A retrospective study evaluated MedJev for in-hospital mortality prediction
Source check: Source checked. Not independent scientific confirmation.
Finding: Using records available 24 hours after admission, MedJev ranked eventual in-hospital deaths with an AUROC of 0.926 (95% confidence interval 0.881 to 0.960). The supervised XGBoost comparison scored 0.902, with a paired interval that included no difference. MedJev’s unadjusted risk estimates were substantially miscalibrated.
Technology: MedJev, based on Qwen3.5-9B; Platt scaling
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- MedJev converted patient-history events available at the prediction time into outcome-token probabilities, which were evaluated against recorded deaths.
- Test and result
- Adult patients still hospitalized at 24 hours were eligible. Evaluation examples and supervised-model/recalibration fitting used the source-described patient-disjoint MIMIC-IV MEDS partitions. Recalibration used separately partitioned labeled records and improved calibration without changing AUROC.
- Limitation / still unknown
- The retrospective single-center data do not establish external performance or patient benefit. MedJev was first adapted on MIMIC-IV Note, but the primary Methods do not explicitly establish that this adaptation excluded the evaluated patients or their later notes. The described MEDS evaluation split therefore does not authenticate checkpoint-level patient independence or absence of note-derived outcome exposure. The discrimination comparison was against supervised models at the tested local-label budget, not unlimited supervised training.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2610.36006v1 · https://arxiv.org/html/2610.36006v1
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: §§3.1–3.4 and 4.2–4.3, Tables 1 and 3–4 and front matter.
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.