A laboratory-data model classifies retrospective COVID-19 severity labels
Source check: Source checked. Not independent scientific confirmation.
Finding: Tuned logistic regression in 8,844 repeated blood tests from 2,935 unvaccinated patients estimated retrospective COVID-19 severity labels, reporting an AUC of 0.8512. Tuning raised sensitivity from 36.59% to 78.05% while lowering specificity from 96.38% to 77.00%. This internal test-level evaluation does not establish early prognosis for new patients.
Technology: Logistic regression
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Logistic regression fitted the represented classification or candidate-selection model; preprocessing and expert reference labels are distinguished from the learned contribution.
- Test and result
- Five-fold stratified evaluation split blood tests, approximately 7,075 for training and 1,769 for testing per fold; model and scaler combinations were compared and logistic regression was tuned.
- Limitation / still unknown
- Patient grouping across folds is not established. Availability-based feature selection and model tuning are not shown to be nested within independent patient-level validation; multiple tests per patient can inflate apparent generalization. Samples after ICU admission were excluded; no prospective benefit or new biomarker mechanism is established. Model/scaler selection used evaluation folds. A figure caption and cohort breakdown describe the nonsevere group differently; no conclusion about hospitalization status is drawn. The operating-point tradeoff is distinct from the small AUC change from 0.8418 to 0.8512.
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/PMC10731083/
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: Primary article full-text bibliographic metadata and relevant computational evaluation Methods/Results: se0030, se0080, se0120, se0150, se0160. Primary-source caveat and denominator review; candidate-panel identity corrected where relevant..
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.