Prognostic significance of EEG patterns in post-cardiac arrest coma: a 1,000-patient multicenter cohort

Abstract

Objective. To quantify the prognostic significance of continuous-EEG (cEEG) patterns for outcome in comatose survivors of cardiac arrest, and how that prognostic value evolves over the first 72 hours, in a large multicenter cohort.

Methods. Retrospective multicenter cohort of 1,000 comatose post-cardiac-arrest patients undergoing cEEG. For each EEG pattern and time window (24, 48, 72 hours, and the remainder), true-positive and false-positive rates and predictive values for poor outcome (Cerebral Performance Category at discharge) were computed with exact Clopper-Pearson 95% confidence intervals, and temporal trends were assessed by linear regression (significance at α=0.05).

Results. The dataset and code reproduce the per-pattern prognostic statistics and their temporal trends, identifying which EEG patterns carry reliable (low false-positive) prognostic information and how this changes over time.

This project provides the de-identified 1,000-patient dataset (linked to BDSP patient IDs) and the analysis code.

---

Background

Continuous EEG is widely used for neuroprognostication after cardiac arrest, but the reliability (false-positive rate) of individual EEG patterns for predicting poor outcome, and how it changes over time, is not well quantified. This project releases the data and code from a 1,000-patient multicenter study addressing this.

---

Software Description

De-identified REDCap dataset (CardiacArrestPrognos_DATA_deidentified.csv; 1,000 x 191; BDSP Patient ID linkage; no names/MRNs/absolute dates) + Python/Jupyter analysis (a0_CalculateStats_v3.ipynb, demographics_code.ipynb) + derived result tables and figures.

---

Technical Implementation

Retrospective multicenter cohort of 1,000 comatose post-cardiac-arrest patients with cEEG. Per EEG pattern and time window, true/false-positive rates and predictive values for poor outcome (CPC at discharge) were computed with exact Clopper-Pearson 95% CIs; temporal trends assessed by linear regression at alpha=0.05.

---

Installation and Requirements

Python 3.9+ with pandas, numpy, scipy, statsmodels, matplotlib, seaborn (and polars for the demographics notebook). Run the notebooks from the repo root. See REPRODUCE.md and DATA_SOURCE.md.

---

Usage Notes

a0_CalculateStats_v3.ipynb reproduces the prognostic-significance tables (exact Clopper-Pearson CIs) and temporal trends from the committed de-identified CSV; demographics_code.ipynb reproduces the demographic tables. Dates are per-patient shifted (intervals preserved).

---

Release Notes

First public release: de-identified 1,000-patient dataset + analysis code + reproduction docs.

---

Ethics

Data collected under IRB approval at participating centers; released de-identified (identifiers removed, dates shifted, BDSP IDs retained).

---

Acknowledgements

The authors thank the participating centers and the BDSP.

---

Conflicts of Interest

See the associated publication (Neurology, accepted; WN9-2025-200289).

---

References

1. Dey RK, Yaramis A, Turley N, Mullan H, Chougule S, Gallagher K, Bayas G, Barot N, Drislane FW, Shafi MM, Pang T, Goldenholz DM, Struck AF, Kim JA, Zafar SF, Lee JW, Amorim E, Westover MB. Prognostic Significance of EEG Patterns in Post-Cardiac Arrest Coma: A Multicenter Retrospective Cohort Study of 1,000 Patients. Neurology. In press (accepted 2026; article WN9-2025-200289).

---