Continuous EEG monitoring enhances multimodal outcome prediction in hypoxic-ischemic brain injury

Abstract

Objective. Hypoxic brain injury is the largest contributor to disability and mortality after cardiac arrest. We aimed to identify EEG characteristics that predict outcome in cardiac arrest patients treated with targeted temperature management (TTM).

Methods. We retrospectively examined clinical data, EEG, functional outcome at discharge, and in-hospital mortality for 373 adults with return of spontaneous circulation after cardiac arrest. Poor outcome was a Cerebral Performance Category of 3–5. LASSO logistic-regression models were fit with and without EEG reactivity/background features.

Results. In-hospital mortality was 68.6% (N=256). An unreactive EEG background together with status epilepticus gave a positive predictive value of 100% (95% CI 0.96–1) and false-positive rate of 0% (95% CI 0–0.11) for poor outcome. A model with demographics, admission exam, status epilepticus, pure suppression-burst, and lack of EEG reactivity achieved AUC 0.92 (0.87–0.95) for poor functional outcome and 0.96 (0.94–0.98) for in-hospital mortality. Pure suppression-burst was confounded by anesthetic use (83.9%) and was not an independent predictor.

Conclusions. An unreactive EEG background and status epilepticus predicted poor functional outcome and in-hospital mortality after cardiac arrest under TTM; the prognostic value of pure suppression-burst is confounded by sedation.

This project provides the MATLAB code and de-identified data to reproduce the analyses.

---

Background

Hypoxic-ischemic brain injury is the leading cause of death and disability after cardiac arrest. This project releases the code and de-identified data for models that predict functional outcome, in-hospital mortality, and discharge disposition, and quantify how continuous-EEG features add to clinical predictors.

---

Software Description

MATLAB source and de-identified data. FULLforLASSO6_deID.mat and EdDataTable2.mat are the modeling tables (surrogate SID, no MRNs/names); MECA_FULL_*.xlsx are the published summary tables. The LASSO scripts (a_LassoCode_*_V3.m) fit models for poor outcome, mortality, and disposition.

---

Technical Implementation

We retrospectively examined clinical data, EEG, functional outcome at discharge, and in-hospital mortality for 373 adults with return of spontaneous circulation after cardiac arrest treated with targeted temperature management. Poor outcome was a Cerebral Performance Category of 3-5. LASSO logistic-regression models were fit for each outcome with and without EEG reactivity/background features; discrimination was assessed by AUC and net reclassification improvement.

---

Installation and Requirements

MATLAB R2016 or later with the Statistics and Machine Learning Toolbox. No installation beyond cloning. Run run_all from the repo root to reproduce the LASSO AUCs. See REPRODUCE.md and DATA_SOURCE.md.

---

Usage Notes

Reproduce with run_all (MATLAB). REPRODUCE.md maps each table/figure to its script and committed input; DATA_SOURCE.md documents provenance and de-identification (the de-identified data reproduce the published AUCs, 0.925 poor-outcome and 0.965 mortality).

---

Release Notes

First public release: MATLAB code + de-identified data + one-command reproduction.

---

Ethics

Data were collected under IRB approval at the University of Pittsburgh; all released data are de-identified.

---

Acknowledgements

Data from the University of Pittsburgh Post Cardiac Arrest Service.

---

Conflicts of Interest

See the associated publication (Resuscitation 2016;109:121-126) for author disclosures.

---