Pilot case study · VigInfo × AKO MED
Every Record, Expert-Verified
AI-assisted literature surveillance with 100% human-in-the-loop review: an AKO MED case study on a Class III surgical suture.
24/25
VigInfo-assisted safety-relevant findings
22/25 in the original manual review
88.2 h
estimated active-specialist time saved
86.0% reduction: 102.5 h manual, 14.3 h VigInfo-assisted
100%
platform record dispositions human-confirmed
including exclusions and no-relevant-finding dispositions
From a paired review of 442 unique records. The 25-item reference set is the adjudicated union of findings from both review approaches, not independent ground truth; specialist times are reviewer estimates. The paper sets out the method and its limitations in full.
What’s inside
- 01Pilot design and scope. The JOST PGA Surgical Suture, PubMed, PubMed Central, FDA MAUDE and one subscription-based biomedical literature database, 574 source records deduplicated to 442 unique canonical records.
- 02Safety-relevant findings. 21 identified by both methods, 3 VigInfo-assisted only, 1 manual review only, and what the discordant findings showed.
- 03Positive performance findings. 30 confirmed clinical performance results; all 29 original manual findings retained.
- 04IMDRF coding. Exact Annex F decision agreement in 17 of 24 cases (70.8%), with close misses and clear mismatches reported.
- 05Active-specialist time. How 102.5 h manual and 14.3 h VigInfo-assisted were reconstructed, and what was excluded.
- 06Auditability, limitations and appendices. The decision lineage retained per record, the limitations of the design, and every calculation behind the figures.