The Impact of FDA Digital Health Guidance on CDS, Medical Software, and Machine Learning
The FDA recently released guidance documents on select provisions of the 21st Century Cures Act (the Cures Act). Digital health products and services providers are most concerned about guidance on the use of clinical decision support (CDS) and medical software. For healthcare organizations working with these fast-changing technologies and constantly pushing digital innovation, federal guidance that hints at changing the regulation of their pioneering efforts may send up warning flags.
However, there is little cause for alarm. The FDA regularly issues clarification on how it is going to enforce statutes and regulations to provide useful interpretive information to the industry. Although the FDA digital health guidance documents are not regulation (they are non-binding, informational communications), digital health executives and developers can still use the FDA’s guidance to understand the current approach it is taking toward regulation, and how this approach impacts CDS and medical software development, as well as provider workflows.
FDA Guidance on CDS
Healthcare organizations use CDS to reduce errors and improve efficiency, standardization, and cost savings. The FDA guidance on CDS (the CDS Guidance) describes what types of software tools the FDA considers to be CDS, and which CDS is excluded from FDA regulation as a medical device.
The CDS Guidance states that software functions must meet the following three criteria to be considered CDS:
- 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;
- Intended for the purpose of displaying, analyzing, or printing medical information about a patient or other medical information (such as peer-reviewed clinical studies and clinical practice guidelines);
- Intended for the purpose of supporting or providing recommendations to a health care professional about prevention, diagnosis, or treatment of a disease or condition.
Having met these three criteria, CDS must then meet a fourth criterion to fall outside the FDA’s definition of a medical device and be exempt from regulation:
- Intended for the purpose of enabling such health care professional to independently review the basis for such recommendations that such software presents so that it is not the intent that such health care professional rely primarily on any of such recommendations to make a clinical diagnosis or treatment decision regarding an individual patient.
The CDS Guidance provides examples of CDS software functions that do and do not meet these criteria. Two key elements are provided in the guidance to further define the criteria for exemption:
- The healthcare practitioner must be able to independently evaluate the basis of the CDS recommendations.
- The CDS recommendations must be based on publicly available clinical guidelines (i.e., studies, published literature, clinical practice guidelines).
These two elements beg the question: what about using algorithms and machine learning as the basis for CDS recommendations?
Machine Learning as a Basis for CDS Recommendations
Healthcare machine learning models produce information that improves patient care. This information may not always be based on published literature, but its reliability is correlated to significant historical patient data and actual outcomes. Providers will exercise clinical judgment before taking any action on the output of machine learning, which is often in the form of recommendations, alerts, and risk prioritizations.
What’s concerning is that an algorithm could operate on unspecified health data within a “black box” and deliver a result, the basis of which a provider might not understand. The FDA is appropriately addressing this issue by requiring disclosure of the inputs used in order for software to be excluded from regulation. Health Catalyst agrees with the principle that any applications using algorithms should have functionality that allows users to view the input features and variables, including the clinical guidelines or findings that helped shape the predictive model.
It is also important to note that, as discussed in an earlier article, a clinical intermediary will always be part of machine learning-enabled CDS to impose judgement and connect any recommendations (produced by the machine learning model) to relevant clinical guidelines.
Health Catalyst and other organizations will seek clarification from the FDA on the use of algorithms and machine learning as the basis for providing CDS recommendations, in lieu of, or in addition to, publicly available clinical guidelines.
The use of the term “providing recommendations” in the CDS Guidance could also be clarified. Many CDS systems, including Health Catalyst’s analytics applications, provide prioritizations and alerts, but they don’t always recommend a course of action. The FDA’s CDS Guidance only addresses recommendations, but innovative healthcare systems may need more nuance to address their advanced clinical environments. The FDA is expected to review and respond to the public comments when it issues its final guidance documents.
The CDS Guidance provides informative examples of CDS functionality that fall into the following three categories:
- CDS functionality that does not constitute a medical device—these are CDS functions that do not meet the definition of a medical device;
- CDS functionality that may meet the definition of a medical device but for which the FDA does not intend to enforce compliance; and
- CDS functionality that the FDA intends to regulate as a medical device.
