Repurposing electroencephalogram monitoring of general anaesthesia for building biomarkers of brain ageing: an exploratory study

David Sabbagh, Jérôme Cartailler, Cyril Touchard, Jona Joachim, Alexandre Mebazaa, Fabrice Vallée, Étienne Gayat, Alexandre Gramfort, Denis A. EngemannBJA 2023 · DOI

TL;DR

This study demonstrates that electroencephalogram (EEG) data collected during general anesthesia can be repurposed to predict a patient's "brain age," offering a novel avenue for understanding brain aging outside of traditional lab settings. While promising, the effectiveness of these brain age biomarkers is significantly influenced by the anesthetic drug used, highlighting the need for careful consideration in future research and clinical applications.

What Problem Does This Paper Address?

Electroencephalography (EEG) is routinely used to monitor anesthetic depth during surgery. However, this rich source of brain activity data is largely underutilized for broader biomedical research. The authors aim to bridge this gap by exploring whether EEG data captured during general anesthesia, a state of induced unconsciousness, can be repurposed to build biomarkers of brain aging. This is particularly interesting because it could allow for brain health assessment in a diverse population without the need for specialized neuroimaging equipment or conscious participation, potentially revealing insights into neurodegenerative disease risk.

Methods

The researchers employed machine learning techniques, specifically focusing on brain age estimation, a method that uses brain data to predict a person's chronological age. They reanalyzed EEG data from 323 patients who underwent general anesthesia using either propofol or sevoflurane. Key to their approach was the extraction of "spatio-spectral features" from stable anesthesia periods, defined by specific frequency band characteristics. They then used ridge regression for model building and compared various feature sets, including total power, alpha power, the entire power spectrum, and spatial patterns derived from electrode covariances. For clinical exploration, they investigated the relationship between the predicted brain age and intraoperative burst suppression, a phenomenon linked to cognitive issues.

Datasets

The study utilized EEG and clinical data from the "Probrain" study, comprising 323 patients. This dataset included demographic information, anesthetic drug type (propofol or sevoflurane), and ASA scores (a measure of patient health status). The data was split into subsets for model construction (moderate-risk patients under propofol) and clinical exploration (including both moderate and high-risk patients). No new benchmark datasets were introduced; instead, the study leveraged existing methods for brain age prediction.

Key Insights and Ideas

  • EEG during anesthesia is a goldmine for brain aging research: The study successfully demonstrates that EEG signals captured during general anesthesia, a state of unconsciousness, contain valuable information for predicting brain age, comparable to findings from laboratory-based EEG studies.
  • Beyond alpha power: While alpha band activity is commonly monitored during anesthesia, the study found that the entire power spectrum and the spatial relationships between electrodes hold more predictive power for brain age. This suggests that anesthesia EEG offers a richer signal than previously appreciated for cognitive research.
  • Anesthetic drug matters: The brain age signatures derived from EEG are highly specific to the anesthetic drug used. Propofol-based EEG data yielded good prediction performance, but this did not generalize well to sevoflurane, indicating that drug choice is a critical factor for repurposing this data.
  • Burst suppression link is complex: Brain age was linked to intraoperative burst suppression, a marker often associated with age-related cognitive decline. However, this relationship showed a surprising inversion in high-risk patients, suggesting that health status and anesthetic interactions can complicate these associations.

Key Results

  • State-of-the-art brain age prediction: Using the entire power spectrum and spatial patterns from 4-electrode EEG during propofol anesthesia, the model achieved a Mean Absolute Error (MAE) of 8.2 years and an R² of 0.65.
  • Propofol specificity: Models trained on propofol EEG performed significantly better than those trained on sevoflurane EEG, with an MAE difference of approximately 4.9 years. Cross-drug generalization was poor, with MAEs around 14-15 years.
  • Burst suppression association: A significant effect of brain age was found on the logit of burst suppression (beta = 0.86, p < 0.001), suggesting that older brain age is associated with increased burst suppression. However, this effect was reversed in high-risk patients (ASA=3), where brain age was negatively correlated with burst suppression (beta = -0.99, p = 0.025).

Limitations and Future Work

The authors acknowledge several limitations. The primary one is the drug specificity, where brain age signatures are not generalizable across different anesthetic agents like propofol and sevoflurane. They also note that confounding effects of drug dosage could not be fully disentangled, particularly for propofol, where dosage might be influenced by age and health status.

For future work, the authors emphasize the need for larger datasets with precisely documented drug dosages to overcome these limitations. They suggest future studies should focus on precise control and measurement of drug dosage, ideally using constant propofol dosage to rule out confounding factors. Furthermore, validation against pre-surgical brain age estimates derived from gold-standard methods like MRI or high-density EEG is crucial to establish the clinical utility of these anesthesia-derived biomarkers.

Summary updated

Abstract

Although EEG from general anaesthesia may enable state-of-the-art age prediction, differences between anaesthetic drugs can impact the effectiveness and validity of brain-age models. To unleash the dormant potential of EEG monitoring for clinical research, larger datasets from heterogeneous populations with precisely documented drug dosage will be essential.
BibTeX
@inproceedings{sabbagh2023repurposing,
  author = {David Sabbagh and Jérôme Cartailler and Cyril Touchard and Jona Joachim and Alexandre Mebazaa and Fabrice Vallée and Étienne Gayat and Alexandre Gramfort and Denis A. Engemann},
  title = {{Repurposing electroencephalogram monitoring of general anaesthesia for building biomarkers of brain ageing: an exploratory study}},
  booktitle = {BJA},
  year = {2023},
  doi = {10.1016/j.bjao.2023.100145},
}
Figure