JSMF ACCESS intraoperative derived EEG data set

@article{PhysioNet-jsmf-access-1.0.0,
  author = {Garcia, Paul and Zinn, Sebastian and Sleigh, Jamie},
  title = {{JSMF ACCESS intraoperative derived EEG data set}},
  journal = {{PhysioNet}},
  year = {2026},
  month = sep,
  note = {Version 1.0.0},
  doi = {10.13026/m2hn-w830},
  url = {https://doi.org/10.13026/m2hn-w830}
}

@article{pollard_physionet_2026, title = {{PhysioNet} as a global platform for biomedical research}, volume = {1}, issn = {3005-0693}, url = {https://doi.org/10.1038/s44360-026-00096-z}, doi = {10.1038/s44360-026-00096-z}, abstract = {Twenty-five years after its first public repository was established, the PhysioNet resource for biomedical and clinical data has enabled tens of thousands of research papers. Further efforts are ongoing to advance secure data sharing, community engagement and scalable research infrastructure.}, number = {8}, journal = {Nature Health}, author = {Pollard, Tom and Moody, Benjamin E. and Lehman, Li-wei H. and Gow, Brian J. and Fernandes, Chrystinne and Xie, Chen and Johnson, Alistair and Mark, Roger G. and Heldt, Thomas}, month = aug, year = {2026}, pages = {792--795},}

Abstract

This resource shares processed frontal electroencephalography (EEG) data collected from adult patients undergoing non-emergency, non-cardiac surgery with general anaesthesia at four hospitals. The dataset was generated to study neurophysiological signatures of emergence from anaesthesia and their association with delirium in the postanaesthesia care unit (PACU). The cohort includes data for 646 patients, and delirium was assessed in 626 patients using the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) after surgery. EEG was recorded intraoperatively.

Summarized EEG parameters are included in this release, while full raw EEG (EDF format) is available upon request.

The dataset is intended for research on perioperative brain-state dynamics and prediction of postoperative cognitive complications.

Background

Postoperative delirium is associated with worse clinical outcomes, including increased morbidity, mortality, longer hospitalization, and readmission risk. The immediate postoperative form occurring in the PACU is of particular interest because it may provide an early clinical signal of subsequent complications. The study underlying this resource was motivated by the hypothesis that the neurophysiological pattern from which a patient emerges from general anaesthesia is associated with PACU delirium. Specifically, the investigators examined whether emergence trajectories containing spindle-dominant EEG activity were associated with lower odds of delirium compared with trajectories lacking such features. The resulting dataset is being shared to facilitate reuse of a multi-site perioperative EEG resource for studying emergence dynamics, burst suppression, anaesthetic neurophysiology, and postoperative neurocognitive risk. The resource may also support methodological work in EEG preprocessing, state classification, and biomarker discovery. The study included four collection sites and was designed as an observational investigation without imposing restrictions on the anaesthetic plan beyond standard emergence procedures used to minimize external stimulation.

Methods

Study design and participants

The source study was a prospective, observational, multi-institutional investigation conducted at four hospitals: Emory University Hospital Midtown, Grady Memorial Hospital, Atlanta VA Medical Center, and Waikato Hospital. Adult patients were eligible if they were expected to recover in the PACU after general anaesthesia for non-emergency, non-cardiac surgery that would not interfere with frontal EEG recording. Data from a total of 646 patients were collected, 626 of which had delirium assessed.

Data acquisition

Frontal EEG was recorded intraoperatively by trained study personnel, who applied electrodes in the operating room and monitored signal quality during the case. In Atlanta, EEG was acquired using a SedLine monitor at 250 Hz. In Waikato, EEG was acquired using a BIS XP monitor at 128 Hz. The observational protocol did not direct clinical management based on EEG, although a standardized emergence approach was used after cessation of anaesthetics to minimize external stimulation such as loud noise and oral suction.

Perioperative and outcome measures

Patient characteristics, comorbidities, and perioperative variables were collected by interview, bedside observation, and chart review. Delirium was assessed using CAM-ICU at approximately 15 minutes after PACU arrival and again about 60 minutes after end emergence. Richmond Agitation and Sedation Scale scores were also measured with CAM-ICU assessments. End emergence was defined as the first observed Observer’s Assessment of Alertness/Sedation score of 2 or greater.

EEG processing

Raw EEG (FP_{7 or 8}–FP_Z) was converted to microvolts during preprocessing. SedLine recordings were low-pass filtered to 47 Hz and downsampled to 125 Hz. The highest possible cutoff frequency was 43 Hz. Analysis was performed on 10-second EEG epochs shifted in 1-second steps. Artifacts were excluded using thresholds based on absolute amplitude, maximum point-to-point amplitude difference, and zero-line detection. Power spectral density was estimated with MATLAB using Welch’s method. EEG spectrograms were then used to derive frequency-domain parameters and to categorize emergence into one of seven trajectory classes. Burst suppression was identified by blinded visual review.

Data Description

This dataset comprises de-identified clinical and derived neurophysiological data collected from 646 adult patients undergoing elective non-cardiac surgery with general anesthesia across multiple international centers. 626 of the patients had delirium assessed.

Data Content

Each record includes:

• Clinical variables

  • Demographics and comorbidities

  • Anaesthetic drug dosing and timing

  • Physiological measurements (e.g., haemodynamics)

  • Surgical characteristics (type and duration)

• Delirium and recovery outcomes

  • Post-anaesthesia care unit (PACU) delirium assessments using validated instruments (CAM-ICU)

  • Recovery metrics (e.g., time to emergence, responsiveness)

  • Postoperative outcomes including length of stay and readmission

• Derived EEG features

  • Spectral power summaries in dB (e.g., delta and alpha bands)

  • Burst suppression indicators

  • Emergence trajectory classifications describing transitions between brain states

EEG Data Availability

Continuous frontal EEG was recorded intraoperatively using clinical monitoring systems and processed to a standardized sampling frequency of 100 Hz to ensure consistency across sites.