Where the FDA “does not intend to enforce compliance” means that the decision to not regulate non-medical-device software is based on the FDA’s discretion. In the future, the FDA could modify its policy, which could result in a more significant impact on vendors that use CDS with machine learning functionality. While the FDA’s approach introduces uncertainty, it is our opinion that, because there have been enough policies issued over a sustained period, radical shifts in enforcement policy are unlikely.
Patient Decision Support
The CDS Guidance also addresses patient decision support (PDS) software, regarding functionality for use by patients rather than clinicians. This guidance has similar criteria to CDS and indicates that PDS will not be regulated as a device. The FDA appears to be drawing the distinction because PDS likely applies to a different audience (patients). For example, PDS guidance applies to patient-accessed modules and functions within Health Catalyst’s Care Coordination application in the Care Management suite.
FDA Guidance on Medical Software
The FDA guidance on changes to existing medical software policies (the Medical Software Guidance) provides clarification on policy changes made by the Cures Act. The Medical Software Guidance builds upon several previous FDA guidance documents and clarifies that software functions classified into any of four areas are exempt from regulation as a medical device:
- Software function intended for administrative support of a healthcare facility.
- Software function intended for maintaining or encouraging a healthy lifestyle.
- Software function intended to serve as electronic patient records.
- Software function intended for transferring, storing, converting formats, or displaying data and results.
The Cures Act had already exempted these software functions, so this guidance simply provides more detail related to the Cures Act provisions, as well as prior guidance on mobile medical applications. Software, like Health Catalyst’s team-based fitness and wellness application, falls under this guidance, which excludes software for maintaining or encouraging a healthy lifestyle, and which is unrelated to the diagnosis, cure, mitigation, prevention, or treatment of a disease or condition.
Additional Ramifications of the FDA Digital Health Guidance
As long as software vendors comply with the FDA guidance, any cost impact should be minimal, provided that their software does not seem to fall within a category that the FDA indicates is a medical device. If vendors aren’t already providing transparency through their applications, then workflow processes could be impacted as they move from a black box approach to one in which data inputs used by algorithms are disclosed to users. Providing these disclosures could potentially involve more time and effort by providers because it could add an input review phase with an internal clinician panel, although this would be discretionary.
The CDS Guidance and Medical Software Guidance were issued in draft form to allow the FDA to consider public comments. While possible, it is not likely that the exemptions will change significantly in the final documents. Vendors might expect to see clarification on how to disclose the basis for recommendations or conclusions derived by CDS, and where machine learning fits into these disclosures. The FDA will also likely include a few more examples of what functionality it considers to be exempt from, or subject to, regulation when it issues its final guidance later in 2018.
Good News for Digital Health Innovators
There are no major surprises with the new FDA guidance documents in terms of what will be regulated or not. The CDS Guidance reassures practitioners that CDS functions that meet the non-device criteria will either be excluded from regulation or the FDA will not enforce regulation. This approach can help promote innovation by ensuring a quicker time to market—and value—for developers and CDS system users.
In its own words, the FDA’s intent is to foster, not inhibit, innovation, and to encourage the development of tools that can help people be more informed about their health. Though these documents don’t establish rules, they should be taken seriously because they contain useful information about how the FDA will enforce its rules, and how it interprets the Cures Act.
The industry response to these new guidance documents should be positive. The policies provide appropriate oversight, promote software transparency, and let digital health developers “know where they stand relative to the FDA’s regulatory framework.” Most importantly, the guidance supports the success of vendors and healthcare organizations as they continue working to improve the features and functionality of their CDS systems and medical software.
Would you like to learn more about this topic? Here are some articles we suggest:
- Machine Learning in Healthcare: How it Supports Clinician Decisions—and Why Clinicians are Still in Charge
- Machine Learning 101: 5 Easy Steps for Using it in Healthcare
- The Why And How Of Machine Learning And AI: An Implementation Guide For Healthcare Leaders
- Healthcare Decision Support: A Modern Tool for Today’s Chief Nursing Officer
- Why Healthcare Decision Support Is No Longer Optional for Chief Operating Officers
Would you like to use or share these concepts? Download this presentation highlighting the key main points.