Prehospital stroke recognition — CPSS facial droop / arm drift / speech exam, RACE & VAN large-vessel-occlusion screens, and time-critical routing
The Cincinnati Prehospital Stroke Scale (CPSS) was designed in the mid-1990s to give EMS providers and bystanders — not neurologists — a fast, teachable way to recognize acute stroke in under a minute. It distills a full neurological exam into three simple maneuvers: look at the face, hold out the arms, and listen to speech. Any one abnormal finding is enough to treat the patient as a stroke code and activate the local EMS stroke protocol.
CPSS is administered as three quick, scripted tests:
1. Facial droop — ask the patient to smile or show their teeth. • Normal: both sides of the face move symmetrically. • Abnormal: one side of the face does not move as well as the other, or does not move at all — a flattened nasolabial fold and a mouth corner that fails to rise.
2. Arm drift — ask the patient to close their eyes and hold both arms straight out in front of them, palms up, for 10 seconds. • Normal: both arms move the same, or both arms do not move at all. • Abnormal: one arm drifts downward compared to the other, or one arm pronates (rotates palm-down) — a sign of unilateral weakness (hemiparesis) that the patient often cannot feel happening because the eyes are closed and proprioception may itself be impaired.
3. Speech — ask the patient to repeat a sentence, classically "You can't teach an old dog new tricks." • Normal: patient uses correct words with no slurring. • Abnormal: patient slurs words, uses the wrong words, or cannot speak at all.
Each maneuver targets a different cortical/subcortical territory — facial droop and arm drift localize to the contralateral motor cortex or corticospinal tract, while abnormal speech can reflect either dysarthria (motor speech, brainstem/cerebellar) or aphasia (language, typically dominant-hemisphere cortical).
The CPSS scoring rule is deliberately binary and forgiving: interpreting "one of three abnormal" as positive is what makes the tool usable by first responders with no neurology training, at the cost of not distinguishing stroke subtype or severity — that refinement is left to hospital-based scales like the NIHSS.
CPSS was adapted from three items of the 15-item NIH Stroke Scale (NIHSS), the standard in-hospital tool neurologists use to quantify stroke severity on a 0–42 scale. Kothari and colleagues at the University of Cincinnati selected the facial palsy, arm motor drift, and best-language/dysarthria items specifically because they could be taught to paramedics in minutes and administered without any equipment.
This design choice reflects a core principle of prehospital medicine: a screening tool only has value if it can be deployed reliably by the actual first responders, under real field conditions (moving ambulance, roadside, poor lighting, anxious bystanders), not just by trained specialists in a controlled clinic setting.
A screening tool is only as useful as its ability to correctly flag true strokes (sensitivity) without generating an overwhelming number of false alarms (specificity). CPSS was validated against physician neurological exam and CT/clinical follow-up in the original 1997 Cincinnati study, and has since been re-tested by many EMS systems with a range of reported performance depending on population, training, and which finding(s) are required to call a positive screen.
The original Kothari et al. (1997) study found that patients with any one of the three CPSS findings had a substantially elevated probability of stroke, with the scale correctly identifying roughly two-thirds of confirmed strokes at the single-finding threshold — a sensitivity figure often cited as approximately 66%. Subsequent validation studies, larger EMS deployments, and re-analyses using a "positive if 1 or more of 3 findings abnormal" cutoff have reported sensitivity ranging as high as 89–100% and specificity in the range of 60–90%, depending on:
• Whether the stroke was anterior circulation (large MCA-territory strokes with obvious face/arm/speech deficits) versus posterior circulation (vertebrobasilar strokes, which often present with vertigo, ataxia, or isolated visual loss and are frequently missed by CPSS) • Paramedic training quality and how strictly the exam maneuvers were performed • How "abnormal" was operationally defined (subtle vs. overt asymmetry) • The base rate of stroke mimics in the population tested (seizure, hypoglycemia, migraine, Bell's palsy, conversion disorder)
In practice, CPSS detects the large majority of anterior circulation strokes — the kind most likely to benefit from thrombolysis or thrombectomy — while being less sensitive to subtle, lacunar, or posterior circulation events.
