Derivation and validation of an artificial intelligence-based plaque burden safety cut-off for long-term acute coronary syndrome from coronary computed tomography angiography.

Summary

Across two registries with median ~7-year follow-up, an AI-quantified percent atheroma volume cut-off of 2.6% achieved ≥90% sensitivity and 99% NPV for future ACS, identifying a large subgroup with low near-term risk. Patients above the threshold had substantially higher adjusted ACS rates.

Key Findings

• Derivation cohort (n=2,271): PAV ≥2.6% yielded 90.0% sensitivity and 99.0% NPV for future ACS over median 6.9 years.
• External validation (n=568): PAV ≥2.6% achieved 92.6% sensitivity and 99.0% NPV over median 6.7 years.
• Patients with PAV ≥2.6% had higher adjusted ACS risk (HR 4.65 derivation; HR 7.31 validation).
• A large fraction had PAV <2.6% (45.2% derivation; 34.3% validation), enabling low-risk identification beyond 'no-plaque' status.

Clinical Implications

CCTA reports could include AI-PAV with a validated safety threshold (2.6%) to reassure low-risk patients and prioritize preventive intensification and surveillance in those above the threshold.

Limitations

• Observational registry design; clinical actionability beyond risk labeling needs prospective testing
• Generalizability to other scanners, AI tools, and diverse populations warrants evaluation

Future Directions

Prospective trials to embed AI-PAV thresholds into care pathways (triage, therapy intensification), multi-vendor validation, and cost-effectiveness analyses.