Correlating EEG Dynamics With Consciousness Alteration Under Anesthesia

Overview

This prospective observational study is designed to investigate and compare the dynamic features of whole-brain electroencephalogram (EEG) during the induction of unconsciousness using various anesthetic agents with distinct pharmacological mechanisms. The primary objective is to identify common, drug-agnostic EEG biomarkers of anesthetic depth and to develop a novel, universal assessment system that addresses the limitations of the currently prevalent Bispectral Index (BIS), which demonstrates variable sensitivity across different anesthetics. Approximately 250 adult patients (ASA I-II) scheduled for elective surgery under general anesthesia will be enrolled. Patients will undergo preoperative cognitive assessment prior to induction. During anesthesia induction, 32-channel EEG signals will be continuously recorded alongside BIS values and behavioral state assessments using the MOAA/S scale as the reference standard. Patients will receive one of the following intravenous anesthetics for induction: Propofol, Ciprofol, Remimazolam, Esketamine, or Fospropofol. Features will be extracted from the preprocessed EEG data. Statistical analyses will compare these features across drug groups and in relation to behavioral state transitions. Machine learning models (e.g., Random Forest) will then be trained to classify states of consciousness based on the extracted EEG features, with model performance validated against the behavioral gold standard. The study aims to establish a more robust and generalizable neurophysiological framework for monitoring anesthetic depth, potentially improving the precision and safety of clinical anesthesia management.

The aim of the study is to identify and validate common whole-brain EEG biomarkers that accurately track the transition between conscious states (wakefulness, sedation, unconsciousness) across five intravenous anesthetics with distinct mechanisms of action: Propofol, Ciprofol, Remimazolam, Esketamine, and Fospropofol. Design: This is a single-center, prospective, observational cohort study. Consecutive eligible patients will be enrolled and grouped based on the clinical choice of anesthetic drug used for induction of general anesthesia. Data analysis will be performed by researchers blinded to the group allocation during the feature extraction and model development phases. Approximately 250 adult patients (aged ≥18 years) scheduled for elective surgery under general anesthesia at Tongji Hospital, Wuhan, China, will be recruited between April 2026 and December 2027. Participants must have an ASA physical status of I or II, normal cognitive function (MMSE score ≥24), and provide written informed consent. Interventions and Procedures: All procedures represent standard clinical care; no experimental interventions are administered. 1. Preoperative Assessment: Demographics, medical history, and MMSE score will be recorded. 2. EEG and Behavioral Data Acquisition: During anesthesia induction, 32-channel EEG will be continuously recorded using a Greentek system with electrodes placed according to the international 10-20 system. Simultaneously, the BIS value (sensor placed infraorbitally)\[referrence\] and the patient's behavioral state will be recorded every 30 seconds using the Modified Observer's Assessment of Alertness/Sedation (MOAA/S) scale as the reference standard and the absence of the pupillary light reflex. Based on these behavioral responses, the depth of anesthesia will be categorized into three stages: sedation, adequate anesthesia depth, and deep anesthesia. 3. Anesthetic Protocol: Induction of anesthesia will be performed by the attending anesthesiologist in accordance with standard institutional practice. One of the five study drugs will be administered as an intravenous bolus, with the induction dose maintained via continuous infusion for 5 minutes. Data Processing and Analysis 1. Data Curation: Data will be checked for quality, and epochs with artifacts or missing clinical data will be excluded. 2. EEG Feature Extraction: Pre-processed EEG data will be analyzed to extract features including but not limited to power spectral density, permutation entropy, phase-lag entropy, and functional connectivity metrics. 3. Data Analysis: During the induction of unconsciousness with five distinct general anesthetic agents, EEG biomarkers corresponding to transitions between three behavioral states-sedation, adequate anesthesia depth, and deep anesthesia-will be identified. The performance of these biomarkers in tracking depth of anesthesia will be quantitatively compared against that of the Bispectral Index (BIS). 4. Machine Learning Modeling: The dataset will be split into training (70%) and validation (30%) sets. A tree-based ensemble model (e.g., Random Forest) will be trained to classify consciousness states based on EEG features. Model performance will be evaluated using AUC, accuracy, precision, and cross-validation.