candidates for endovascular therapy, every minute is critical. Time lost by triaging these patients to primary stroke centers (PSCs) without endovascular capability is time and brain lost. Unfortunately, the range of stroke scales is wide and score cutoffs are inconsistent, and data on their predictive value for detecting LVO is limited.
The authors of this study assessed 13 different clinical scales for their ability to predict LVO. The cutoff score for each scale which was associated with an under-10% false negative rate (FNR) was also calculated. The false negative rate would include patients with LVO who were not detected by the score cutoff, and so this number would ideally be minimal. Of over 1000 acute stroke patients seen from 2008-2015, about one-third had large artery occlusion (ICA, M1, or basilar). Patients transferred from a primary stroke center for endovascular therapy were excluded, as the authors mention this would have led to too high a prevalence of large vessel occlusions.
The scales included were: modified NIHSS (mNIHSS), 3-item stroke scale (3I-SS), Rapid Arterial oCclusion Evaluation Scale (RACE), Cincinnati Prehospital Stroke Scale (CPSS), Cincinnati Prehospital Stroke Severity Scale (CPSSS), Maria Prehospital Stroke Scale (MPSS), shortened versions of the NIHSS (sNIHSS-1, sNIHSS-5, sNIHSS-8), abbreviated NIHSS (aNIHSS), out-of hospital NIHSS (OoH-NIHSS), retrospective NIHSS profiles (rNIHSS: A to F), and Recognition of Stroke in the Emergency Room (ROSIER). The NIHSS was calculated for each patient on admission (median of 7), and the remaining stroke scale scores were retrospectively calculated from NIHSS score components. The published accepted cutoffs for each score were used to assess predictive value for LVO, and if there was not a published value, the cutoff which maximized the sum of specificity and sensitivity for LVO was used.
The scores with the highest accuracy were NIHSS (11 and over) and RACE (5 and over), which had 79% accuracy, but these cutoffs were associated with false negative rates (FNR) around 30%. Using NIHSS >= 11 as a cutoff in this cohort would have led to sending 35% of the cohort to a CSC, but 27% of LVO patients would have been inappropriately triaged to primary stroke centers when they should have been sent to a CSC. This false negative rate of 27% falls in the published range for NIHSS >= 11, which has been reported between 12-35%.
Arguably, it is worse to have a high FNR than a high FPR. A high FNR means time (and brain) lost, while a high FPR means overburdening the CSCs (but not necessarily worse care for patients). To achieve a FNR under 10%, the cutoff scores had to be quite low – the NIHSS cutoff was 5, mNIHSS was 3, RACE was 1, and Cincinnati Prehospital Stroke Severity Scale (CPSSS) was zero. As expected, using published cutoff scores (such as NIHSS 11) for triage would inappropriately send about 25% of patients with LVO to centers without endovascular capability. If, however, we used the calculated cutoffs which reduced the FNR to under 10% (such as NIHSS of 5), 60% of patients would have been sent to a comprehensive stroke center, 46% of whom would have been futile transfers, overburdening the system.
Limitations noted by the authors include the changing cutoff score for LVO as time passes, the lower predictive value of NIHSS for LVO in the posterior circulation, and the lack of training of emergency medical teams in performing the NIHSS. The simpler scales seemed to be a solution to address the complexity of the NIHSS, but these scores posed similar problems in this analysis as did the NIHSS. This important study highlights the pitfalls of using the existing clinical scales to predict LVO. The authors recommend that intracranial artery imaging should be performed in all stroke patients presenting within 6 hours of onset, since the scores cannot be reliably used. They bring up the idea of a mobile stroke unit, which could be used to image in the field and triage patients. Additionally, biomarkers and TCDs have potential roles in the future for detecting LVO stroke patients in the field. This study is limited by its population, in that the patients were all admitted to a CSC and had a diagnosis of stroke. A study on the predictive value of stroke scales performed in the prehospital setting would provide real-world data along with the ability to quantify the role of the examiner’s proficiency with the stroke scale. In the meantime, optimizing the transfer process to swiftly identify and transport LVO patients from PSCs to a CSC with endovascular capability is critical to ensure that our patients get the best stroke care possible.