The abstract is read at the publisher; the summary is JClub's.
Across 14,736 CTPA exams in a six-site health system, the Aidoc BriefCase PE triage model achieved 64.1% sensitivity and 98.8% specificity, with sensitivity varying by 13 points between sites.
- Why it matters: Post-market performance for an FDA-cleared PE triage tool has lacked subgroup, site, and temporal stability analyses, limiting confidence in real-world monitoring.
- What they did: In a retrospective study (July 2023–March 2026), radiology-report PE labels were extracted using a zero-shot LLM pipeline validated on 600 reports (kappa=0.98), then model performance was assessed by demographics, embolus features, site, and time using logistic and segmented regression.
- The result: Sensitivity dropped by embolus peripherality (central 88.6% to subsegmental 33.3%) and chronicity (acute 65.5% to chronic 57.4%), fell between two high-volume sites (68.7% vs 55.7%, P<.001), and showed temporal decline at site A (72.5% to 65.0%), while false-positive rate decreased 40% per year (P<.001).
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.