Study: AI-Assisted Device Reviews Could Reduce Recalls
A machine learning-based approach to FDA's 510(k) medical device clearance process could reduce device recalls by nearly one-third while cutting the agency's regulatory workload by more than 40%, according to a study published in Management Science. The study evaluated a proposed system that combines machine learning with human regulatory review to identify devices at greater risk of recall before they receive FDA clearance.
Researchers from Indiana University, Harvard University and Emerging Health Consulting developed algorithms to estimate recall risk using information available when manufacturers submit 510(k) applications. Their proposed framework would recommend clearance, rejection or referral for more extensive evaluation by FDA review committees.
The researchers analyzed more than 31,000 medical device submissions using data assembled from FDA and the Centers for Medicare and Medicaid Services. Compared with FDA's existing clearance practices, which the researchers associated with a 10.3% recall rate, their proposed approach produced an estimated 32.9% improvement in the recall rate and a 40.5% reduction in regulatory workload.
The findings suggest that machine learning could help FDA allocate review resources more efficiently by identifying lower-risk submissions while directing greater regulatory scrutiny toward devices with elevated recall risks.
The proposed system is a research-based approach and would require further evaluation before they could be established in routine regulatory practice, the authors note.