CPSS is intentionally tuned toward sensitivity over specificity: missing a treatable stroke (false negative) costs a patient irreversible brain tissue and disability, while a false positive costs a stroke-center evaluation, a CT scan, and a few hours of a physician's time for a stroke mimic. This asymmetry of costs is why prehospital stroke tools accept a meaningful false-positive rate — commonly 10–50% of CPSS-positive patients turn out to have a stroke mimic rather than true stroke — as the price of not missing treatable ischemic strokes.
Common stroke mimics that trigger a false-positive CPSS: hypoglycemia, seizure with postictal (Todd's) paralysis, complex migraine, Bell's palsy (isolated facial nerve palsy with no other deficit), conversion/functional neurological disorder, and prior stroke deficits being re-triggered by an unrelated illness ("recrudescence").
CPSS answers "is this a stroke?" but says almost nothing about "is this a large vessel occlusion (LVO)?" — the subset of ischemic strokes, caused by a clot blocking a proximal artery (internal carotid, M1/M2 middle cerebral artery, basilar), that respond dramatically to mechanical thrombectomy but require a comprehensive stroke center with an interventional neuroradiology suite. RACE and VAN were built specifically to be administered in the field, in under two minutes, to help EMS decide whether to bypass the nearest hospital for one capable of thrombectomy.
RACE extends the basic CPSS exam with graded severity scoring and an added gaze/language item, producing a 0–9 point scale:
• Facial palsy: 0 = absent, 1 = mild, 2 = moderate-to-severe (graded, not just present/absent) • Arm motor function: 0 = normal/mild, 1 = drifts but does not fall, 2 = falls rapidly or no effort against gravity • Leg motor function: scored the same way as the arm, 0–2 • Head and gaze deviation: 0 = absent, 1 = present — forced conjugate eye deviation toward the side of the lesion is a strong marker of a large cortical territory being knocked out • Aphasia (if right-sided weakness, i.e. suspected left-hemisphere lesion): 0 = can name objects, 1 = cannot name/identify, 2 = does not understand commands • Agnosia/neglect (if left-sided weakness, i.e. suspected right-hemisphere lesion): scored 0–2 the same way, testing for hemispatial neglect or asomatognosia
A RACE score of ≥5 is the most widely used field cutoff for suspecting LVO and triggering direct transport to a thrombectomy-capable center, with reported sensitivity around 85% and specificity around 68% for LVO detection in validation cohorts — notably better at LVO discrimination than CPSS or the NIHSS cutoffs used alone.
RACE deliberately mirrors NIHSS scoring logic (graded 0–2 severity rather than binary present/absent) because higher-grade motor deficits and gaze deviation are disproportionately associated with the large, proximal clots that thrombectomy can retrieve — a small lacunar stroke rarely produces gaze deviation or severe (grade-2) drift.
VAN was designed as an even faster alternative to RACE, built around a simple two-part rule rather than a summed point score:
Step 1 — Is there any arm or leg weakness (motor deficit)? • If no motor deficit at all → VAN negative, LVO very unlikely, no bypass indicated on VAN criteria alone.
Step 2 — If motor weakness is present, is there ALSO at least one of: • Vision: a new visual field cut or gaze preference • Aphasia: new difficulty producing or understanding speech • Neglect: inattention to one side of the body or space, denial of the deficit
If motor weakness AND at least one of vision/aphasia/neglect are present → VAN positive, suspicious for LVO, consider bypass to a comprehensive/thrombectomy-capable center per local protocol.
VAN's appeal is cognitive simplicity in a high-stress field environment: it is a yes/no decision tree rather than a point-scoring exercise, at some cost of the finer severity gradation that RACE provides.
Large vessel occlusions account for roughly a quarter to nearly half of anterior circulation ischemic strokes but are disproportionately disabling — LVOs are responsible for the majority of severe post-stroke disability because they knock out large territories of brain tissue rather than a small perforator-vessel distribution. Mechanical thrombectomy for LVO has one of the largest treatment effects in modern medicine: pooled trial data (the 2015 HERMES meta-analysis of MR CLEAN, ESCAPE, SWIFT PRIME, EXTEND-IA, and REVASCAT) showed a number-needed-to-treat of approximately 2.6 for a meaningfully better disability outcome.
