General
Determining the Regulatory Pathway for Your Diagnostic & PGx SaMD
For a company developing a novel Software as a Medical Device (SaMD) intended for both diagnostic screening and pharmacogenetic assessment, what are the key considerations for determining the appropriate premarket submission pathway and evidence requirements?
For instance, consider a hypothetical prescription SaMD that uses an AI/ML algorithm to analyze retinal images for disease detection, a function similar in principle to devices classified under 21 CFR 886.1100. However, this device also incorporates a patient's genetic data to predict their likely response to a specific therapeutic class, touching on principles relevant to pharmacogenetic systems like those described under 21 CFR 862.3364.
Given this dual-functionality, how should a sponsor approach the following regulatory challenges?
First, in establishing the device classification and choosing between a 510(k) and a De Novo request, how is the overall risk profile evaluated when the device combines two distinct diagnostic functions with no single, clear predicate? If pursuing a 510(k), what is the strategy for demonstrating substantial equivalence when referencing multiple predicates from different device types and product codes?
Second, what type of performance data is generally expected for such a hybrid device? How can sponsors design validation studies that adequately characterize the analytical and clinical performance of both the image analysis algorithm and the pharmacogenetic assessment component? This includes addressing the AI/ML-specific considerations, such as managing algorithm training data and demonstrating the algorithm's generalizability.
Finally, how do existing special controls for related device types inform the potential requirements for the new device? Engaging with the FDA through the Q-Submission program is often recommended for novel technologies; what specific questions should be prioritized in a Pre-Submission meeting to gain clarity on the required validation evidence and regulatory pathway before committing to a final submission strategy?
---
*This Q&A was AI-assisted and reviewed for accuracy by Lo H. Khamis.*
💬 1 answers
👁️ 26 views
👍 1
Asked by Lo H. Khamis
Answers
Lo H. Khamis
👍 2
Determining the Regulatory Pathway for a Dual-Function Diagnostic and PGx SaMD
For companies developing novel Software as a Medical Device (SaMD) with multiple, distinct functions, such as combining diagnostic screening with pharmacogenetic (PGx) assessment, determining the appropriate U.S. FDA premarket submission pathway is a critical strategic challenge. These hybrid devices often lack a single, clear predicate, making the choice between a 510(k) and a De Novo request complex. The overall regulatory strategy hinges on a thorough evaluation of the device's combined risk profile, the feasibility of demonstrating substantial equivalence to multiple predicates, and the design of robust performance validation studies that address each unique function.
Consider a hypothetical prescription SaMD that uses an AI/ML algorithm to analyze retinal images for disease detection, a function with principles similar to devices classified under 21 CFR Part 886. At the same time, it incorporates a patient's genetic data to predict their response to a specific class of drugs, a function related to pharmacogenetic systems like those described under regulations such as 21 CFR 862.3360. For such a device, a sponsor must navigate how to classify the device, what evidence to generate, and how to best engage with the FDA to de-risk the submission process. Success depends on a proactive, well-documented strategy that addresses the validation needs of both the imaging and the genetic components, especially considering the unique data requirements for AI/ML algorithms.
### Key Points
* **Risk Profile Dictates the Pathway:** The device's overall risk classification, which drives the choice between a 510(k) and De Novo pathway, is determined by the cumulative risk of all its functions. The intended use and potential harm from an incorrect output from either the diagnostic or the PGx component must be carefully evaluated.
* **A "Hybrid" Predicate Strategy May Be Necessary:** For a 510(k), it is unlikely a single predicate device will exist for a dual-function SaMD. Sponsors often need to identify a primary predicate for the device's main function or highest-risk component and use secondary predicates to address other technological features or indications.
* **Comprehensive, Component-Specific Validation is Required:** Performance data must be generated to validate each major component of the device. This means separate but integrated analytical and clinical validation for the image analysis algorithm and the pharmacogenetic assessment algorithm.
* **AI/ML Validation Has Unique Requirements:** For SaMD utilizing AI/ML, FDA guidance documents emphasize the need for rigorous management of data. This includes demonstrating the quality and independence of training, tuning, and testing datasets, ensuring the data is representative of the intended use population, and proving the algorithm's generalizability.