This PhysioNet release includes summarized EEG features only. The raw EEG recordings (EDF format) are available separately upon reasonable request to the study investigators, subject to data use agreements and ethical considerations.

For a more detailed description of the data, please see the README file, and for detailed information about missing data for each variable, please see the data_dictionary_annotated.csv file.

Usage Notes

This dataset contains processed intraoperative frontal EEG data for secondary analysis of anaesthetic brain states, emergence trajectories, and postoperative delirium risk. The data are intended for feature-based analysis, statistical modeling, and method development rather than reconstruction of the original raw EEG acquisition pipeline.

External documentation pages

Primary documentation is provided in the project README and accompanying variable descriptions. The related source publication should also be cited when using the dataset.

How has this data already been used?

This dataset is derived from the parent study published in [1].

Data were collected at four sites (Atlanta VA Medical Center, Grady Memorial Hospital, Emory University Hospital Midtown, and Waikato Hospital, New Zealand) by the ACCESS (Anesthesiologists Concerned with Cognition, Emergence, Sleep, and Sedation) research consortium [3]; Hesse et al. is the primary analysis of this collaborative group and portions of this database are freely available for secondary analysis.

This PhysioNet release (646 eligible patients, with the superset of a 626-patient cohort with delirium-outcome-complete cohort and 20 missing delirium outcome) also supports a secondary analysis examining whether intraoperative frontal EEG alpha power, measured during the maintenance phase of general anesthesia, predicts postoperative (PACU) delirium.

Potential future uses

Development or external validation of postoperative delirium risk-prediction models that combine demographic, comorbidity, anaesthetic, and EEG-derived features.

Secondary analysis of EEG biomarkers of emergence and maintenance (e.g. spectral edge frequency, alpha peak frequency, burst suppression) independent of the delirium outcome.

Epidemiological study of burst suppression occurrence and its relationship to age, comorbidity burden, and anaesthetic technique.

Comparative analysis of EEG spectral power across surgical disciplines, anaesthetic maintenance techniques (volatile gas vs. TIVA), or study sites.

Benchmark dataset for tabular machine-learning methods handling a mix of binary comorbidity flags, categorical variables, and continuous EEG features with substantial variable-dependent missingness.

Meta-analysis in combination with other perioperative delirium or intraoperative EEG datasets.

Known limitations

Site/generalizability: Data come from four sites of different types (a VA medical center, a safety-net county hospital, an academic university hospital, and an international site in New Zealand — see the parent publication for site-level detail); case mix, anaesthetic practice, and delirium-assessment procedures may not generalise beyond these settings. The site column in this release is a de-identified numeric code and is intentionally not mapped to the named institutions here, to avoid row-level linkage between individual patients and a specific site.

Derived EEG features only: The dataset contains spectral/summary EEG features (e.g. band power, spectral edge frequency, burst-suppression indices), not raw EEG waveforms. Re-derivation with different signal-processing choices (e.g. different frequency-band edges) is not possible from this release.

Variable-dependent missingness: Missingness ranges from <1% for basic demographics to over 30% for some sleep-history and phase-specific EEG variables. See the Value counts column in data_dictionary_annotated.csv for the exact missingness of every variable before use; no imputed values are included in this release, so users must apply their own missing-data strategy.

Primary outcome missingness: The delirium outcome (DELIRIUMtf_NaN) is missing for 20 of 646 patients (3.1%). The associated study's own primary analysis further restricted to 502 patients with complete outcome, age, and alpha-power data; outcome-based reuse should expect similar attrition.

One top-coded age value: One patient's age is recorded as the text value 90+ rather than a number. Analyses treating age as continuous must handle this value explicitly (e.g. exclude, or bin all ages into ranges).

Retrospective, secondary-analysis design: This is a secondary analysis of prospectively collected clinical data, not data collected to answer a pre-registered question about EEG and delirium; anaesthetic and surgical practice patterns reflect the study sites' practice at the time of data collection and may not reflect current practice elsewhere.

Related software

No dataset-specific software is required. The data can be analyzed using standard tools such as Python, MATLAB, or R.

Special software required

No special software is required beyond standard scientific computing tools capable of reading the released file formats.

Release Notes

Version 1.0.0

Initial Release

Ethics

The study protocol was approved by local Ethics or Institutional Review Boards (both at Emory University (CR3_IRB00063456) and Waikato Hospital (Ref. 12/CEN/56). A written informed consent was obtained from each patient.

Acknowledgements

P.S.G. research efforts are supported by the James S. McDonnell Foundation (St. Louis, Missouri, USA) grant number: 220023046 [4].

Conflicts of Interest

Paul S Garcia is named inventor on several patents owned by Columbia University and TUM related to human EEG. P.S.G. is also a co-founder of Lantern Laboratory, Inc., a company that is licensed by Columbia University to develop EEG technology. None of these disclosures have any financial relevancy to the work presented.

References

1. Hesse S, Kreuzer M, Hight D, et al. Association of electroencephalogram trajectories during emergence from anaesthesia with delirium in the post-anaesthesia care unit. Br J Anaesth. 2018.
2. Hight DF, Gaskell AL, Kreuzer M, et al. Transient electroencephalographic alpha power loss during maintenance of general anaesthesia. Br J Anaesth. 2019.
3. AccessHQ [Internet]. [Accessed September 14, 2026]. Available from: https://www.accesshq.org
4. Jackson S. Moulton Foundation [Internet]. [Accessed September 14, 2026]. Available from: https://www.jsmf.org.