Stakeholders seek clarity, risk-based approach in AI/ML PCCP guidance

Monday 10 July, 2023

Various stakeholders are urging the US Food and Drug Administration (FDA) to use a risk-based and least burdensome approach to regulating artificial intelligence/machine learning (AI/ML) products through the use of Predetermined Change Control Plans (PCCPs). They are also asking regulators to clarify the recently published draft guidance applies to combination products and to ensure additional guidances are developed to address broader PCCP topics.
 
When Congress passed the Food and Drug Omnibus Reform Act (FDORA) late last year, it directed FDA to allow certain products on the market with PCCPs that would not require a resubmission every time there is an update to the product. Following passage of FDORA, the agency published a draft guidance outlining how sponsors of AI/ML products can submit PCCPs in their product submissions. (RELATED: FDA draft guidance allows AI/ML devices to evolve without requiring new submissionsRegulatory Focus 31 March 2023)
 
So far, the agency has received 30 comments on the draft guidance, including a letter from the Combination Products Coalition (CPC), which has been a proponent of enabling PCCPs for certain products.
 
“Although the Agency has clearly communicated its intention to apply the tenets of the draft guidance to device constituent parts of a combination products, the level of detail provided on implementing the guidance is lacking specifics in how the agency expects to implement practices for drug-device and biologic-device combination product and software used in combination with medicines,” the group wrote in its comments.
 
The CPC notes that section V of the draft guidance only discusses the issue in the context of FDA’s device center, and later on when giving examples of PCCP uses, it only provides examples relevant to standalone medical devices.
 
“Further, while the guidance suggests early engagement with [the Center for Drug Evaluation and Research] or [the Center for Biologics Evaluation and Research] for drug-device and biologic-device combination products, the interaction pathways discussed are limited to [the Center for Devices and Radiological Health] specific Q-submission process,” CPC added.

The group says future versions of the guidance should include details on implementing PCCPs in the context of drug-device and biologic-device combination product marketing applications and the software that is used.
 
The drug lobby group PhRMA also commented that it wants FDA to clarify that the recommendations in the draft guidance may also apply to drug-device or biologic-device combination products marketed under a new drug application (NDA) or biologics license application (BLA).
 
“PhRMA recommends that FDA specifically confirm that sponsors have the flexibility to submit a PCCP as part of the NDA or BLA for drug-device or biologic-device combination products and that a separate device marketing submission is not required in order to submit a PCCP for the device constituent of such products.”
 
PhRMA also notes that while FDA recommends early engagement with its drug and biologics centers to discuss combination products that may include Artificial Intelligence/Machine Learning-Enabled Device Software Functions (ML-DSF), the agency recommends sponsors use the Q-Submission process to get feedback on a proposed PCCP. The group says it supports the use of Q-Submission meetings in such cases but also wants Prescription Drug User Fee Act (PDUFA) meetings, such as Type C meetings, to be available for such discussions.
 
PhRMA adds that it would like to see additional guidance on how to establish a PCCP through an NDA or BLA, what to consider in terms of the components of a PCCP for an NDA or BLA, as well as guidance on submitting an application for a drug- or biologic-led combination product where the device component already has an approved PCCP.
 
The medtech lobby group AdvaMed also wants FDA to publish more guidances related to PCCP that is not technology specific.
 
“Additional guidance will offer greater clarity regarding how FDA intends to implement this new authority,” said the group. “Without additional guidance, the medical device industry would need to extrapolate the principles in this guidance, which is intended solely for devices with ML-DSF, to other device types and PCCP change types.”
 
AdvaMed is also concerned that phrasing in the draft guidance stating that it expects modifications included in the PCCP to keep the device within its indication for use to be too limiting to the use of PCCP. The group says that goes against how Congress intended FDA to use its statutory authority in FDORA.
 
“We recommend the draft guidance remove this limitation to the PCCP and align with the authorities outlined in the statute and with long-standing policies for the 510(k) program,” said AdvaMed.
 
“Further, the guidance should also make clear that a PCCP may be authorized under any premarket submission types, including Special and Abbreviated 510(k),” the group added. “The Special and Abbreviated 510(k) programs exist under the same statutory authority as the Traditional 510(k) program, and therefore the PCCP authorities apply equally to all 510(k) submission types.”
 
The concerns about the limitations of the guidance were also echoed by the medtech lobby group the Medical Device Manufacturers Association (MDMA) which also asked for additional guidances targeting PCCP issues that are not covered in the current proposed guidance.

“Indeed, Congress intended that the predetermined change control plan (PCCP) framework could apply more broadly to other types of devices, including those with software functions that do not utilize AI or ML, such as signal processing algorithms,” said MDMA. “Therefore, MDMA calls on FDA to develop further guidance that clarifies the circumstances in which PCCPs may be utilized for beyond AI/ML-enabled software functions.”
 
“Further, MDMA also calls on FDA to provide guidance on other topics potentially affected by the PCCP framework, including with respect to manufacturing, sterilization, and recordkeeping,” it added.
 
MDMA also said that it wants FDA to continue to take a risk-based approach to medical device premarket reviews and use of least burdensome principles by adopting those same positions when looking at the content of PCCPs. The group says the agency should adopt a flexible framework that allows for PCCPs to include verification and validation testing or other activities to support planned modifications.
 
“Final guidance concerning the PCCP framework should also provide greater flexibility to permit certain modifications to a device’s intended use to be effectuated through a PCCP, consistent with a risk-based approach and Congressional intent,” said MDMA.
 
Finally, the group also argued that FDA should clarify that a PCCP may be established through a premarket pathway such as a special 510(k). It also wants clarification in the guidance on when a single device with multiple ML models could be documented in a PCCP; when a PCCP could be proposed in response to an additional information request or deficiency letter; and how to modify a PCCP.
 
Another medtech lobby group, the Medical Imaging & Technology Alliance (MITA) also raised concerns the guidance is limited to AI/ML-enabled medical devices and argues that PCCP would also benefit Software as a Medical Device (SaMD) and Software in a Medical Device (SiMD) products. The group asked FDA if it would consider expanding PCCPs to non-AI/ML devices.
 
As with other commentors, MITA also urged FDA to use a risk-based and least burdensome approach to regulating products with PCCP.
 
“The draft guidance has recommendations for inclusion of vast amounts of details and information as part of the PCCP,” said MITA. “The requirements and expectations should be risk based and aligned with least burdensome principles.”
 
“Not all ML-DSFs present the same risk, with that in mind, the depth/breadth of information being requested in this draft guidance may be disproportional for cases where the risk posed by a ML-DSF is low,” the group argued. “We urge the agency to make such recommendations proportional to the risk profile of the ML-DSF.”

Regulatory Focus ,A RAPS Publication