Regulatory Framework and Legislative Background
The FDA’s final guidance on Clinical Decision Support Software, issued on September 28, 2022, builds upon the amendments introduced by the 21st Century Cures Act. Enacted on December 13, 2016, this act modified section 520 of the Federal Food, Drug, and Cosmetic Act (FD&C Act) to exclude specific categories of medical software functions, including certain decision support software, from the definition of a device under section 201(h) of the FD&C Act. This exclusion applies only to software functions that meet the criteria outlined in section 520(o)(1)(E) of the FD&C Act, referred to as the Non-Device CDS criteria.
The guidance clarifies the FDA’s interpretation of which software functions fall under these criteria, distinguishing between software that qualifies as a non-device and software that remains subject to device regulations. This distinction is critical for developers, as it determines whether their software requires premarket review or other regulatory oversight.
The guidance also emphasizes that existing digital health policies continue to apply to software functions that do not meet the Non-Device CDS criteria. This includes software intended for use by patients or caregivers, which remains subject to device classification and regulatory requirements.
The FDA explicitly states that this guidance does not alter the agency’s broader authority to regulate software as a device when necessary to ensure public health and safety.
Non-Device Clinical Decision Support Software Criteria
The Non-Device CDS criteria, as defined in section 520(o)(1)(E) of the FD&C Act, exclude certain clinical decision support software functions from the definition of a device.
Key characteristics of Non-Device CDS software include transparency and clinician-reviewability. The software must allow clinicians to review, modify, or override its recommendations without undue difficulty.
The guidance provides examples to illustrate how the FDA interprets these criteria. Conversely, software that autonomously executes clinical decisions, such as adjusting medication dosages without clinician intervention, would likely be classified as a device and subject to premarket review.
The FDA’s examples highlight the importance of distinguishing between software that assists clinicians and software that operates independently. Predictive and image-analyzing tools, such as those used for radiology or pathology, are typically classified as devices because they perform diagnostic or treatment-related functions without direct clinician oversight.
Software Functions Subject to Device Regulations
The guidance explicitly states that most predictive and image-analyzing AI tools are considered devices under the FD&C Act. Similarly, predictive algorithms that forecast patient outcomes or recommend treatment plans are generally treated as devices if they influence clinical decisions without clinician intervention.
Software that escapes premarket review under the Non-Device CDS criteria must adhere to strict conditions. It must not interfere with the clinician’s ability to exercise independent judgment and must provide clear, actionable recommendations that can be easily reviewed or modified. The FDA’s focus on transparency and clinician control ensures that such software remains a tool for assistance rather than a replacement for professional expertise.
The guidance also underscores that software intended for patient or caregiver use is subject to device regulations, regardless of its classification under the Non-Device CDS criteria. This includes mobile applications or platforms that provide health recommendations directly to consumers, as these tools may impact patient care and require regulatory oversight to ensure safety and efficacy.
Developers are encouraged to consult the guidance for specific examples and clarifications on how the FDA evaluates different types of software functions.
Public Comment and Regulatory Compliance
The FDA invites public comments on this guidance, allowing stakeholders to provide feedback on its interpretation and implementation. Comments can be submitted online or in written form, with instructions provided in the guidance document. The docket number for this guidance, FDA-2017-D-6569, should be included in all submissions to ensure proper routing.
Written comments can be mailed to the FDA if online submission is not feasible.
Compliance with this guidance is voluntary but serves as a critical reference for developers seeking to navigate the regulatory landscape. By adhering to the outlined criteria, manufacturers can determine whether their software qualifies for exclusion from device classification or requires premarket review.
The guidance’s emphasis on transparency and clinician involvement aligns with the FDA’s broader goal of fostering innovation in digital health while maintaining rigorous standards for software that directly impacts patient care.
Required actions
- Software that provides a single recommendation, analyses medical images or signals, or is intended for time-critical decisions remains a regulated device - Non-device CDS must let clinicians independently review the basis of recommendations and not replace their judgement