* **Early FDA Engagement is Essential:** For a novel device with this level of complexity, the Q-Submission program is the most effective tool for gaining clarity and alignment with the FDA. A Pre-Submission meeting should be used to confirm the regulatory pathway, get feedback on a proposed predicate strategy, and vet the clinical validation plan *before* committing to a final submission.
## Determining the Device Classification and Premarket Pathway
The first and most critical step is to determine the device's classification and the corresponding premarket pathway. This decision is based on the device's intended use and the level of risk it presents to patients.
### Evaluating the Overall Device Risk
For a dual-function device, the risk assessment must consider each function independently and then as a whole.
1. **Diagnostic Imaging Component:** The risk here is associated with a false positive (leading to unnecessary follow-up procedures) or a false negative (leading to a missed or delayed diagnosis). The severity of the target disease is a major factor. A screening tool for a life-threatening condition is inherently higher risk than one for a less critical condition.
2. **Pharmacogenetic (PGx) Component:** The risk is tied to the clinical action taken based on the device's output. If the PGx result is used to guide dosing for a drug with a narrow therapeutic index or to select a therapy where an incorrect choice could lead to severe adverse events, the risk is significant.
The FDA will view the device's risk profile based on the highest-risk function. If the PGx component is determined to be moderate-risk (Class II) and the diagnostic component is also moderate-risk (Class II), the overall device will likely be regulated as Class II. However, if either function introduces novel risks or has a high-risk intended use (e.g., guiding therapy for a life-or-death condition with no alternative diagnostics), it could push the entire device into Class III, requiring a Premarket Approval (PMA) submission.
### The 510(k) vs. De Novo Decision
* **510(k) Pathway:** This pathway, governed by regulations in 21 CFR Part 807, is appropriate if the sponsor can identify one or more legally marketed "predicate" devices and demonstrate that the new device is "substantially equivalent" (SE) in terms of intended use, technological characteristics, and performance. For a dual-function SaMD, the challenge is finding predicates that cover all aspects of the new device.
* **De Novo Classification Request:** This pathway is for novel low-to-moderate-risk devices for which no predicate exists. If a sponsor cannot identify a suitable predicate or if the new technology raises different questions of safety and effectiveness, the De Novo pathway allows the FDA to classify the new device as either Class I or Class II and establish new special controls, if applicable.
For the hypothetical SaMD, if a sponsor can find a cleared device for AI-based retinal image analysis AND a cleared device for the specific type of PGx assessment, a 510(k) using a multi-predicate strategy might be viable. If no such PGx predicate exists, the novelty of that function would likely require a De Novo request.
### Crafting a Substantial Equivalence Argument with Multiple Predicates
When pursuing a 510(k) for a complex device, sponsors can reference multiple predicates to build a comprehensive SE argument.
1. **Select a Primary Predicate:** Identify the cleared device that is most similar to the new device's core function or highest-risk component. In our example, this might be a cleared AI SaMD for retinal disease detection.
2. **Use Secondary Predicates:** Identify other cleared devices to address the remaining features. A cleared PGx test system (similar to those regulated under 21 CFR 862.3360) could serve as a predicate for the genetic analysis component. Another device might be referenced for its software architecture or cybersecurity controls.
3. **Perform a Detailed Comparison:** The 510(k) submission must include a detailed, side-by-side comparison of the new device to each predicate. The goal is to demonstrate that any differences in technology or intended use do not raise new questions of safety or effectiveness. This is typically done through robust performance testing.
## Assembling the Necessary Performance Data
A submission for a dual-function SaMD requires a comprehensive data package that validates both the analytical and clinical performance of each component.
### Analytical Validation
Analytical validation demonstrates that the device's algorithms perform accurately and reliably from a technical standpoint.
* **For the Image Analysis Algorithm:** This involves testing the algorithm against a curated, well-characterized dataset with a known "ground truth" (e.g., diagnoses confirmed by a panel of expert clinicians). Key metrics include:
* Sensitivity and Specificity
* Positive and Negative Predictive Value (PPV/NPV)
* Receiver Operating Characteristic (ROC) curve analysis
* **For the PGx Algorithm:** This involves testing the device's ability to correctly identify genetic variants from sample data. This is typically done by comparing the SaMD's output to results from validated, established methods like Sanger sequencing using well-characterized human genetic samples.
