Table 4 EEG during acute assessment to predict outcomes after confirmed stroke

Sainio 1983 [41]

Admission and 7-day disability

15 Ischaemic Stroke patients

TIA

 < 48

16 electrodes, eyes closed with checking for wakefulness

Online only, > 30 Hz, time constant 0.3 s, sampling 100 Hz

Relative spectral power (all bands), focal and background slowing

Poorer admission outcome associated with background (p = 0.00016) and focal (p = 0.0099) abnormalities, greater ipsilesional rolandic and occipital delta₂ (p’s = 0.005) and less ipsilesional rolandic and occipital alpha (p = 0.005 and p = 0.025 respectively). Poorer 7-day outcome associated with background abnormalities (p = 0.0089) greater ipsilesional (p = .025) and contralesional (p = 0.025) delta₂ and less ipsilesional alpha (p = 0.025)

4

Charlin 2000 [42]

Day 90 mRS

47 Ischaemic Stroke Patients

Epilepsy, cirrhosis, cancer, pre-stroke dependence; sedatives

 < 24

16 electrodes

None

PLEDs plus and PLEDS proper

Worse outcome (mRS > / = 3) associated with PLEDs (p = 0.03, AUC = 0.62, sensitivity = 30.8%, specificity = 93.75%). (prognostic accuracy extrapolated from true and false positive and negative values)

3

Cuspineda 2003 [43] ^a

mRS at discharge and within three months

28 Ischaemic Stroke patients (MCA territory)

Haemorrhage

 < 72

19 electrodes, awake, eyes open and closed, reclining, temperature controlled

Online filters 0.3-30 Hz, notch 60 Hz, manual artifact removal, 2.56 s epochs

Absolute spectral power (absolute energy)

Discharge and 3-month outcome (mRS) predicted by assessment of EEG absolute energy variables with 100% accuracy (r = 0.99). QEEG predicted outcome at discharge better than the CaNS (p = 0.03)

2

Cuspineda 2007 [44] ^a

mRS at discharge and within three months

28 Ischaemic Stroke patients (MCA territory)

Haemorrhage

 < 72

19 electrodes, awake, eyes open and closed, reclining, temperature controlled

Online filters 0.3-30 Hz, notch 60 Hz, manual artifact removal, 2.56 s epochs

Absolute spectral power (all bands Absolute Energy)

Poorer outcome at discharge (mRS) predicted by.less alpha (Accuracy = 92.3% r = 0.95) and beta (Accuracy = 69.2%, r = 0.76) and greater theta (Accuracy = 92.3%, r = 0.94) and delta (Accuracy = 84.6%, r = 0.85) power within 24 h. Poorer outcome at 3 months predicted by less alpha (Accuracy = 88.9%, r = 0.97) and beta (Accuracy = 77.8%, r = 0.83)and greater delta (Accuracy = 88.9%, r = 0.92, r 0.87) and theta (Accuracy = 77.8%, r = 0.83) within 24-48 h

4

Sheorajapanday 2011 [15]

Day 7 mRS

60 Ischaemic Stroke patients

Mass lesion; ICH; seizure(s); hypo/hyperglycaemia

Most < 72

19 electrodes, eye closed, awake/alert

Online montage re-referencing; filters > 0.3 Hz, < / = 30 Hz, manual artifact removal, 128 s epochs, FFT

Relative spectral power (DTABR), BSI

Greater DTABR predicted unfavourable outcome (mRS score >  = 2) in LACS (AUC = 0.88; accuracy = 0.83%, p = 0.01) but not in POCS

5

Su 2013 [45]

Three-month mRS

162 Ischaemic Stroke patients (large MCA infarct)

Pre-stroke dependence, concurrent illness affecting outcome, sedatives; extraneous factors affecting consciousness

 < 72

8 electrodes; pain and auditory stimulation

Online filter 0.5-70 Hz, time constant 0.3 ms

Dominant fast/slow wave with/without reactivity, RAWOD, epileptiform activity, burst and general suppression; alpha/theta coma

