Aortic & peripheral vascular diseaseDiagnostic or prognostic study
Adaptive Artificial Intelligence (AI) for Cardiovascular Event Adjudication: Cardiovascular Event Adjudication Across Different Definitions in the ODYSSEY OUTCOMES and EUCLID Trials
- Diagnostic accuracy study
- external validation
Authors' conclusion
After brief adaptation, a single trial-derived AI algorithm can adjudicate similar (MI and stroke) and new end points (CV death and bleeding) in a second trial and replicate the EUCLID primary outcome treatment effect. A hybrid approach with humans adjudicating those suspected events with the lowest 30% of ADAPT-CEC prediction certainty was superior to ADAPT-CEC alone or GPT 4o alone and replicated the EUCLID primary outcome treatment effect.
Study context
- Population
- Suspected cardiovascular (CV) end point events in the EUCLID trial (PAD)
- Participants
- 13,885 suspected events
- Comparison
- ADAPT-CEC vs hybrid vs Generative Pretrained Transformer (GPT) 4o adjudication, vs human adjudication
Performance
As reported in the abstract
Classification vs human adjudication
| Test or model | End points correctly classified | Non-end points correctly classified | F1 bleeding | F1 CV death | F1 myocardial infarction (MI) | F1 stroke |
|---|---|---|---|---|---|---|
| ADAPT-CEC | 86.4% | 99.4% | 0.78 [95% CI, 0.77-0.79] | |||
| Hybrid | 95.6% | 99.6% | 0.83 [95% CI, 0.82-0.85] | 0.94 [95% CI, 0.92-0.96] | 0.80 [95% CI, 0.77-0.82] | 0.82 [95% CI, 0.78-0.86] |
| GPT 4o | 76.3% | 99.8% |
Further findings
| Outcome | HR (95% CI) | |
|---|---|---|
| EUCLID primary treatment effect (CV death, MI or stroke) by adjudication strategy — Human adjudication vs Human adjudication | 1.02 (0.93–1.13) | |
| EUCLID primary treatment effect (CV death, MI or stroke) by adjudication strategy — Hybrid vs Human adjudication | 1.04 (0.94–1.15) | |
| EUCLID primary treatment effect (CV death, MI or stroke) by adjudication strategy — ADAPT-CEC vs Human adjudication | 0.98 (0.88–1.09) | |
| EUCLID primary treatment effect (CV death, MI or stroke) by adjudication strategy — GPT 4o vs Human adjudication · printed as GPT 4.0 | 1.06 (0.95–1.19) | |
| 0.512Hazard ratio (log scale) |
EUCLID primary treatment effect (CV death, MI or stroke) by adjudication strategy
- Treatment effect similar across strategies
Original publication
Adaptive AI for Cardiovascular Event Adjudication: Cardiovascular Event Adjudication Across Different Definitions in the ODYSSEY OUTCOMES and EUCLID Trials. Circulation. Published 30 March 2026. doi:10.1161/circulationaha.126.080072
Summarised from the published abstract; every number is checked against it