### Clinical Validation
Clinical validation demonstrates that the device performs as intended in the target patient population and clinical workflow. For a dual-function device, the clinical study must be designed to generate evidence for both intended uses.
* **Study Design:** The study should enroll patients representative of the intended use population. The study protocol must clearly define the endpoints for both the diagnostic and PGx functions.
* **Evidence for Diagnostic Function:** This requires comparing the SaMD's output to the clinical gold standard for diagnosing the disease in question.
* **Evidence for PGx Function:** This is more complex and depends on the specific claim. It might involve correlating the algorithm's prediction with actual patient outcomes, drug metabolite levels, or observed clinical response to the targeted therapy.
### Special Considerations for AI/ML Algorithms
FDA guidance on AI/ML-based SaMD highlights several critical areas for validation:
* **Data Management:** Sponsors must provide a detailed description of the datasets used to train, tune, and test the algorithm. It is crucial that the final testing dataset is independent and was not used during algorithm development.
* **Algorithm Generalizability:** The sponsor must demonstrate that the algorithm's performance is maintained across different patient demographics (age, sex, ethnicity), disease severities, and clinical settings (e.g., different types of imaging hardware).
* **Algorithm Change Protocol (ACP):** For algorithms that are expected to learn and adapt over time, sponsors may propose a pre-specified ACP in their submission. This "predetermined change control plan" outlines the specific modifications the sponsor intends to make and the validation plan for those changes, potentially allowing for implementation without a new 510(k) submission for each update.
## Strategic Considerations and the Role of Q-Submission
Given the novelty and complexity of a dual-function SaMD, early and effective communication with the FDA is paramount. The Q-Submission program is the formal mechanism for requesting this feedback.
A Pre-Submission meeting allows sponsors to present their device, proposed regulatory strategy, and validation plans to the FDA review team and receive targeted, non-binding feedback. This can save significant time and resources by identifying potential agency concerns long before the final marketing submission is filed.
### Prioritizing Questions for the FDA
When preparing a Q-Submission for a dual-function SaMD, the briefing package should be detailed and the questions posed to the FDA should be specific and well-supported. Key questions to prioritize include:
1. **Regulatory Pathway:** "Based on the device description, intended use, and risk analysis provided, does the Agency concur with our assessment that the device is moderate-risk and that a [510(k) or De Novo] is the appropriate premarket pathway?"
2. **Predicate Strategy (for a 510(k)):** "We propose using Predicate X for the imaging function and Predicate Y for the PGx function. Does the Agency agree that this is a reasonable predicate strategy? Are there specific technological differences that the Agency believes constitute a new question of safety or effectiveness that would require a De Novo submission?"
3. **Clinical Validation Plan:** "We have provided our proposed clinical study protocol. Does the Agency have any feedback on our proposed patient population, primary and secondary endpoints, and statistical analysis plan to support the proposed intended use for both functions?"
4. **AI/ML Evidence:** "Does the Agency find our plan for managing and curating the training, tuning, and testing datasets for our AI/ML algorithm to be adequate? Are there specific recommendations for demonstrating the generalizability of our algorithm?"
## Key FDA References
Sponsors developing SaMD should familiarize themselves with the latest FDA guidance and regulations. While device-specific guidances exist, the following general documents provide a foundational framework:
* FDA's Q-Submission Program guidance
* FDA's 510(k) Program guidance on the substantial equivalence framework
* 21 CFR Part 807, Subpart E – Premarket Notification Procedures
* Relevant FDA guidance on Software as a Medical Device (SaMD), including AI/ML-enabled devices
* FDA guidance on Cybersecurity in Medical Devices
## Finding and Comparing REACH Only Representative Providers
Navigating global regulatory landscapes requires specialized expertise. Whether you are entering the US market with the FDA or expanding into other regions like the European Union, finding the right regulatory partner is crucial for success. Different markets have unique requirements, and selecting a qualified representative or consultant ensures you can meet all compliance obligations efficiently.
To find qualified vetted providers [click here](https://cruxi.ai/regulatory-directories/reach_only_rep) and request quotes for free.
---
This article is for general educational purposes only and is not legal, medical, or regulatory advice. For device-specific questions, sponsors should consult qualified experts and consider engaging FDA via the Q-Submission program.
---
*This answer was AI-assisted and reviewed for accuracy by Lo H. Khamis.*