How the FDA could impact machine learning, IoT in healthcare
With a new president, the Food and Drug Administration may well reconsider whether mobile apps that deliver healthcare information to patients should be regulated. Scott Gottlieb, MD, the new FDA Commissioner, has used software as an example of where a new approach is needed.
In a 2015 article for Forbes, Dr. Gottlieb wrote, “Regulatory pathways should distinguish between digital tools that are trying to function like an existing medical device, versus those that are performing patient-specific analyses and provide patient-specific information based on data that can be accessed already, albeit far less efficiently. For the latter, we can take a page from the setup with EMRs, where the feds certify standards and an independent nonprofit ensures that software platforms meet those requirements.”
Since 2015, the technology has continued to zoom ahead, with a greater emphasis on using machine learning algorithms as well as scads of data presented by the Internet of Things. The question is, what exactly can we anticipate from this Administration regarding the future of FDA, machine learning and the Internet of Things?
It’s easy to get caught up in dreams when talking about machine learning in healthcare, but let’s take asthma management as a relatively near-term goal. Researchers are well on their way to devising software that would operate in the cloud, analyzing data for the purpose of sending messages via patients’ cell phones on when they should take their asthma medicine.
The software has a growing amount of data available to it through the Internet of Things, which ultimately will include body sensors monitoring such biosignals as heart rate, respiration sounds, hydration and blood oxygen saturation. In the near future, the sources of data may well include irritants for people with asthma, including pollen and pollution. Further, through a smart inhaler connected via Bluetooth to a patient’s cell phone, the software will know exactly when the patient has used inhaled corticosteroids—used to manage asthma over the long term—and short-acting beta antagonists.
The Federal Food Drug and Cosmetic Act gives the FDA oversight of both drugs and medical devices. Broadly speaking, regulated medical devices include those items, such as software, that are intended for use in the cure, mitigation, treatment or prevention of disease. In this example, software that directs the patient on when to take specific medicines in the management of asthma would seem to fit that definition. Indeed, FDA has regulated various types of software for decades.
Because of the 21st Century Cures Act passed last December, software that advises healthcare professionals is no longer regulated by the FDA, as long as the professional has the ability to independently review the basis for the recommendation. However, much of the intelligent software used in connection with the Internet of Things will target consumers to change behaviors. One study, for example, has shown that average adherence among children using inhaled corticosteroids for asthma improved from 30 percent to 84 percent with the introduction of smart inhalers that produce audiovisual reminders. The opportunity to impact patient behavior is too alluring not to nudge the patients directly. So it is likely that much of the machine learning-based software used in connection with the Internet of Things will not qualify for this exclusion from FDA oversight.
While FDA has said that the agency will use enforcement discretion to avoid wasting time regulating low-risk software, at least prior to the new Administration, it is unlikely that FDA would characterize this use case as low risk. These drugs need to be carefully used—research shows that long-acting beta antagonists, for example, may actually increase the risk of a severe asthma attack. And because these drugs can mask asthma deterioration, they can cause problems if used with an acute asthma attack. The Centers for Disease Control estimates that asthma attacks lead to 1.6 million emergency room visits each year in the United States and about 3,600 deaths.
So the FDA likely would regulate this particular software presently. The question is, under the new Administration, will that change? The new administration really has not shed much light on this particular question yet, but here’s how I see the forces for change lining up.
We will certainly continue to hear the White House talk about the FDA and its proclivity for discouraging innovation. Pressure will mount on the FDA to do something, and Commissioner Gottlieb already clearly has this area in his sights.
The FDA is an agency of laws, and the underlying laws have been stitched together over nearly a century. Congress does not change them quickly. Only four months ago, Congress amended the definition of a medical device to exclude certain software, the first time Congress has amended the statute with regard to software since the medical device definition was originally enacted in 1976. It seems unlikely that Congress is going to re-examine that definition any time soon.
When the agency is operating at its best, it focuses on science and the public health. The rank-and-file at the FDA are driven by risk. So in looking at the use case, the agency would typically focus on risks such as:
- False positives. What if the software tells the patient to administer a drug when the drug isn’t necessary? There is some risk that excessive medicine could actually make asthma worse.
- False negatives. What if the software tells the patient nothing, because the software hasn’t identified the need for medication? The software may fail to get the asthma under control, although the patient can also listen to her own body and medicate without being told to do so by software. Some at the FDA may be concerned about patients becoming reliant on the software and not being as vigilant as they would be without the software.
Ideally, the agency is driven by the need for consistency. So what kind of regulatory oversight is given to decisions regarding when a drug should be recommended for administration? Presently, for the drugs, this is typically done through a new drug application which represents a high level of FDA oversight. If FDA requires significant regulatory compliance for changes in drug labeling with regard to recommended timing of administration, for the sake of consistency if nothing else, the FDA will be inclined to regulate timing recommendations done by software to a similar level.
Properly interpreting comments made by people in power, from the time before their rise to power, is always difficult. When a person is not in authority but running for office, there is a tendency to discuss issues at a more rhetorical level. We all have to understand that when someone then takes power, he is confronted by unexpected realities.
That’s not to say the prior comments should be dismissed. On the contrary, certainly the new Administration and Commissioner Gottlieb have an important message regarding the impact of regulation on much-needed innovation. While removing software such as this from FDA oversight seems unlikely, the agency is likely to take a critical look at how it can improve its regulatory processes to exert the lightest possible touch on software while still providing effective oversight. Indeed, that thought process has already begun at the FDA.
Some creative people at the FDA have asked whether software premarket clearance could be done through a process akin to TSA pre-check. The innovation behind pre-check was the realization that for frequent travelers, less intrusive security measures could be required on the day of travel if the traveler underwent more in-depth prescreening.
The software industry and the FDA are now discussing whether a similar innovation could be used to prequalify software vendors so that the product-specific premarket reviews could be lessened. The CDS Coalition, for example, has suggested that FDA consider expedited review for software as long as the software vendor commits to employ software design and development processes that will enable:
- Frequent and relevant feedback from patients, clinicians and other users.
- Timely updates as necessary to ensure safety and effectiveness in real-world deployments.
Such an approach takes one of the great strengths of the software industry, its ability to continually improve the performance of its product quickly, and uses that ability as a basis for reducing the premarket regulatory burdens.
Further, over the last several years we’ve seen a preference at the FDA for focusing on health and safety, and leaving economic injuries to other agencies to oversee. When software is sold to consumers and only embodies low risk, such as an app that uses the light a cell phone emits to treat acne, agencies like the Federal Trade Commission and even the New York Attorney General’s office are more likely to get involved than the FDA. These agencies focus on truth in advertising to make sure that consumers are protected from fraud, in lieu of requiring premarket approval. This shifting of responsibility may accelerate under the new Administration.
There will undoubtedly be much discussion about the pre-check idea before settling on a particular approach. It’s very important that small companies, which have been a source of so much innovation, are able practically to make use of this expedited process. Hopefully, the FDA will embrace its own form of regulatory innovation.