Learning models show no clear gain in hemorrhage outcome prediction
Source check: Source checked. Not independent scientific confirmation.
Finding: In 388 patients with aneurysmal subarachnoid hemorrhage, tree boosting, neural networks and other models did not clearly outperform traditional models using admission clinical and imaging information. The best reported AUC was 0.78 for combined features, versus 0.76 using Glasgow Coma Scale information alone. This is a retrospective clinical comparison, not proof of equivalence or evidence that learning cannot help other cohorts or inputs. Single-institution sampling and incomplete outcome records limit generalization.
Technology: CatBoost tree boosting; multilayer perceptron; support-vector machine; generalized linear models
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Boosted trees, a multilayer perceptron and other fitted classifiers learn admission-feature/outcome relationships; their held-out clinical performance is compared with regression and models using established severity scores.
- Test and result
- Retrospective admission-feature prediction against later favorable/unfavorable modified Rankin outcomes. Patient-level 80:20 splits repeated fifty times, with ten-fold tuning and training-set imputation/scaling; 388 analyzed after twenty patients lacking outcomes were excluded.
- Limitation / still unknown
- Generalization to other centers, incremental value of different inputs and clinical decision benefit remain unestablished. Absence of a clear performance advantage in this cohort is not formal statistical equivalence, and no intervention allocation was tested.
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/PMC8490233/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 21 Nov 2026. Review scope: Primary relevant patient/reference, predictor, preprocessing and model-validation Methods, Results and limitations; supplementary detailed metric tables not independently reviewed. Exact primary identity/date and fresh primary PubMed correction/retraction relationships checked..
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 2/5 · 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.