The use of artificial intelligence (AI) in health care, and specifically, as the utilization of software in medical devices continues to increase. Machine learning (ML) is a branch of AI in which computer systems automatically adapt, improve, and make predictions by applying algorithms to analyze and draw inferences from data. The FDA regulates software that qualifies as a medical device and recently issued guidance that provides examples of software-based devices that may incorporate AI/ML, and thus, fall within its purview.
What You Need to Know:
- AI/ML innovations that improve patient outcomes may be considered a medical device and subject to regulation by the FDA.
- In 2021, the FDA issued an action plan that laid out a framework for the regulation of AI/ML-based software as a medical device.
- In September 2022, the FDA provided guidance about the scope of its regulatory authority over software that is intended to support the diagnosis, treatment, prevention, cure, or mitigation of diseases or other conditions. Such software may utilize AI/ML as part of its functionality.
The FDA is responsible for regulating the safety and effectiveness of medical devices, a number of which use software. Also, software itself may be a medical device. The term Software as a Medical Device (SaMD) is defined by the International Medical Device Regulators Forum to mean “software intended to be used for one or more medical purposes that perform these purposes without being part of a hardware medical device.” SaMD is different from software that is integrated into or used in the manufacture and maintenance of a medical device.
The FDA considers software that provides support to treat, diagnose, cure, mitigate, or prevent disease or other conditions to be a medical device. This type of software may be referred to as clinical decision support (CDS) software. The FDA has historically regulated CDS software under section 201(h) of the Federal Food, Drug, and Cosmetic Act (FDCA). The number of applications to market medical devices that utilize or are AI/ML software has grown over the last decade. As recently as October 5, 2022, the FDA added 178 devices to its list of AI/ML-enabled medical devices marketed in the US.
In January 2021, the FDA issued its “Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) Action Plan,” outlining its vision and a framework for regulating SaMD that utilizes AI/ML. As part of its action plan, the FDA intends to, among other things, encourage the harmonization of Good Machine Learning Practice development, hold a public workshop on how medical device labeling supports transparency to product users, and support efforts to evaluate and improve ML algorithms.
Recently, the FDA took another step in the regulation of medical devices that utilize AI and ML. On September 28, 2022, the FDA issued guidance to the industry and FDA administration staff that clarified the agency’s oversight of CDS software as medical devices (the “Guidance”). The Guidance analyzes CDS software that would be excluded from the definition of a device by Section 520(o)(1)(E) of the FDCA. To be excluded, the software must satisfy all four of the following criteria:
- Criterion One - the software function is not intended to acquire, process, or analyze a medical image or a signal from an in vitro diagnostic device or a pattern or signal from a signal acquisition system;
- Criterion Two - the software function is intended for the purpose of displaying, analyzing, or printing medical information about a patient or other medical information;
- Criterion Three - the software function is intended for the purpose of supporting or providing recommendations to a health care professional about prevention, diagnosis, or treatment of a disease or condition; and
- Criterion Four - the software functions are intended to allow the health care professional to independently review the basis for any recommendations made by the software, such that the professional does not primarily rely on any of the recommendations to make a clinical decision about a patient.
In addition, the Guidance provides examples of medical device software functions on which the FDA intends to focus its regulatory oversight. These are software functions that do not meet all of the four criteria listed above, and some of which, based on their description, may utilize AI/ML. The Guidance provides more instances of device software functions than those discussed below. The following examples are worth noting, however, because they are illustrative of SaMD that may utilize AI/ML.
- Software functionality that identifies patients who may be diagnosed with an opioid addiction, based on analysis of patient-specific medical information, family history, prescription patterns, and geographical data. This type of functionality does not satisfy Criterion Three because it recommends a diagnosis about a patient.
- Software function that analyzes sound waves captured when users cough or recite certain sentences to diagnose bronchitis or a sinus infection. This software does not meet Criterion One because it is intended to analyze a signal or pattern, and it does not satisfy Criterion Three because it recommends a diagnosis.
- Software functionality that is intended for healthcare professional (HCP) management of heart failure patients that analyzes patient-specific medical information to predict heart failure hospitalization.
- Software that analyzes hourly pulse oximetry and heart rate measurements to identify signs of patient deterioration and alert an HCP. This software does not meet Criterion One because it analyzes a pattern.
- Software functionality that analyzes patient-specific medical information to detect a life-threatening condition, such as a stroke, and generate an alarm or an alert to notify an HCP.
The functionality in these examples describes the type of task performed by AI or ML - software that makes a prediction based on the analysis of different types of real-world data or the recognition of a pattern. ML could be used to automate this type of analysis, minimizing or eliminating human HCP involvement in data selection, and may lead to improved patient treatment through comprehensive analyses, helpful recommendations, and accurate diagnoses.
Innovation in AI/ML is growing rapidly, especially in the healthcare industry. The number of applications submitted to the FDA to market medical devices that are or use AI/ML-based software is surely to continue increasing. The FDA has provided a framework to the industry and the end-users, and continues to refine its approach to regulating the safety and effectiveness of software that qualifies as a medical device, including software that utilizes AI/ML innovations.
