AI-assisted CT risk score retains mortality discrimination in a later COVID-19 cohort
Source check: Source checked. Not independent scientific confirmation.
Finding: A five-variable model combining demographic data, oxygen saturation and automatically extracted CT measurements distinguished recorded mortality in a later COVID-19 cohort. Developed from 1,125 first-wave patients, AI-SCoRE achieved a reported AUC of 0.808 in 213 second-wave patients. This supports a bounded clinical prognostic validation, without demonstrated treatment benefit or proof that the predictor was locked before validation outcomes accrued.
Technology: Pretrained U-Net lung segmentation with quantitative CT feature extraction, logistic risk modeling and CART risk binning
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Pretrained neural segmentation converted patient CT scans into quantitative lung features used with clinical variables in the represented mortality score; learned risk modeling and binning enabled the patient-outcome discrimination analysis.
- Test and result
- A multicenter first-wave dataset was reduced from 1,575 enrolled patients to 1,125 complete records, including 292 deaths. Repeated train/test splits and repeated cross-validation compared candidate models and selected a five-variable logistic model. Pretrained learned segmentation enabled automated lung measurements. A later, prospectively collected single-center cohort of 213 patients, including 31 deaths, yielded AUC 0.808 with reported 95% confidence interval 0.740–0.877.
- Limitation / still unknown
- Prospective cohort collection does not establish that the model and thresholds were fixed before its outcomes were known. Complete-case selection, different pandemic waves, one-center later validation and historical treatment conditions limit transportability. The paper inconsistently describes 213 versus 214 validation patients and presents four low-risk deaths as 1.8% without a clearly appropriate low-risk denominator; those percentages and post-hoc exclusions are not adopted here. Supplemental acquisition/endpoint details were not independently read. No causal CT mechanism, intervention benefit or current clinical utility is established.
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/PMC9423702/
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: Read primary front identity/date, clinical study design, modeling and segmentation Methods, cohort/model/later-wave mortality Results and limitations. Supplemental Methods were not independently read. Later cohort collection was prospective; temporal predictor locking was not established. Exact original article front DOI/title and fresh PubMed primary relationship metadata 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.