Abstract
This cross-sectional study describes predetermined change control plan use among artificial intelligence (AI)–enabled medical devices and the comprehensiveness of public reporting documenting predetermined change control plans.
Introduction
Artificial intelligence (AI) can improve existing medical devices by dynamically modifying functionalities in response to real-world data.1 However, this iterative process exists in tension with the current approach by the US Food and Drug Administration (FDA) to authorize devices, which generally requires manufacturers to resubmit devices for review when modifications affect safety or effectiveness.2
To reduce this tension, the FDA drafted guidance in April 2023 and finalized guidance in December 2024 clarifying their expectations regarding predetermined change control plans (PCCPs) specifically for AI-enabled devices.3 Within PCCPs, which were first statutorily permitted for devices generally in December 2022,3 manufacturers prespecify device modifications and methodologies for evaluating them. If authorized, manufacturers can then implement these modifications without additional FDA review, even when modifications affect safety or effectiveness. This cross-sectional study describes PCCP use among AI-enabled medical devices and the comprehensiveness of public reporting documenting PCCPs.
Methods
From public FDA databases,4,5 we identified AI-enabled medical devices authorized from 2023 through 2025 and extracted information on device characteristics, including authorization pathway, authorization date, clinical specialty, review time, count of prior AI-enabled devices from the manufacturer (manufacturer experience), and PCCP use. Using public manufacturer summary documents, we manually assessed disclosure of 4 key PCCP elements: (1) planned modifications (modification list), (2) how manufacturers test/validate modifications (modification protocol), (3) how manufacturers communicate modifications to users (notification plan), and (4) risks/benefits of modifications (impact assessment). Additionally, the FDA recommends that manufacturers publicly summarize modification lists, modification protocols, and notification plans (but does not direct manufacturers to summarize impact assessments)3; we assessed whether manufacturers disclosed all recommended elements. Finally, within notification plans, we identified whether manufacturers contacted users through updates to labeling/instructions for use, device interface notifications, direct correspondence, and/or release notes.
Descriptive statistics and 2-sided tests were calculated using R, version 4.4.2 (R Project for Statistical Computing), with statistical significance defined as P < .05. The eMethods in Supplement 1 provide additional details. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. This study used publicly available materials and did not involve humans; therefore, institutional review board review was not required.
Results
Among 794 AI-enabled medical devices, 43 (5.4%) had FDA-authorized PCCPs (Table). Clinical specialty varied by PCCP use, while there were no meaningful differences in authorization pathway, review time, or manufacturer experience. The first PCCP for an AI-enabled device was observed in the third quarter of 2023, with 2 devices (3.5%) with PCCPs that quarter (Figure). By the fourth quarter of 2025, 7 (9.7%) AI-enabled devices included PCCPs.
Table. Characteristics of Artificial Intelligence–Enabled Medical Devices by Predetermined Change Control Plan (PCCP) Authorization.
| Characteristic | No. (%) | P valuea | |
|---|---|---|---|
| PCCP | No PCCP | ||
| Total No. | 43 | 751 | NA |
| No. of unique manufacturers | 33 | 345 | NA |
| Review pathway | |||
| 510(k) | 42 (97.7) | 730 (97.2) | .86 |
| De novo | 1 (2.3) | 16 (2.1) | |
| Premarket approval | 0 | 5 (0.7) | |
| Specialty | |||
| Anesthesiology | 3 (7.0) | 15 (2.0) | .006 |
| Cardiovascular | 9 (20.9) | 53 (7.1) | |
| Gastroenterology-urology | 1 (2.3) | 18 (2.4) | |
| Hematology | 0 | 5 (0.7) | |
| Neurology | 4 (9.3) | 41 (5.5) | |
| Radiology | 25 (58.1) | 589 (78.4) | |
| Otherb | 1 (2.3) | 30 (4.0) | |
| Review time, median (IQR), dc | 150 (118-202) | 145 (93-216) | .68 |
| Manufacturer experience, median (IQR)d | 2 (1-3.5) | 2 (0-9) | .43 |
| Public elements of PCCPe | |||
| Modification list | 39 (90.7) | NA | NA |
| Modification protocol | 36 (83.7) | NA | NA |
| Notification plan | 32 (74.4) | NA | NA |
| Label/instructions for use | 30 (69.8) | NA | NA |
| Interface | 14 (32.6) | NA | NA |
| Correspondence | 13 (30.2) | NA | NA |
| Release notes | 8 (18.6) | NA | NA |
| ≥2 Methods | 24 (55.8) | NA | NA |
| Impact assessment | 14 (32.6) | NA | NA |
| All FDA-recommended elements | 31 (72.1) | NA | NA |
Abbreviations: FDA, US Food and Drug Administration; NA, not applicable.