Significant associations between worse outcome (mRS > 4) and RAWOD (OR = 2.47, sensitivity = 37%, specificity = 85%) and good outcome and dominant alpha with reactivity (OR = .08, but poor sensitivity = 7.4%, specificity = 49.3%). All other markers had > 80% specificity but < 40% sensitivity in predicting poor outcome. Modified grading most accurate (Kappa = 0.61, p = 0.04, sensitivity = 77.9%, specificity = 89.6%, accuracy = 91.4%)

4

Lima 2017 [46]

Three-month mRS

157 Ischaemic Stroke patients (19 with seizures)

Previous seizures, debilitating neurological disorders, hypo/hyperglycaemia

 < 45.5

19 electrodes

None

Epileptiform activity (IED and PP)

Worse outcome (mRS > / = 3) associated with epileptiform activity (OR = 2.94, p = 0.001) but not when seizures excluded (OR = 2.13, p = 0.07). AUC = 0.60, sensitivity = 51.3%, specificity = 69%). (prognostic accuracy extrapolated from true and false positive and negative values)

4

Bentes 2017 [47] ^a

mRS (including mortality) at discharge and within 1 year

151 Ischaemic Stroke patients (ICA; NIHSS 4–42)

Prestroke dependence, traumatic brain injury or surgery, hydrocephalus, history of epilepsy

 < 72

64 electrodes, eyes open and closed, resting, hyperventilation and photic stimulation

Not Reported

Asymmetry, Suppression, focal slow-waves, epileptiform activity; periodic discharges

Worse outcome (mRS > / = 3) at discharge associated with EEG background (OR = 5.55, p = 0.002) slowing, asymmetry (OR = 11.91, p < 0.001) and periodic discharges (OR = 10.39, p = 0.027). Worse outcome at 1 year predicted by background slowing (OR = 14.50, p < 0.001) and asymmetry (OR = 22.73, p > 0.001) and periodic discharges (OR = 14.1, p = 0.002). Clinical and radiological predictors plus background asymmetry (AUC = 0.91, sensitivity = 81.1%, specificity = 88.7%) was a better model than clinical data plus past seizures (AUC = 0.83, sensitivity = 72.1%, specificity = 77.5%), clinical (AUC = 0.82, sensitivity = 70.3%, specificity = 73.2%), asymmetry (AUC = 0.81, sensitivity = 72.7%, specificity = 89%) and past seizures (AUC = 0.59, sensitivity = 25.7%, specificity = 93.2%) in isolation. 12-month mortality associated with EEG acute symptomatic seizures (OR = 4.55, p = 0.015) and EEG suppression (OR = 7.48, p = 0.019). Clinical/radiological predictors plus EEG suppression (AUC = 0.84, sensitivity = 31.8%, specificity = 99.2%) were a better predictor than clinical data plus acute seizures (AUC = 0.82, sensitivity = 40.9%, specificity = 100%), and clinical data (AUC = 0.81, sensitivity = 22.7%, specificity = 98.4%), acute seizures (AUC = 0.64, sensitivity = 0%, specificity = 100%), and suppression (AUC = 0.61, sensitivity = 26.1%, specificity = 96.1%) in isolation

5

Xin 2017 [48]

BI/mRS at 21 days

29 Ischaemic Stroke patients

TIA, ICH, previous stroke, cardiovascular disorders, traumatic brain injury, tumour, ‘serious disease’, pregnancy

 < 72

16 electrodes, < 3 h after meal; sedatives discontinued 3 days prior

Online and offline, filters < 0.53 Hz, > 50 Hz. Sampling 100 Hz, EOG, ECG, EMG, visual and wavelet transform artifact removal, 10 s epochs

r-BSI

Worse outcome (lower BI and higher mRS) associated with higher r-BSI at admission (BI -2.070, P = 0.049, mRS 2.256, P = 0.033)

3

Bentes 2018 [49] ^a

mRS at discharge and one year

151 Ischaemic Stroke patients (ICA;NIHSS 4–42)

Prestroke dependence, traumatic brain injury or surgery, hydrocephalus, history of epilepsy

