Time-series model tests sepsis prediction with an indeterminate option
Source check: Source checked. Not independent scientific confirmation.
Finding: An attention-based LSTM evaluated hospital-record sepsis labels retrospectively. External-dataset AUC was 0.971 six hours before recorded onset; withholding uncertain classifications reduced false positives. This does not establish improved care or prospective non-ICU performance.
Technology: Attention-based long short-term memory neural network; conformal prediction
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- The LSTM learns laboratory trajectories; conformal sets flag unfamiliar inputs.
- Test and result
- MIMIC-IV development and eICU-CRD retrospective evaluation. The November2024 preprint already reports this result; peer-reviewed version followed in May2025.
- Limitation / still unknown
- External labels used diagnosis strings rather than Sepsis-3. About 17% were indeterminate. Missingness eligibility used the six-hour window even for earlier forecasts. ICU-onset cases were included; preprocessing isolation is incompletely documented. No measured workflow benefit.
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/PMC11601686/ · https://pmc.ncbi.nlm.nih.gov/articles/PMC12174992/
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: Exact primary metadata/version relationship; indexed primary clinical/model Methods, Results and limitations..
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.