Regulatory time-lag between CE-marking and FDA approval of artificial intelligence softwares
G Siontis, D Windecker, A Fraser, E Caiani, Y RenAbstract
Background
The translation of artificial intelligence (AI) technologies into clinical practice is strongly influenced by regulatory pathways. It remains unclear whether previously observed authorization gaps between Europe and the US persist under the EU Medical Device Regulation (MDR), and whether the timing of a second approval varies depending on the approval sequence.
Purpose
To examine the time difference between CE-marking and FDA clearance for AI-enabled medical devices in radiology under the contemporary MDR framework.
Methods
We performed a cross-sectional analysis of AI-enabled medical devices with CE-mark and/or FDA clearance. Device information was collected via web scraping from a publicly available database (Health AI Register), complemented by additional data from FDA website. We calculated the corresponding time-lags for each device as the difference in months between the two approval dates. Kaplan-Meier methods were used to compare time-to-second-approval before versus after MDR implementation (05/2021) and by regulatory approval sequence. Multivariable associations with time to second authorization were assessed using Cox proportional hazards regression.
Results
In total, 239 devices were included: 128 (54%) obtained CE marking only, 95 (40%) received CE marking followed by FDA clearance, and 16 (7%) received FDA clearance followed by CE marking. The pooled median time to second approval across all devices was 54 months (Fig. Panel A). The time to second approval did not differ between devices approved before and after MDR implementation (log-rank p=0.17) (Fig. Panel B). It differed significantly by regulatory approval sequence (p<0.01) (Fig. panel C), with longer absolute delays for devices receiving CE marking first compared with those receiving FDA clearance first (median 17.5 vs 3.5 months, p<0.01). In the multivariable Cox regression analysis stratified by regulatory approval sequence, compared with computed tomography related AI-enabled medical devices, conventional X-ray related MDAI had a lower hazard of obtaining a second approval (hazard ratio 0.46; 95%CI 0.25–0.82), indicating slower progression to second authorization. Compared to EU Class I devices, EU Class IIa devices exhibited a higher hazard of second approval (2.91; 95%CI 1.52–5.56). Devices for which clinical evidence was available in a peer-reviewed publication were more likely to have been approved by both regulatory agencies, with a higher early hazard of second approval (6.08; 95%CI 3.04–12.19) that attenuated over time.
Conclusions
Devices initially approved in the EU had longer time lags before subsequent approval in the US compared with devices first approved in the US. Efforts to streamline and harmonize regulatory processes can facilitate timely and equitable deployment of AI-enabled medical technologies worldwide.Fig.