NIH Stroke Scale and Stroke Pathway

1. What the Score Measures, and What It Does Not

Fifteen scored rows across eleven numbered categories, total 0 to 42. It measures neurological deficit at one moment: not infarct volume, not vessel status, not a diagnosis. Baseline NIHSS is the strongest clinical predictor of outcome after ischaemic stroke and enters almost every eligibility rule that follows it.

• The five-band scheme is teaching shorthand. Minor 1 to 4, moderate 5 to 15, moderate to severe 16 to 20, severe 21 to 42: it could not be traced to any primary NINDS or AHA document. Printed here because residents are examined on it, and labelled as informal.
• The sourced anchors are from TOAST, n = 1,281. NIHSS 16 or above forecasts a high probability of death or severe disability; 6 or below forecasts good recovery.
• Indian outcome weight. In IRIS-TNK, 1,015 thrombolysed patients, each additional point carried an odds ratio of 0.823 for a good outcome, ahead of age at 0.983 per year.

2. The Left and Right Asymmetry

The scale gives 7 of its 42 points to language, rows 1b, 1c and 9, and 2 points to neglect, row 11. Language is a dominant hemisphere function and neglect the signature non-dominant syndrome, so the same anatomy scores lower on the right.

The disparity lives exactly where a threshold rule such as "NIHSS 6 or more" does its work. A right-hemisphere patient at 7 does not have a mild stroke: get the vessel imaged and let the angiogram, not the number, decide.

3. Reliability, and Why Four Points Sits Inside the Noise

Reliability is not uniform across the instrument. Certification restores agreement on the total, intraclass correlation above 0.95 in the TOAST programme, and leaves the individual rows as the residual disagreement.

• Lyden and colleagues, certification videos: limb ataxia is the worst row in the scale, kappa 0.21.
• Josephson and colleagues, 7,405 raters: best language and facial palsy contribute most to variance in the total, kappas 0.60 and 0.65.
• Same series: for 7 of 11 videotaped patients the total differed by four points or more between the fifth and ninety-fifth centile of raters.

Four points is the conventional definition of neurological deterioration, and four points with haemorrhage on imaging the conventional definition of symptomatic intracranial haemorrhage; DEFUSE 3 uses it in both senses and is the traceable source. It therefore sits inside the instrument's own between-rater noise: it should trigger a repeat scan and a reassessment, not a conclusion. Score serially with the same examiner where the roster allows, and record the examiner's name against each score.

4. Untestable Is Not Zero, Except Where It Is

Only six rows carry a UN code: 5a, 5b, 6a, 6b, 7 and 10.

• Motor rows. UN only for amputation or joint fusion at the shoulder or the hip. A hand amputation, a wrist fusion, an intravenous line or a plaster cast below the shoulder does not authorise it.
• Row 10. UN requires intubation or another physical barrier to producing speech. Severe aphasia does not qualify.
• Rows 7 and 11 collapse to zero instead, and can only lower the total. Row 7 scores 0 in the patient who is paralysed or cannot comprehend. Row 11 has no UN code at all and states in its own instructions that it is never untestable.

Point to note. The instrument prints no rule for how a UN row enters the total, and no maximum.

• Trials and registries treat UN as 0, which is what this tool does, but that convention could not be verified from a primary NINDS document.
• A total of 18 with an untestable arm is not the same clinical object as a total of 18 fully scored, so the UN rows are printed beside the total and in the EMR text.

5. Bedside Stroke Type Scores in Indian Practice

Scores that separate supratentorial intracerebral haemorrhage from infarction without a CT. Siriraj, Bangkok 1991, is the one Indian residents are taught; the Greek score, 2002, uses neurological deterioration, vomiting, a raised white cell count and depressed consciousness. Only a minority of Indian stroke patients reach a scanner inside the treatment window, which is why these persist.

• The published sensitivities exclude the equivocal cases first. Siriraj is equivocal in roughly 17 to 26 per cent of Indian patients, so those figures overstate what the score does for an unselected patient at the door. This tool reports the equivocal band as an answer in its own right.
• The Greek score is not a door score. Six of its points ride on neurological deterioration within three hours of admission and four more on a white cell count, so it answers a question three hours later than Siriraj does. It is computed only when the leucocyte count has been entered, rather than reading a blank field as a normal count.

6. Tenecteplase in India: Two Doses, One Licence

A fibrin-specific variant of alteplase with a longer half-life, given as a single bolus instead of a bolus and an hour-long infusion. That takes a pump and an hour of nursing attention out of the door-to-needle pathway.

