Networked Radiology Imaging and Treatment Planning Systems and FDA Cybersecurity Regulation: A Difference-in-Differences Analysis of 510(k) Clearance Delays.

Authors

Chuelwon Lee

Published in

Journal of imaging informatics in medicine. Sep 08, 2026. Epub Sep 08, 2026.

Abstract

The objective of this study is to determine whether 510(k) clearance duration for AI-enabled devices changed differentially relative to the general market around October 1, 2023-the date FDA began Refuse-to-Accept enforcement of Section 524B cybersecurity documentation completeness-using a freshly re-extracted, complete FDA 510(k) dataset. We analyzed 83,675 FDA 510(k) records (De Novo excluded; decisions dated January 2000 through August 2, 2026). AI-enabled codes were identified via FDA's officially published AI-Enabled Medical Device List (26 codes, N = 7070). Log-transformed clearance duration was modeled with a difference-in-differences (DiD) regression (product code-clustered standard errors, adjusted for submission type and review pathway); a half-year event study specification tested parallel pre-rule trends and timing, and Bonferroni-corrected Mann-Whitney tests (15 comparisons) assessed code-level heterogeneity; AI-enabled status was assigned at the product code level and was not confirmed at the individual-submission level. After adjustment, the overall market showed no significant post-rule change (Post coefficient = 0.029, p = 0.141), while the AI-enabled cohort showed a robust increase (AI × Post = 0.212, p < 0.001; ≈24% longer). The two half-years nearest the cutoff showed no significant pre-rule trend, though the most distant pre-rule half-year (six half-years prior) showed a significant negative coefficient, an anomaly discussed in Limitations; the AI x Post effect itself became significant approximately 24 months post-implementation. Five codes-diagnostic ultrasound (IYN), CT scanners (JAK), radiology image processing software (LLZ), angiography/fluoroscopy systems (OWB), and radiotherapy-planning software (MUJ)-were Bonferroni significant (all p < 0.001) and together drove the aggregate effect; all five are radiology panel codes. The highest-volume AI code (QIH) showed no significant change after correction. The observed clearance time increase for AI-enabled submissions is associated with, and concentrated in, networked radiology panel imaging and treatment planning systems rather than AI-enabled software broadly, and it accumulates over roughly 6 to 24 months rather than appearing immediately. Because AI-enabled status is defined at the product code level, submissions in the control group may include non-AI cyber devices also subject to the cybersecurity rule, and the concurrent mandatory eSTAR transition is not separately identified; these findings should therefore be read as an association warranting device category-specific, rather than blanket AI/ML, regulatory attention, not as established evidence that Section 524B specifically caused the observed delay.

PMID:
42711639
Bibliographic data and abstract were imported from PubMed on 09 Sep 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.