Machine-learning models estimate one-year target-lesion failure risk after coronary intervention
Source check: Source checked. Not independent scientific confirmation.
Finding: Five models were trained on 35,389 patients in the global e-ULTIMASTER percutaneous coronary intervention registry. On a random 20% test split, gradient boosting had an AUC of 0.72 (95% CI 0.69–0.75) for one-year target-lesion failure; component AUCs were 0.82 for cardiac death, 0.75 for target-vessel myocardial infarction, and 0.68 for clinically driven revascularization. The study lacks external validation outside this registry and calls for prospective evaluation; it does not establish clinical benefit.
Technology: Gradient boosting, linear discriminant analysis, naive Bayes, logistic regression, and k-nearest neighbors
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Learned risk models estimate one-year target-lesion failure and its component outcomes from patient and procedural characteristics.
- Test and result
- Five risk models were evaluated on a held-out random test subset of the e-ULTIMASTER registry, with discrimination and calibration assessed.
- Limitation / still unknown
- The authors report no external validation outside the registry and state that prospective evaluation is warranted; clinical utility is unestablished.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://academic.oup.com/ehjdh/article/4/6/433/7257453
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: 22 Nov 2026. Review scope: Full primary publisher article, including Methods, Results, author affiliations and publication history.
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 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.