Authors: Gerald L. Klein, MD(1); Michael Fath, PhD(1,2); Freddy Byrth, BA(1); Lee Schacter, PhD, MD(1); Emilia Jones Amaowei(1,3); Patrick Loebs, MSW/MPH(1); Roger Morgan, MD(1)
Affiliations: MedSurgPI(1); Cavabio Consulting(2); EJ MedInsights(3)
Guidance for Drug Development and Medtech
This edition focuses on two decisions in product development: when to amend a protocol and how to evaluate AI used in a medical device.
Drug Development
When a Protocol Change Requires an Amendment
For a study under a U.S. investigational new drug application (IND), the FDA amendment submission threshold depends on the phase:
· Phase 1: A change that significantly affects participant safety.
· Phase 2 or 3: A change that significantly affects participant safety, the scope of the investigation, or the scientific quality of the study.1
Examples include dose or exposure increases beyond the current protocol, significant change in enrollment criteria or trial design changes, adding safety criteria, or discontinuation of safety tests.1
Before implementation, submit a qualifying change to FDA and obtain Institutional Review Board (IRB) approval, in either order. A change intended to eliminate an apparent immediate hazard to subjects may be implemented immediately, followed by FDA notification through a protocol amendment and notification of the reviewing IRB. Other changes to approved research still require appropriate IRB review; the IRB may use expedited procedures for eligible minor modifications.1,2
For Australian studies, submit amendments to the reviewing Human Research Ethics Committee (HREC) for ethics approval and to each site’s Research Governance Office for governance authorization. Depending on the change and whether the trial runs under the Clinical Trial Notification (CTN) or Clinical Trial Approval (CTA) scheme, a CTN variation or Therapeutic Goods Administration (TGA) approval of a CTA variation may also be needed.3 For EU substantial modifications affecting Part I (scientific and medicinal-product aspects assessed jointly by the Member States concerned) or both Parts I and II (national aspects, such as informed consent, recruitment, and site suitability), the basic time sequence is 6 days for validation, 38 days for assessment, and 5 days for the decision. Assessment questions may add up to 31 days; other applicable extensions can also lengthen the process.4
A study published in 2024 reported amendments in approximately 76% of protocols, averaging 3.3 amendments among protocols that were amended, and a mean of approximately 260 days from internal approval of an amendment to the last required ethics or oversight approval, which may precede implementation at every site. These are historical operational benchmarks, not regulatory review deadlines.5
Medtech Development
Building Evidence for AI in a Medical Device
When AI performs a medical device function, its performance is a core part of the evidence on the device’s safety and effectiveness. Build the evaluation around its intended patients, users, and clinical setting.6
· Use data that represent relevant patient groups, sites, equipment, and conditions of use.
· Maintain appropriate independence between training and test datasets, addressing potential dependence arising from shared patients, sites, or data-acquisition processes.
· Evaluate clinically meaningful performance in the intended workflow, including human-AI interactions where applicable, and examine subgroups in which performance may be inadequate. Define acceptable performance criteria in advance.
· Evaluate how clinicians or patients understand and act on the output.
· Plan post market monitoring for performance degradation and dataset drift, and control model updates after deployment, including through an FDA-authorized Predetermined Change Control Plan (PCCP) where appropriate.6,7
The 2025 International Medical Device Regulators Forum (IMDRF) Good Machine Learning Practice principles provide an internationally agreed framework that builds on the principles issued jointly by FDA, Health Canada, and the UK Medicines and Healthcare products Regulatory Agency (MHRA) in 2021. FDA’s January 2025 lifecycle management document remains draft guidance, with nonbinding recommendations and a designation of “not for implementation.”6,8
References
Regulations and online sources accessed October 8, 2026.
1. Protocol amendments. 21 CFR §312.30(b). Current regulation.
2. IRB functions and operations. 21 CFR §56.108(a)(4), (b). Expedited review procedures. 21 CFR §56.110(b). Section 56.108 | Section 56.110.
3. Therapeutic Goods Administration. Clinical Trial Notification (CTN) scheme: Variations. TGA CTN scheme.
4. Central Committee on Research Involving Human Subjects. Timelines for substantial modifications under Regulation (EU) No 536/2014. CCMO timelines.
5. Getz K, Smith Z, Botto E, Murphy E, Dauchy A. New benchmarks on protocol amendment practices, trends and their impact on clinical trial performance. Ther Innov Regul Sci. 2024;58(3):539–548. doi:10.1007/s43441-024-00622-9. Published article.
6. International Medical Device Regulators Forum. Good machine learning practice for medical device development: Guiding principles. IMDRF/AIML WG/N88 FINAL:2025. 2025. Final principles.
7. US Food and Drug Administration. Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions. Final guidance. December 2024.
8. US Food and Drug Administration. Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations. Draft guidance. January 2025. FDA draft guidance.