Because thrombectomy is only available at comprehensive stroke centers with 24/7 interventional neuroradiology coverage, correctly identifying LVO in the field — and being willing to bypass a closer primary stroke center that cannot perform the procedure — is often the single highest-leverage decision an EMS crew makes for that patient's long-term outcome.
"Time is brain" is not just a slogan — it is a quantitative description of ongoing tissue death. Jeffrey Saver's widely cited 2006 analysis estimated that, on average, an untreated ischemic stroke destroys about 1.9 million neurons, 14 billion synapses, and 12 km of myelinated fibers every minute reperfusion is delayed. Every subsequent minute of delay — in recognition, dispatch, transport, triage, imaging, and treatment decision — has a real, countable neurological cost, which is exactly why treatment eligibility is defined by strict, shrinking time windows rather than by clinical judgment alone.
Saver's 2006 Stroke editorial modeled the average large-vessel ischemic stroke (affecting a territory of roughly 1.2 billion neurons) and estimated the pace of neuronal, synaptic, and myelinated-fiber loss per minute without reperfusion:
• ~1.9 million neurons destroyed per minute • ~14 billion synapses lost per minute • ~12 km (7.5 miles) of myelinated fibers destroyed per minute • At this rate, each hour of delay ages the brain by the equivalent of roughly 3.6 years of normal age-related neuron loss
These figures are population averages from core-and-penumbra modeling, not a literal per-patient countdown — the true rate depends heavily on collateral blood flow, which varies between patients and determines how quickly the ischemic penumbra (salvageable tissue) converts into permanently infarcted core. But the directional message is unambiguous: every minute of prehospital delay is not neutral, it is actively costing brain tissue.
The "1.9 million neurons per minute" figure is the single most quoted statistic in stroke systems-of-care advocacy precisely because it converts an abstract urgency ("hurry") into a concrete, memorable number that motivates EMS dispatch prioritization, scene-time minimization, and prenotification of receiving hospitals.
Intravenous alteplase (recombinant tissue plasminogen activator, tPA) dissolves the clot pharmacologically and was first proven effective in the landmark 1995 NINDS trial within a 3-hour window from last-known-well. The ECASS III trial (2008) extended proven benefit out to 4.5 hours in appropriately selected patients (with some additional exclusion criteria such as age >80, prior stroke plus diabetes, oral anticoagulant use, or severe stroke, depending on guideline version), and this 3-to-4.5-hour window is now the standard operating target cited in most EMS and ED protocols worldwide.
Beyond 4.5 hours from clear symptom onset, imaging-based selection can still identify candidates for thrombolysis: the WAKE-UP trial (2018) used MRI diffusion-weighted imaging/FLAIR mismatch — tissue that shows acute ischemic change on DWI but has not yet turned bright on FLAIR — as a surrogate for "stroke likely began within the last 4.5 hours" in patients who woke up with a deficit and have no reliable last-known-well time. The EXTEND trial (2019) used perfusion-imaging mismatch to justify alteplase out to 9 hours from onset (or mid-point of sleep for wake-up strokes) in selected patients with salvageable penumbra. These extended-window pathways require CT or MR perfusion imaging and are only available at centers with the capability to perform and interpret them rapidly.
Mechanical thrombectomy, in which an interventional neuroradiologist threads a stent-retriever or aspiration catheter through the femoral or radial artery up into the occluded cerebral vessel to physically remove the clot, was traditionally offered within 6 hours of onset, based on the 2015 HERMES-era trials (MR CLEAN, ESCAPE, SWIFT PRIME, EXTEND-IA, REVASCAT).
Two practice-changing 2018 trials dramatically extended this window using perfusion/clinical mismatch imaging:
• DAWN (DWI or CTP Assessment with Clinical Mismatch in the Triage of Wake-Up and Late Presenting Strokes Undergoing Neurointervention) — enrolled patients 6–24 hours from last-known-well with a clinical deficit severity that was disproportionately large relative to a small measured infarct core on imaging, implying a large territory of salvageable penumbra despite the late hour.
