It has long been conceptually desirable for at least some medical devices to run as closed-loop systems where patient sensors feed a control algorithm, and the control algorithm drives the device based on the feedback from the sensors. This patient-based sensing is distinct from some internal sense and respond capability such as a device thermostat. Canada has an official definition of such a device: A “closed-loop system,” in respect of a medical device, means a system that enables the device to sense, interpret, and treat a medical condition without human intervention. I have not found a comparable general definition in the FDA’s codified definitions, although the term has been used with respect to an artificial pancreas.
Assuming adequate sensors and effectors, the challenge here is the complexity of the control problem and the necessary robustness of the control algorithm. That is, under how broad a range of circumstances and patients will the control algorithm produce the “right” result in the sense of optimizing the outcome for the patient? Related issues are the algorithm’s ability to detect that it is operating at or near its own validation boundaries. Of course whatever the strength of the underlying science/medicine, control theory tells us that implementing a control algorithm can be a complex task. Even when that task is accomplished, the system then has to be correctly implemented in both hardware and software. This wouldn’t be a good place for a long string of software “upgrades,” a term I love when it means fixing software that wasn’t right in the first place. A few examples of closed-loop control have been with us for some time, such as physiologically responsive pacemakers and implanted defibrillators. Similarly anti-epilepsy implanted devices also are responsive to pre-seizure indicators. There is also at least one closed-loop ventilator that I am aware of, but it is not available in the U.S., presumably for regulatory reasons. More recently, there has been discussion of closed-loop medication delivery, although in that context there is a human that closes the loop between order and delivery.
In the truly closed-loop sense, the FDA recently approved a (Class III) self-regulating propofol sedation system that reflects a cautious foray into this arena. The device in question is for certain endoscopy procedures in healthy adult patients for which it will provide minimal to moderate sedation. Beyond convenience, the system claims reduced risk and faster patient recovery time, suggesting that it is actually better than at least some humans doing the same task. Another goal of the system is to more closely match the skill level of the sedation delivery team with the actual requirements of less complex cases, which in real life apparently means to use lower skilled (i.e., lower paid) workers. Stated limitations (for at least on-label use and initial distribution) are that an anesthesiologist must be “immediately available,” and that direct users will have to have special training. The approval is also subject to post-approval clinical studies to determine how the device performs in the less fettered post-market environment as opposed to the pre-market clinical trial environment. The warnings in a product information two-pager include that in the clinical trial, there were “non-sustained, unintended episodes of deep sedation and/or complete unresponsiveness or non-purposeful response to painful stimulation.” Is this a control algorithm problem? It doesn’t say.
Beyond single close-looped devices, part of the connectivity discussion is that more than one device could be involved in a prescribed dialogue of sense and control. The challenges of developing comprehensive algorithms is then compounded by the question of who is it that defined the conversation, especially if the devices involved are from more than one manufacturer. If the conversation software doesn’t belong to any of the manufacturers ,then it is likely to itself be an independent medical device.
With the newly approved device, and perhaps with others, closed-loop control is coming into our ranks. Are such devices ready for prime time? Are we ready for them?
William Hyman, ScD, is professor emeritus of biomedical engineering at Texas A&M University. He now lives in New York where he is adjunct professor of biomedical engineering at The Cooper Union. Hyman may be contacted at email@example.com.