A guide to FDA Diagnostic Radiology Product Codes
When it comes to the US market, three-letters stand in the way of getting your products to market. FDA Product Codes are essential in understanding and classifying your medical device. Here we breakdown FDA product codes and analyse the existing product codes for diagnostic radiology AI devices currently available on the market.
Classification begins with the Code of Federal Regulations, specifically Title 21 - commonly referred to as CFR Title 21. There are over 1000 parts to the CFR, with 98 parts dedicated to medical or surgical devices (Part 800 - 898). The regulations cover various aspects of design, clinical evaluation, manufacturing, packaging, labelling and post market surveillance of medical devices. In essence they are the US equivalent of the EU Medical Device Regulations (MDR) in that they are federal requirements for all US states. Within each part of the regulations, single or multiple āproduct codesā can exist, with each code having slight nuances in definition.
What are FDA product codes?
Product codes are 3-letter codes assigned to a product, based on identified predicate devices, with similarities in device description and intended use. Product codes were initially introduced as a way for regulators to accurately monitor and track medical devices on the market. For developers, product codes can often provide a framework to their productās intended use and how it would be regulated by the FDA.
How does the FDA come up with their codes?
Before a medical device can be considered for market access, the FDA requires a premarket submission using an existing product code from predicate devices that exist on the market. Whilst the 3-letter nomenclature seen within product codes seems to be randomly generated, medical devices similar in intended use and device description are categorised into the same product code or a completely new product code can be generated if there are no identified predicate devices (more about this later). Radiological AI device product codes tend to start with the letter Q, though isnāt always the case as we shall see below.
Submissions made under a certain product code will be directed to the relevant classification panel for review. As new product codes are made, existing devices may be re-assigned into a more suitable product code. In cases where an incorrect product code is assigned or a more suitable product code is identified, the agency will reallocate the product code to the most accurate code. For specific products, there is full guidance on the FDAās current thinking within the FDA Guidance Documents.
Radiology AI Product Codes
Letās take a deep dive into Radiology (CFR Part 892). Radiology is made up of two main sub-parts, diagnostics (892.1000 - 892.2080) and therapeutics (892.2050 - 892.2090). It is important to note that multiple product codes can exist under a CFR section.
Product Codes within CFR section 892.2050 are medical devices that perform image processing for "disease detection, diagnosis or patient managementā by a clinician. Image processing can include image manipulation, analysis or quantification. More complex functions include image segmentation, image registration or 3D visualisation. Product code LLZ includes medical devices that process radiological images for all pathologies not including cancer. Product code QIH is more specific within its methods, including the use of AI to perform image processing, for all pathologies not including cancer.
CFR section 892.2060 describes computer-assisted diagnostic (CADx) software that uses product code POK for medical devices intended to characterise lesions on radiological images suspicious for cancer. Information from radiologist-derived regions of interest, such as size, density etc is extracted from an image, processed and presented to clinical users for assistance with diagnosis. Note, these products do not triage, prioritise or highlight regions of interest. For example Koios DS For Breast (2019) allows radiologists to draw a boundary box around a lesion on breast ultrasound and receive a BiRADS categorisation.
CFR section 892.2070 defines a computer-assisted detection (CADe) device using product code MYN, not intended for diagnosis or triage. This product code is limited to specific cancers mentioned. For example, Auto Lung Nodule Detection (2021) is a chest x-ray CADe device under product code MYN, which identifies and marks (encircles) suspected pulmonary nodules from 10 to 30 mm in size. These devices cannot suggest a diagnosis, only highlight potential lesions for radiological review.
CFR section 892.2080 includes medical devices intended for computer assisted triage and notification. Product code QAS performs computer assisted triage using pre-specified clinician inputs and notifies the user without suggesting prioritisation. QFM performs computer assisted triage and prioritisation, using machine machine learning, artificial intelligence or other image analysis algorithms based on high sensitivity and specificity >95% AUC (in contrast, no performance metrics are mentioned for QAS). Examples of QFM include Annalise Enterprise CXR Triage Pneumothorax (2021) which triages and prioritises pneumothorax and tension pneumothorax and Lunit CXR (2021) which notifies and triages pleural effusion and/ or pneumothorax. It is important to note that QAS and QFM products do not direct a radiologistās attention to a specific part of the image - the intended use is solely to triage and/or prioritise images in a workflow.