There has been no randomised comparison of the two. The evidence for the Indian dose is IRIS-TNK, a 1,015-patient single-arm safety registry mandated as a condition of approval and funded by the manufacturer, whose authors themselves call for a trial.

• What IRIS-TNK showed. Symptomatic intracranial haemorrhage 0.6 per cent, three-month mortality 1.0 per cent, modified Rankin 0 to 2 at three months 75.1 per cent. Functional independence 93.4 per cent at NIHSS 7 or below, 70.9 per cent at 8 to 14, 46.4 per cent at 15 or above.
• 0.4 mg/kg has been tested and should not be used. EXTEND-IA TNK Part 2 found no added benefit over 0.25 mg/kg; NOR-TEST 2 Part A stopped early for higher mortality and symptomatic haemorrhage.
• Ayushman Bharat PM-JAY carries a separately coded package for tenecteplase thrombolysis in acute ischaemic stroke.

7. ASPECTS: Ten Regions, and Where the Threshold Now Sits

Ten points on a normal scan, one subtracted for each region showing early ischaemic change, so a lower score means a larger infarct. Seven regions at the ganglionic level: caudate head, lentiform nucleus, internal capsule, insular ribbon, and the cortical M1 anterior, M2 lateral to the insula and M3 posterior. Three about 2 cm above the ganglia: M4, M5 and M6, anterior, lateral and posterior.

Six large-core trials between 2022 and 2024 moved that line: RESCUE-Japan LIMIT, SELECT2, ANGEL-ASPECT, TENSION, LASTE and TESLA. TESLA alone selected on non-contrast CT within 24 hours, the imaging reality in most Indian district hospitals, and missed its primary endpoint; TENSION used non-contrast CT but capped the window at 12 hours and was positive. Where perfusion imaging is unavailable, the evidence for treating a very large core is weaker than the Class 1 recommendation suggests.

8. The Prognostic Scores Computed, and the Ones Refused

• DRAGON, computed. All six weights are stated in the primary paper and externally validated: dense artery or early infarct signs, pre-stroke Rankin above 1, age, glucose, onset-to-treatment time, baseline NIHSS. Area under the curve 0.84 in derivation, 0.80 in validation.
• SPAN-100, computed. Age plus NIHSS at or above 100, carried with the authors' own caveat that it must never be used to withhold thrombolysis.
• ICH score, computed. Its five components are quoted verbatim in the primary paper.
• HAT, SEDAN, SITS-SICH, THRIVE, ASTRAL and iScore, refused. The point weights within the components could not be obtained from the primary papers, only the component names and the score-to-risk mapping. For HAT, the two calculator sites that publish weights give the NIHSS bands as "15 to 20" and "20 or above", which overlap at 20 and cannot both be right.
• Guy's Hospital (Allen), refused, although it sits beside Siriraj in every Indian validation study. The 1983 paper is closed access and the secondary restatements disagree: the constant appears as both −12 and −12.6, the infarction cut-off as both "below 4" and "at or below −4". It also needs the diastolic pressure and conscious level at 24 hours, so it cannot inform a decision made in the first four and a half hours, a point Kochar and colleagues made from Bikaner in 2000.

A numeric band with an ambiguous edge is the defect pattern that has recurred most often in this system, so none of the refused scores is computed until its own table has been read.

9. Stroke Mimics, and the Score That Runs the Other Way

Mimics account for close to half of hospital admissions for suspected stroke and 22 to 33 per cent of acute stroke activations in telestroke series. The common ones are seizure and the post-ictal state, migraine with aura, functional disorder, metabolic disturbance and above all hypoglycaemia, sepsis and syncope.

FABS scores six items one point each: absence of Facial droop, no Atrial fibrillation, Age below 50, systolic Blood pressure below 150, history of Seizures, isolated Sensory symptoms without weakness.

• Three or more identifies a mimic at 90 per cent sensitivity and 91 per cent specificity, area under the curve 0.95.
• It runs in the opposite direction to every other score here: a high FABS argues against a stroke.
• It was derived only in patients whose CT was already negative and who presented within 4.5 hours, so it is raised as a caution and never withholds treatment.

10. What This Tool Does Not Cover

• Paediatric stroke, cerebral venous sinus thrombosis, subarachnoid haemorrhage beyond recognising it and stopping.
• Cervical artery dissection; carotid endarterectomy and stenting decisions.
• The full aetiological workup for stroke in the young.
• Post-stroke rehabilitation, and the driving and occupational advice that follows discharge.
• Any reading of a scan. Every imaging finding here is entered by the clinician, and the tool has no way of knowing whether it was entered correctly.