Algorithm Development Through AI for the Triage of Stroke Patients in the Ambulance With EEG

Algorithm Development Through Artificial Intelligence for the Triage of Stroke Patients in the Ambulance With Electroencephalography

Endovascular thrombectomy (EVT) enormously improves the prognosis of patients with large vessel occlusion (LVO) stroke, but its effect is highly time-dependent. Direct presentation of patients with an LVO stroke to an EVT-capable hospital reduces onset-to-treatment time by 40-115 minutes and thereby improves clinical outcome. Electroencephalography (EEG) may be a suitable prehospital stroke triage instrument for identifying LVO stroke, as differences have been found between EEG recordings of patients with an LVO stroke and those of suspected acute ischemic stroke patients with a smaller or no vessel occlusion. The investigators expect EEG can be performed in less than five minutes in the prehospital setting using a dry electrode EEG cap. An automatic LVO-detection algorithm will be the key to reliable, simple and fast interpretation of EEG recordings by ambulance paramedics. The primary objective of this study is to develop one or more novel AI-based algorithms (the AI-STROKE algorithms) with optimal diagnostic accuracy for identification of LVO stroke in patients with a suspected acute ischemic stroke in the prehospital setting, based on ambulant EEG data.

RATIONALE

Large vessel occlusion (LVO) stroke causes around 30% of acute ischemic strokes (AIS) and is associated with severe deficits and poor neurological outcomes. Endovascular thrombectomy (EVT) enormously improves the prognosis of patients with LVO stroke, but its effect is highly time-dependent. Because of its complexity and required resources, EVT can be performed in selected hospitals only. In the Netherlands, approximately half of the EVT-eligible patients are initially admitted to a hospital incapable of performing EVT, and - once it has been ascertained that the patient requires EVT - the patient needs to be transported a second time by ambulance to an EVT-capable hospital. Interhospital transfer leads to a treatment delay of 40-115 minutes, which decreases the absolute chance of a good outcome of the patient by 5-15%. To solve this issue, a prehospital stroke triage instrument is needed, which reliably identifies LVO stroke in the ambulance, so that these patients can be brought directly to an EVT-capable hospital. Electroencephalography (EEG) may be suitable for this purpose, since it shows almost instantaneous changes in response to cerebral blood flow reduction. Moreover, significant differences between EEGs of patients with an LVO stroke and those of suspected AIS patients with a smaller or no vessel occlusion have been found. A dry electrode EEG cap enables ambulance paramedics to perform an EEG in the prehospital setting, with significant reduced preparation time compared to conventional wet electrode EEG. An automatic LVO-detection algorithm will be the key to reliable, simple and fast interpretation of the EEG by paramedics, enabling direct admission of suspected AIS patients to the right hospital.

HYPOTHESIS

An EEG-based algorithm, developed with artificial intelligence (AI), will have sufficiently high diagnostic accuracy to be used by ambulance paramedics for prehospital LVO detection.

OBJECTIVE

The primary objective of this study is to develop one or more novel AI-based algorithms (the AI-STROKE algorithms) with optimal diagnostic accuracy for identification of LVO stroke in patients with a suspected AIS in the prehospital setting, based on ambulant EEG data.

STUDY DESIGN

AI-STROKE is an investigator-initiated, multicenter, diagnostic test accuracy study.

STUDY POPULATION

Part A: Adult patients with a (suspected) AIS, in the prehospital setting. Part B: Adult patients with a (suspected) AIS, in the in-hospital setting.

INTERVENTION

A single EEG measurement with a dry electrode cap (approximately 2 minutes recording duration) will be performed in each patient. In addition, clinical and radiological data will be collected. EEG data will be acquired with a CE approved portable dry electrode EEG device.

MAIN END POINTS

Primary end point: Based on the EEG data, and using the final diagnosis based on CT angiography data established by an adjudication committee as the gold standard, one or more novel AI-based EEG algorithms (the AI-STROKE algorithms) will be developed with maximal diagnostic accuracy (i.e. area under the receiver operating characteristic curve; AUC) to identify patients with an LVO stroke of the anterior circulation in a population of patients with suspected AIS.

Secondary end points:

• AUC, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of the AI-STROKE algorithms based on ambulant EEG for diagnosis of LVO of the anterior circulation in suspected AIS patients in the prehospital setting;
• AUC, sensitivity, specificity, PPV and NPV of existing EEG algorithms based on ambulant EEG for diagnosis of LVO stroke of the anterior circulation in suspected AIS patients in the prehospital setting;
• AUC, sensitivity, specificity, PPV and NPV of existing and newly developed EEG algorithms based on ambulant EEG for detection of LVO stroke of the posterior circulation, intracerebral hemorrhage, transient ischemic attack, and stroke mimics;
• Technical and logistical feasibility (e.g. in terms of EEG channel reliability) of paramedics performing ambulant EEG in patients with a suspected AIS in the prehospital setting.

Inclusion Criteria:

• Suspected AIS, as assessed by the attending ambulance paramedic, or a known LVO stroke;
• Onset of symptoms or last seen well < 24 hours before EEG acquisition;
• Age of 18 years or older;
• Written informed consent by patient or legal representative (deferred).

Exclusion Criteria:

• Skin defect or active infection of the scalp in the area of the electrode cap placement;
• (Suspected) COVID-19 infection.