 < 72

64 electrodes, eyes open and closed, resting, hyperventilation and photic stimulation

Offline filters < / = 0.5 Hz, > 70 Hz, notch 50 Hz, manual and automatic artifact removal, 2.05 s epochs; FFT

Absolute spectral power (all bands, DAR, DTABR); BSI

Worse outcome (mRS > / = 3) associated with greater delta (discharge AUC = 0.812, OR = 125; 12 months AUC = 0.836, OR = 129.8), and DTABR (discharge AUC = 0.827, OR = 1.702; 12 months AUC = 0.859, OR = 1.668) and less alpha (discharge AUC = 0.814, OR = 0.221; 12 months AUC = 0.852, OR = 0.16) and beta (discharge AUC = 0.803, OR = 0.28; 12 months AUC = 0.829, OR = 0.28) power (all p > 0.001; theta not significant). The best discharge models combined clinical/radiological predictors with background asymmetry (AUC = 0.831, sensitivity = 81.3%, specificity = 68%), DTABR (AUC = 0.827, sensitivity = 87.5%, specificity = 60%), alpha power (AUC = 0.756, sensitivity = 86.9%, specificity = 46.2%) and background slowing (AUC = 0.787, sensitivity = 82.3%, specificity = 60%). The best 12-month models combined clinical/radiological predictors with background asymmetry (AUC = 0.89, sensitivity = 81.1%, specificity = 88.7%), background slowing (AUC = 0.866, sensitivity = 78.4%, specificity = 87.3%), DTABR (AUC = 0.859, sensitivity = 79.7%, specificity = 74.6%) and alpha (AUC = 0.852, sensitivity = 75.7%, specificity = 78.9%). Isolated clinical data, followed by DTABR and alpha were good predictors (AUC’s = 0.768–0.794, sensitivity = 70.1–76.6%, specificity = 64.4–71.8%) (all p > 0.001)

4

Kuznietsov 2018 [50]

21-day mRS

103 Ischaemic Stroke patients (supratentorial)

Cardiovascular or psychiatric disorders, traumatic brain injury, ICH, tumour, past seizure(s)

 < 72

19 electrodes

Offline independent component analysis artifact removal, 60 s epochs; FFT

Absolute and relative spectral power (All bands, RSRP; FORG; IHRA)

Worse outcome post-stroke (mRS) associated with higher RSRP of delta band in contralesional hemisphere > 18.4% (OR = 1.31, p = 0.0004; AUC = 0.94, sensitivity = 87.0%, specificity = 87.7%, p < 0.0001), lower FORG of alpha band in ipsilesional hemisphere > -0.066 (OR = 29.07, p = 0.0224; AUC = 0.74, sensitivity = 67.4%, specificity = 70.0%, p < 0.0001) and IHRA of alpha band ≤ -0.066 (OR = 0.01, p = 0.0402; AUC = 0.66, sensitivity = 60.9%, specificity = 70.2%, p < 0.0039). No significant differences for other biomarkers

3

Rogers 2020 [51]

30 and 90-Day mRS and mBI

12 Ischaemic Stroke patients, 4 Haemorrhagic Stroke patients

Neurological/psychiatric disorders

 < 72

Single electrode at 10–20 FP1, eyes closed

Online sampling and amplification, Offline filter 0.5-30 Hz, manual and automatic artifact removal, 4 s epochs; FFT

Absolute and relative spectral power (all bands, DAR, DTR, DTABR)

Only relative theta power significantly negatively correlated with mRS (30-day r = -0.54; 90-day r = -0.53) and positively with mBI (30-day r = 0.60; 90-day r = 0.45). Better outcome post-stroke (mBI > / = 95; mRS < / = 1) associated with higher theta values >  = 0.25 for 30-day mRS (AUC = 0.81, sensitivity = 71.4%, specificity = 88.9%, p = 0.04), mBI (AUC = 0.90, sensitivity = 83.3%, specificity = 90%, p < 0.01) and 90-day mBI (AUC = 0.82, sensitivity = 80%, specificity = 81.8%, p = 0.05) but not 90-day mRS (AUC = 0.75, sensitivity = 62.5%, specificity = 87.5%, p = 0.09). EEG theta power was a no more accurate predictor than NIHSS

