How the FDA process is biased against new technology

By MHN Staff
09:00 am

Bradley_Merrill_Thompsonby Bradley Merrill Thompson

Let’s say no one had ever invented the tongue depressor, but this year, in a flash of inspiration, I came up with the idea. Indeed, let’s say there’s nothing else like it. To bring the first tongue depressor to market, it’s quite likely I would have to spend years conducting clinical trials, to be followed up by perhaps a year of FDA review. For a tongue depressor. Why? Because it’s new and novel.

Perhaps you don’t believe me. I don’t blame you, it sounds hard to believe. But I can support what I say.

The Big Picture

In the Medical Device Amendments of 1976, Congress directed FDA to divide up all then-existing medical devices and put them into one of three classes, ranging from class I for the lowest risk up to class III for the highest. In response to that directive, FDA eventually promulgated roughly 1,700 regulations that each define a then existing category of medical device. In turn, each regulation identifies the applicable class for that particular technology.

Once classified, the law provides three different corresponding levels of regulation. Generally speaking, most:

  • Class I medical devices can go right to market so long as the manufacturer observes certain quality manufacturing practices.
  • Class II devices have to go through a modest FDA review to determine whether the device is in fact “substantially equivalent” to existing devices. In a sense, this is similar to a generic drug pathway. The manufacturer doesn’t have to re-prove the fundamental safety and effectiveness of the device type.
  • Class III devices have to be proven safe and effective through a much more rigorous premarket approval process that requires substantial clinical trials.

All of that makes perfect sense. But here’s the rub. What if your device is new and novel? More specifically, what if nothing like it was in existence in 1976 and therefore nothing like it has been classified? Section 513(f) of the 1976 Medical Device Amendments says such new products are automatically in class III unless they can be fit in an existing classification. That’s right, a device can be plainly low risk but still end up in class III simply because it’s new and novel. As FDA itself observes, “devices of a new type that FDA has not previously classified … are ‘automatically’ or ‘statutorily’ classified into class III…, regardless of the level of risk they pose.”

To be fair to Congress, clearly our elected officials in 1976 were concerned about new technology and the unknown risks it might present. Further, simply as a matter of administrative functioning, Congress wanted to make sure that FDA was the gatekeeper and had the chance to say no to new technology that may present unacceptable risks. Congress couldn’t very well say that all new technology was automatically in class I, the lowest level of regulatory oversight, and let every new device go directly to market regardless of the risk it presented.

Unfortunately, though, despite best intentions, the system isn’t working. The system ends up discouraging innovative technology because changing the default class III determination for a new technology to a lower, more reasonable level is expensive and time-consuming. By starting a new technology off automatically in class III, the law plainly discourages new technology.

The Flaw in the System

The problem is the cost and time delays associated with getting out of class III. If you’ll permit me again to generalize a bit, for new technology automatically placed in class III, there are two primary ways to lower the classification:

  1. 1. The first is by filing a petition for reclassification (there are actually about five flavors of the petition, but they are all very similar.)  The goal of one of these petitions is to create a new classification for the new category of technology. Typically the petitioner is an industry coalition trying to improve the regulatory environment for a generic category of devices.  While the trade group tries to spread the cost among members of the industry, inevitably that is difficult due to the free rider problem. There are always companies that don’t want to contribute even though they will clearly benefit.
  2. 2. The second is by filing what is referred to as a “de novo” submission. De novo is Latin for "afresh" or "anew." Through a de novo submission, an individual company can pursue a new classification for its particular product. Instead of only approving the one manufacturer’s product, a de novo submission also establishes a new classification for all products like the new product. Thus, a company that does this needs to be magnanimous enough to want to help all of its competitors. It doesn’t happen very often.

Part of the reason companies avoid those pathways is the fact that they are so expensive. This is true because (1) the procedures are complicated to follow, and (2) the substantive evidence the petitioner needs to submit is similar to the evidence a manufacturer needs to submit to get a class III product approved. Often the company needs to submit clinical trials or something similar. Indeed, in many ways the reclassification procedures are more difficult than a class III premarket approval application because instead of proving that a given manufacturer’s product is safe and effective, the petitioner must prove that all of the products in a given category are safe and effective.

