An EEG model retrospectively estimates outcomes across critical-illness etiologies
Source check: Source checked. Not independent scientific confirmation.
Finding: A one-dimensional convolutional network analyzed EEGs from 358 adults with acute consciousness impairment. Patient-held-out cross-validation yielded AUCs of .721 for six-month survival and .703 for favorable functional outcome. Higher-confidence subsets had better performance at reduced coverage; this was retrospective modeling, not a randomized test of AI-guided care.
Technology: t-VGG GAP one-dimensional convolutional neural network
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- The CNN learned EEG features directly to classify clinical outcomes and etiologies; Grad-CAM and visual review explored model behavior without proving a new causal mechanism.
- Test and result
- This post-hoc analysis reused the four-hospital CERTA EEG trial. A fixed previously published CNN architecture used overlapping epochs, while stratified five-fold splitting kept all epochs from a patient together; patient predictions averaged epoch outputs. Prognosis used survival at six months and the best Cerebral Performance Category measured at four weeks or six months. All 358 patients contributed prognostic evaluation; diagnostic etiology classification used 277 eligible patients and reached 54.5% accuracy.
- Limitation / still unknown
- Original trial randomization concerned EEG recording strategy, not this learned model. The best-of-two-timepoints functional endpoint is not a fixed six-month recovery measure. Model-output confidence selected progressively smaller subgroups and does not establish calibrated clinical certainty; higher-confidence performance must not be reported as whole-cohort accuracy. Earlier random-forest work used the same cohort, so this is not an independent replication of its findings. Treatment decisions and withdrawal of life support may affect observed prognosis; authors argue that using the first EEG reduced but did not remove that concern. There was no held-out new hospital, prospective model-guided treatment or demonstrated patient benefit. The represented date is the documented August 2022 online journal version, without an earliest-priority assertion.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC9441331/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: Original publication/affiliation metadata, computational Methods2.1–2.4, clinical endpoints and Results3.2/Table2/Table3; bounded Discussion limitations actually read..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 1/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.