CFR section 892.2090 is similar to section 892.2060, but where section 892.2060 is specifically CADx for cancer, section 892.2090 is for CADe and CADx. Product code QDQ is intended for devices that use machine learning algorithms to detect and diagnose suspicious lesions for cancer. There are nuanced similarities between QDQ (breast only), MYN (breast, CXR and dental caries) and POK (cancer), so it is important to compare individual devices within each product code when identifying a predicate. Product code QBS is the only code for a single non-cancerous pathology - CADe and CADx software specifically for bony fractures on radiographs. For example, BoneView (2022) is categorised under product code QBS, performing image detection and diagnosis on chest x-rays of multiple anatomical areas, including the ribs.
Of note, there are currently (at the time of writing) no FDA product codes for radiology AI devices that can find, highlight and diagnose single non-cancerous pathologies, nor indeed multiple pathologies. To date, only products that either triage cases or notify users for single non-cancerous pathologies are FDA cleared. We imagine that the FDA will be reviewing several competitor devices with the intended use of diagnosing multiple pathologies in the near future, via the De Novo pathway. Once that ground is broken, others will be able to follow suit using the 510(k) pathway.
510(k) pre-market approval - the need to demonstrate equivalence
Medical devices are classified by risk within three classes. Class I devices are considered the lowest risk, whilst class II and III are considered to be moderate and high risk respectively. For class I and II devices, 510(k) pre-market submissions must be made, unless a product is found to be 510(k) exempt. All the product codes for radiology AI devices are class II. For class III devices, premarket approval (PMA) must be obtained as well as a 510(k) submission. Determining whether your product is 510(k) exempt or requires a PMA, begins with identifying your product code and searching the pre-market approval database (PMA) and the 510(k) pre-market notification database.
Pre-market submissions are required to demonstrate substantial equivalence in safety and effectiveness with another existing product on the market with FDA approval. Whilst new devices do not need to be identical to predicate devices, three main comparators are used to compare if products are similar: intended use, device features, performance testing. If any questions are raised regarding the product, a reclassification process may occur.
De-Novo Application
Sometimes, thereās no legally marketed predicate device for which substantial equivalence can be compared against. In these cases, a De Novo classification request is made, specifically for the intended product. Products which fall under De Novo classification request are identified by the submitter before 510(k) submission, or by the review through a not substantially equivalent (NSE) determination after a 510(k) submission. If a request for a Class I or Class II device classification is approved future submissions can be compared with the newly marketed device as a predicate. In other instances, products which have been automatically classified as Class III devices can be considered for reclassification under a De Novo application.
Prior to submission, the FDA recommends a pre-submission meeting as part of the FDA Q-Submission Program. This involves a formal request for feedback for 510(k)s, De-Novo applications and Pre-Market Approvals. Hardian will be with you on these calls to navigate the questions and nuances around these submissions.
Hardian Hot Takes
Product codes provide a starting point for classifying your product against possible predicates on the market.
Product codes are specific to intended use, with some crossover between codes.
Currently, there are no product codes that exist for the detection or diagnosis of multiple pathologies, such devices should be considered for a De Novo application.
Following this framework allows for a better understanding of your product, including its intended use, performance outcomes and risk classification.
Getting healthtech to market doesn't have to be difficult. We are a team of multidisciplinary experts supporting innovators to accelerate healthtech to healthcare, safely, and securely. We are no strangers to the nuances of FDA product codes for all kinds of software and AI medical devices.
Hardian Health is clinical digital consultancy focussed on leveraging technology into healthcare markets through clinical strategy, scientific validation, regulation, health economics and intellectual property.