4

Juhasz 1997 [52]

Modified NIHSS at 1 month

40 Ischaemic Stroke patients

Bilateral stroke

 < 48

16 electrodes

Online filters < / = 0.3, > 30, 4 s and 80 s epochs, artifacts removed

Absolute spectral power (alpha, beta); APF

Worse outcome (NIHSS) post stroke significantly associated with > 0.5 Hz difference in interhemispheric APF (p < 0.02)

3

Vespa 2003 [38]

 < 72 h NIHSS and GOS at discharge

46 Ischaemic Stroke patients, 63 Haemorrhagic Stroke patients (NIHSS 8–42)

Traumatic haemorrhage, SAH, ICH; Brainstem stroke

 < 24

14 electrodes

Online (hospital staff) or offline (EEG segment review or total power trend) seizure detection and classification (focal, hemispheric or generalised)

Epileptiform activity

EEG seizures showed no association with GOS 4–5 (p = 0.25) but differed significantly according to NIHSS < 72 h (p = 0.05)

4

Finnigan 2004 [12]

30 Day NIHSS

11 Ischaemic stroke patients

Fever, encephalitis, seizures, ICH, non-cortical stroke, confounding neurological condition (e.g. previous stroke) or medication

 < 9

64(62) electrodes, alert or drowsy

Online filter .01-100 Hz, artifacts 0.2- 40 Hz, automatic artifact removal, 4 s epochs, sampling 500 Hz, FFT .5-50 Hz

Relative spectral power (aDCI)

Worse outcome (higher NIHSS) associated with greater aDCI (rho = 0.80, P < 0.01)

3

Finnigan 2007 [53]

30 Day NIHSS

13 Ischaemic Stroke patients

Fever, encephalitis, seizures, ICH, confounding neurological condition (e.g. previous stroke) or medication

 < 52

62 electrodes, alert or drowsy

Online filter .01-100 Hz, artifacts 0.2- 40 Hz, EOG artifact removal, 4 s epochs, sampling 500 Hz, FFT .5-50 Hz

Relative spectral power (delta, theta, alpha; beta); DAR

Worse outcome (NIHSS) was associated with greater DAR (r = 0.91, P < 0.001) and less relative alpha power (r = -0.82, P < 0.01). These correlations were also observed in a 19-channel subset

3

Wolf 2016 [40]

Admission and discharge NIHSS

69 Ischaemic Stroke patients

Epileptic seizures

 < 48

10–20 system

Not Reported

Epileptiform activity; focal slowing

Worse outcome post-stroke (deterioration of NIHSS > 3 points admission vs discharge) associated with generalised EEG slowing (p = 0.003)

2

Yang 2017 [24]

7, 14 & 90 Day NIHSS

86 Ischaemic Stroke patients (NIHSS 4–24)

Cardiovascular disorders, pregnancy

 < 4.5

20 electrodes

Online filter .16-70 Hz, sampling 250 Hz, FFT

Relative spectral power (DAR, DTABR), BSI

Neurological improvement of patients post-thrombolysis (decrease in NIHSS by 8 points or return to normal) significantly associated with early decrease in BSI (2 h), DAR (2 h) and DTABR (24 h) (both p < 0.01)

4

De Herdt 2018 [54]

Day 7 NIHSS

29 Ischaemic Stroke patients, 2 Haemorrhagic stroke patients

Not Reported

 < 72

Not Reported

Not Reported

Epileptiform activity (spikes, spike-waves; seizure, PLEDs)

Epileptiform activity not associated with outcome, only useful for predicting seizure incidence (abstract only—no statistics provided)

2

Gur 1994 [55]