For its part, FDA tries to be as accommodating as it can be. Above I explained that the automatic class III designation is a problem for those items that are both new and novel. That may seem redundant, but it’s not. There can be new inventions that are simply incremental improvements. For incremental improvements, FDA can get quite creative in trying to fit them into existing classifications. For example, relevant to clinical decision support, here are a few of the existing classifications into which FDA could shoehorn new standalone software:

  • A calculator/data processing intended to store, retrieve, and process laboratory data. (21 CFR 862.2100)
  • A cardiac monitor used to measure the heart rate from an analog signal produced by an electrocardiograph, vectorcardiograph, or blood pressure monitor. (21 CFR 870.2300)
  • A medication reminder intended for medical purposes to provide alerts to patients or healthcare providers for pre-determined medication dosing schedules. (21 CFR 890.5050)
  • An automated blood grouping and antibody test system used to group red blood cells and to detect antibodies to blood group antigens. (21 CFR 864.9175)
  • A calculator used to calculate pulmonary-function values based on actual physical data obtained during pulmonary-function testing. (21 CFRR 868.1880)
  • A predictive pulmonary-function value calculator used to calculate normal pulmonary-function values based on empirical equations. (21 CFR 868.1890)
  • A computer that can be programmed to compute various physiologic or blood flow parameters based on the output from one or more electrodes, transducers, or measuring devices. (21 CFR 870.1425)

There are roughly speaking about 60 classifications that could potentially house clinical decision support software, but they are each focused on products that have been around typically for decades.

As you can see, for a developer of new clinical decision support software, those are pretty slim pickings. And when something is truly novel -- when something truly hasn’t been created before -- there is only so far that FDA can go in fitting that new square peg into one of those old round holes.

The Impact on Novel Software

I’ll give you an example of the route through FDA that the maker of new and novel software must traverse. A company called MELA Sciences, Inc. brought to market a product called MelaFind. MelaFind is a software program that uses algorithms to analyze images of moles and other suspicious skin lesions to assess whether they might be melanoma. Because it is new and novel, following the statute FDA placed the software in class III and required the company to pursue premarket approval.

In seeking approval, the company requested the following relatively bland intended use for MelaFind: “when a dermatologist chooses to obtain additional information for decision to biopsy.” In its approval application, the company explained that MelaFind is only for use by trained dermatologists who are otherwise capable of assessing the lesion to determine the need to conduct a biopsy.

There is a hardware component to the system, in that there is a camera that takes 10 multispectral digital images that are used for the analysis. The camera by itself is probably only a class II medical device because there are other similar cameras out there already. Beyond the fancy camera, however, the system is basically a computer with software.

According to the FDA’s Summary of Safety and Effectiveness for MelaFind, the company was granted expedited review status in October 2006. That might sound encouraging, until you also learn that the product was approved on November 1, 2011, more than five years later. To get that approval, the company had to conduct six clinical trials, that together included almost 9500 images from almost 7000 patients, at 40 clinical study sites over the course of seven years.

The actual time for review at FDA commenced on June 2009, and took over two years. During the review, the company got so frustrated with FDA that it filed a citizens petition in May 2011, trying to force the agency to act on its application.

How many software companies would have the resources to pursue this level of FDA approval? And how many patients could have been helped over this period of time?

Attempts at Reform Have Failed

Both Congress and FDA are aware of this issue. In the past, their efforts to address it have focused mostly on trying to make the de novo reclassification process quicker and cheaper to use. But their efforts have fallen far short of solving the problem.

Congress’ first attempt to address the problem was the creation of the de novo process itself in 1997. Unfortunately, as explained above, the process had so many problems, and was so time-consuming and expensive to pursue, that not many companies even attempted to pursue it. For the 15 years from 1997 until 2012, only 75 organizations successfully navigated the de novo process at FDA’s Center for Devices and Radiological Health.  That is on average five per year. Not many, given the rate of innovation.

Recognizing that the initiative had fallen far short of its goal, in a second attempt to solve the problem Congress amended the de novo law in 2012 in the Food & Drug Administration Safety and Innovation Act. Section 607 of that Act did away with the previous requirement that a company needed to pursue FDA clearance first and fail before being able to pursue the de novo reclassification process.

Even with that improvement, fundamentally the de novo process still requires that an applicant demonstrate with clinical evidence that the product can be regulated as safe and effective in either class I or class II.  That means providing FDA with evidence from clinical trials to address such things as:

  • The safety and effectiveness of the device.
  • Identification of each risk associated with the device and the reason for each risk (tracing back to risk analysis, clinical testing, etc.).
  • The proposed mitigations or controls for each risk based on the best available information at the time of the submission.