P values associated with review pathway and specialty were derived from χ2 test for differences in distributions by presence/absence of an authorized PCCP. P values associated with review time and manufacturer experience were derived from Wilcoxon rank-sum test for difference in median by presence/absence of an authorized PCCP.
The other specialty category includes clinical chemistry, dental, general hospital, general and plastic surgery, microbiology, obstetrics and gynecology, ophthalmic, orthopedic, and pathology.
Review time is defined as the difference between date of application receipt and date of final authorization, as reported in FDA databases.
Manufacturer experience is defined as the number of previously authorized artificial intelligence–enabled medical devices brought to market by a manufacturer before the authorization date for a given device.
See the eMethods in Supplement 1 for details on how PCCP elements were identified.
Figure. Bar Graph Showing the Number of Artificial Intelligence–Enabled Medical Devices by Predetermined Change Control Plan (PCCP) Authorization.
The sample includes 794 artificial intelligence–enabled medical devices authorized from 2023 through 2025. Q indicates quarter.
Reporting of PCCP elements varied: 39 devices (90.7%) reported modification lists, 36 (83.7%) reported modification protocols, 32 (74.4%) reported notification plans, 14 (32.6%) reported impact assessments, and 31 (72.1%) reported all FDA-recommended elements (Table). For devices with public notification plans, 30 (69.8%) notified users via labeling/instructions for use changes, 14 (32.6%) used interface notifications, 13 (30.2%) used direct correspondence, 8 (18.6%) issued release notes, and 24 (55.8%) used 2 or more methods.
Discussion
In this cross-sectional study, only 5% of AI-enabled medical devices had an FDA-authorized PCCP during the study period. However, in contrast to previous work,1,6 we observed a modest increase in PCCP use starting in 2025. Among devices with PCCPs, public reporting of PCCP elements was reasonably high: nearly three-quarters reported all FDA-recommended elements. Additionally, many manufacturers used multiple methods to communicate updates to users.
PCCPs allow manufacturers to quickly implement device modifications while prespecifying plans for mitigating risk.1,6 As such, low PCCP use may represent a missed opportunity for improving device performance. The PCCP authorization process was generally accessible to manufacturers with differing levels of experience; experience with AI-enabled devices was not associated with PCCP use, nor did manufacturers undergo meaningfully different review times when submitting a PCCP. Nonetheless, low overall adoption indicates that there remain unidentified barriers to PCCP use or that certain AI-enabled devices may simply be unamenable to prespecified design changes that preserve safety and effectiveness.
This study was limited by using public summaries, which are less detailed than full regulatory submissions received by the FDA, and the inability to assess the value of PCCPs. Nonetheless, these findings indicate that manufacturers have generally been comprehensive in documenting PCCPs when used. This work supports further efforts by the FDA to engage stakeholders to facilitate appropriate PCCP adoption.
eMethods
eReferences
Data Sharing Statement
References
- 1.Carvalho E, Mascarenhas M, Pinheiro F, et al. Predetermined change control plans: guiding principles for advancing safe, effective, and high-quality AI-ML technologies. JMIR AI. 2025;4:e76854. doi: 10.2196/76854 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Darrow JJ, Avorn J, Kesselheim AS. FDA regulation and approval of medical devices: 1976-2020. JAMA. 2021;326(5):420-432. doi: 10.1001/jama.2021.11171 [DOI] [PubMed] [Google Scholar]
- 3.Marketing submission recommendations for a predetermined change control plan for artificial intelligence–enabled device software functions. US Food and Drug Administration . August 18, 2025. Accessed January 16, 2026. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/marketing-submission-recommendations-predetermined-change-control-plan-artificial-intelligence
- 4.Medical device databases. US Food and Drug Administration . August 16, 2023. Accessed January 16, 2026. https://www.fda.gov/medical-devices/device-advice-comprehensive-regulatory-assistance/medical-device-databases
- 5.Artificial intelligence–enabled medical devices. US Food and Drug Administration . December 5, 2025. Accessed January 16, 2026. https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices
- 6.Almarie B, Gonzalez-Gonzalez LF, Dos Santos Barbosa LA, Lutz A, Grosse U, Fregni F. Machine learning–enabled medical devices authorized by the US Food and Drug Administration in 2024: regulatory characteristics, predicate lineage, and transparency reporting. Biomedicines. 2025;13(12):3005. doi: 10.3390/biomedicines13123005 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eMethods
eReferences
Data Sharing Statement
