Clinical-record models show modest discrimination of cardiac events after checkpoint therapy
Source check: Source checked. Not independent scientific confirmation.
Finding: Elastic-net regression, gradient-boosted trees and random forests showed similar exploratory discrimination of coded cardiac events after immune checkpoint therapy, with test AUCs around 0.72. Positive predictive values remained below 5%. The selected retrospective cohort and outcome-informed preprocessing limit interpretation as a ready-to-use baseline risk tool.
Technology: Elastic-net logistic regression; gradient-boosted trees; random forest
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Learned regression/tree models mapped recorded baseline characteristics to coded 90-day events; comparisons measured their modest retrospective prognostic discrimination and false-positive limitations.
- Test and result
- TriNetX retrospective cohort of 61,117 patients receiving checkpoint therapy in 2010–2023, including 1,172 coded cardiac inflammatory events within 90 days. A stratified 70:30 patient split yielded 18,335 test patients. Models used baseline demographic, diagnosis and medication features with resampling and cross-validated tuning. Test AUCs were 0.718, 0.718 and 0.719; all models had positive predictive values below 5%.
- Limitation / still unknown
- Diagnostic codes plus hospitalization, not adjudicated drug causation, defined events. Event-free patients needed at least 60 days on therapy, conditioning eligibility on post-baseline exposure. Initial feature prevalence screening used event cases from the entire cohort before splitting, so evaluation is not completely isolated from feature selection. Missing diagnoses/exposures were treated as absent; calibration and rare-event precision were limited. No independent prospective validation, prevention benefit or causal risk factors were 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/PMC13415296/
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 actual primary outcome/exposure definitions, selection window, feature construction, cohort-wide initial screening, patient partition, named model/tuning Methods, performance Results and limitations. Supplementary modeling tables were not independently inspected. Exact primary identity/date and fresh primary PubMed correction/retraction relationships 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.