Dementia diagnosis, checked every 6 months for 2 years

199 Ischaemic Stroke patients

Cognitive impairment, TIA, ICH, previous stroke

 < 48

18 electrodes

Not Reported

Abnormal EEG patterns, foci, background slowing

Worse outcome (development of dementia) associated with abnormal EEG (OR = 2.6, p = 0.003, AUC = 0.38, sensitivity = 63.4%, specificity = 12.2%) (prognostic accuracy extrapolated from true and false positive and negative values)

3

Wang 2013 [39]

MoCA at two weeks and 2 years

110 Ischaemic Stroke patients

Cognitive impairment, psychiatric disorders, traumatic brain injury, tumour, infection, multi-infarct, systemic disease, psychoactive drug use

 < 10

16 electrodes

Sampling 250 Hz, offline filter 0.5-50 Hz, computer, visual and EOG artifact removal, 2 s epochs,

Relative spectral power (beta only)

Significantly lower beta power with cognitive impairment and larger infarct size (P < 0.01). Sensitivity: 92.3% for predicting impairment and 93.3% for predicting normal cognition. Good concordance between MoCA scores and beta power (Kappa statistic = 0.851, p < 0.001)

3

Song 2015 [56]

MoCA (Beijing version) 11 months—7 years

105 Ischaemic Stroke Patients

Cognitive impairment, psychiatric disorders, traumatic brain injury, tumour, infection, multi-infarct, systemic disease, psychoactive drug use

 < 12

16 electrodes, eyes closed with checking for wakefulness

Online filter 0.5-50 Hz, Offline 2 s epochs, EOG artifact removal, FFT

Relative spectral power (all bands)

Worse outcome associated with high background rhythm frequency (HR = 14 (3.8, 41), P < 0.001) or greater median theta power (HR = 5 (1.4, 7.8), P = 0.002)

4

Aminov 2017 [25]

90 Day MoCA

15 Ischaemic Stroke patients, 4 Haemorrhagic Stroke patients

Neurological/psychiatric disorders, previous stroke

 < 72

Single electrode at FP1, eyes closed

Online filter 0.5-30 Hz, manual artifact removal, 4 s epochs; FFT

Relative spectral power (DAR, DTR)

Better outcome moderately correlated with higher relative theta power (r = 0.50, p = 0.01), lower DAR (r = -0.45, p = 0.03), DTR (r = -0.57, p = 0.01) and relative delta power (r = -0.47, p = 0.02)

4

Yan 2011 [28]

Mortality

22 Stroke patients

Not Reported

 < 48

16 electrodes, eyes closed, resting

Offline visual artifact removal followed by digital filter, 10 s epochs. FFT

BBSI

BBSI > 0.082 predicted mortality with an accuracy of 86.36%

2

Chen 2018 [30]

Mortality at Day 90

47 Haemorrhagic Stroke patients

Aneurysm, vascular malformation, traumatic head/brain injury, tumour, infection/encephalitis

 < 59

16 electrodes, eyes closed and awake; supine

Offline filters > 0.3, < / = 30 Hz, artifacts removed. FFT

Relative spectral power delta, alpha, DAR, DTABR), BSI

Mortality at Day 90 was associated with higher DAR (OR 5.306, p = 0.008). AUC for TCD-QEEG(DAR) model = 0.949

4

Jiang 2019 [57]

Mortality at discharge and six months

58 Ischaemic Stroke patients

Prestroke dependence, consciousness altering drugs, haemorrhage, tumour, encephalitis, epilepsy

 < 72

16 electrodes

Online filters 0.5-30 Hz and offline visual artifact rejection. FFT

Relative spectral power (All bands, DTABR), BSI

Mortality at discharge and six months post-stroke associated with greater contralateral electrode theta power > / = 25.53 (discharge p = .038, accuracy = 68%, sensitivity = 69.2%, specificity = 66.7%), 6-month p = 0.026, accuracy = 64%, sensitivity = 45.2%, specificity = 94.7%). No other biomarkers significantly contributed to the model

4