• DEFUSE 3 (Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke) — enrolled patients 6–16 hours from last-known-well using automated perfusion imaging (RAPID software) to identify a small infarct core with a large surrounding penumbra.
Both trials were stopped early for overwhelming efficacy, and together they moved international guidelines to endorse thrombectomy out to 24 hours in carefully imaging-selected patients — the single largest expansion of a stroke treatment window in the modern endovascular era.
Once a positive CPSS screen and a suspicious RACE/VAN result are in hand, the EMS crew faces the same category of decision trauma systems have used for decades: transport to the nearest facility, or bypass it for a farther, higher-capability center. For suspected LVO, that means weighing the extra transport minutes of driving past a primary stroke center against the minutes saved by avoiding an inter-facility transfer later — a calculation increasingly formalized into regional stroke routing protocols.
Stroke systems of care are typically organized into tiers, mirroring trauma system design:
• Acute Stroke-Ready Hospitals: can administer IV alteplase and stabilize, but must transfer for anything beyond that • Primary Stroke Centers (PSC): certified for full stroke unit care and IV thrombolysis, but without in-house endovascular thrombectomy capability • Thrombectomy-Capable / Comprehensive Stroke Centers (CSC): everything a PSC offers, plus 24/7 interventional neuroradiology or neurosurgery able to perform mechanical thrombectomy
When a suspected LVO patient is closer to a PSC than a CSC, two competing transport strategies exist:
"Drip-and-ship": transport to the nearest PSC, start IV alteplase there if eligible, then transfer by ambulance or helicopter to a CSC for thrombectomy. Minimizes time-to-needle for thrombolysis but adds substantial inter-facility transfer delay (commonly 1.5–2.5 hours) before the clot can be mechanically retrieved.
"Mothership" (direct bypass): transport straight to the CSC, skipping the closer PSC, so thrombectomy can begin as soon as possible, with IV alteplase given at the CSC if still eligible. Minimizes time-to-thrombectomy but may add extra minutes of ground transport up front and modestly delay IV thrombolysis if the CSC is meaningfully farther away.
Most regional protocols now set an explicit bypass rule of thumb: bypass the nearest primary stroke center for a comprehensive/thrombectomy-capable center only if the added transport time is within a fixed threshold (commonly 15–30 extra minutes, protocol-dependent) AND the field LVO screen (RACE ≥5 or VAN positive) is met — balancing the benefit of skipping a transfer against the cost of a longer initial transport.
A positive CPSS/RACE screen does more than choose a destination — it triggers a cascade at the receiving hospital before the patient even arrives. EMS prenotification (a radio or app-based "stroke alert" call) allows the receiving emergency department to:
• Activate the stroke team and have the CT scanner cleared and ready on arrival • Pre-mix or have alteplase immediately available at bedside • Alert the interventional neuroradiology/neurosurgery team if RACE/VAN suggests LVO, so the angiography suite can be prepared in parallel with imaging rather than sequentially after it
Hospitals track "door-to-needle" time (arrival to IV alteplase administration, target under 45–60 minutes per most quality benchmarks) and "door-to-groin-puncture" time (arrival to start of thrombectomy) as core quality metrics — both of which prenotification measurably shortens, often by 10–20 minutes, translating directly into salvaged brain tissue under the "time is brain" framework.
The cumulative effect of fast recognition, accurate LVO screening, and correct destination routing is measured in modified Rankin Scale (mRS) outcomes at 90 days — the standard disability metric in stroke trials, ranging from 0 (no symptoms) to 6 (death). Every increment of delay to reperfusion — whether from slow recognition, a missed LVO screen, or an unnecessary detour through a non-capable hospital — shifts the population-level distribution of 90-day mRS scores toward greater disability.
Conversely, systems that combine early CPSS-based recognition, RACE/VAN-based LVO triage, prenotification, and correct primary-vs-comprehensive routing have been shown in registry data to meaningfully increase the proportion of LVO patients achieving functional independence (mRS 0–2) at 90 days compared to systems without organized stroke routing — the entire chain of tools covered in this simulation exists to protect exactly that outcome.