Indeed, just as with a class III device, FDA encourages companies that plan to submit a de novo petition to meet with FDA first to review the clinical trial protocols before commencing any needed studies. Thus, the total process for developing the evidence and obtaining reclassification remains long.

The Dysfunctional De Novo Process

How long? Let’s look at three software examples since the 2012 legislative improvements. Consider:

  • FDA spent over 24 months reviewing a Computerized Cognitive Assessment Aid, a prescription device that uses an individual’s scores on a battery of cognitive tasks to provide an interpretation of the current level of cognitive function. The Computerized Cognitive Assessment Aid is used only as an assessment aid to determine level of cognitive functioning (a) for which there exists other valid methods of cognitive assessment and (b) which does not identify the presence or absence of clinical diagnoses. The Computerized Cognitive Assessment Aid is not intended as either a stand-alone or adjunctive diagnostic device.
  • FDA, in 15 months, reviewed and reclassified software used to analyze images of patients to estimate how much blood they lost.
  • In a period of only 4 months, FDA reviewed and reclassified into class I “continuous glucose monitor retrospective data analysis software” intended to analyze and correlate retrospective data from a continuous glucose monitoring device. In connection with that decision, FDA made it clear that the software must be intended for display of retrospective glucose data and not for real-time display of glucose results. Further, the software could not be used for (a) making treatment decisions, (b) calculating insulin or other drug doses, or (c) controlling insulin pumps or other drug delivery systems. Such software is very closely related to medical device data system software that are now unregulated. This decision, while apparently only requiring four months, was related to a class III device with a multiyear FDA history.

Outside of software, the experience is pretty similar.

  • FDA took 19 months to review a pen that a surgeon could use during surgery to mark important lines, like where to cut.
  • FDA spent about 48 months reviewing a treatment for itching due to mosquito bites using limited output transcutaneous piezoelectric stimulator.
  • FDA took about 22 months to review an oral electronic vision aid in conjunction with other assistive devices (e.g., the white cane, guide dog, etc.).  The system translates images of objects captured by a digital camera into electro-tactile signals that are presented to the user's tongue. It’s pretty cool.
  • FDA spent approximately 33 months reviewing cloth used to wrap a patient’s foot to apply pressure in a way that would moderate the effects of restless leg syndrome. When the review was complete, FDA placed the sock in class I.

In each case, it’s important to underscore that the times listed above are only the FDA review times for the final submission, and do not capture the more considerable amount of time that the petitioners may have spent developing the evidence to submit to the agency. Further, stated review times can be deceptive because the review times listed on the agency’s website are not necessarily for the first submission by the company. Indeed, many of the de novo petitions followed fairly lengthy 510(k)s that failed to gain FDA clearance and so were converted to de novo submissions.

More than just the delay in getting a response, the de novo process has limited appeal because it requires manufacturers to act against their own financial interest. As already observed, putting together a de novo submission requires collecting data on the safety and effectiveness of the technology, and often generating new data to justify a lower classification. That costs substantial money. While that by itself is a deterrent, the real problem is that reclassification helps every company that wants to sell a given product. Pursuing reclassification means directly helping your competitors, without your competitors having to financially support the reclassification. That’s why even though the de novo process has been around for almost 18 years, only 135 companies have used it.

Informal Consultations Are Not the Answer

I know that some at the agency would very genuinely say, if you are developing software and you’re having trouble figuring out the classification, please come talk to us. Perhaps we can figure something out.

That is not a viable solution for at least four reasons.

  1. 1. The existing classifications can only be creatively extended so far. It is true that FDA, for example, has been using a classification for laboratory analytics as a classification for software that has nothing to do with laboratories. But I think FDA would agree that it can only finagle the wording of the classification regulations to a certain point. For the vast majority of the regulated CDS presently being developed, there simply are no viable classifications.
  2. 2. Taking a month or two to arrange a phone call with the FDA simply does not allow for the quick developmental work that most of these projects require. When a development team is working on a piece of software, and they have a fundamental question about how the piece of software will be treated at FDA, the regulatory affairs professional cannot simply say, “I have no idea, let me arrange a meeting with FDA and get back to you in a couple of months.” That simply is not the pace at which innovative software businesses work.
  3. 3. It is hard to base important company decisions purely on oral statements by FDA regulators. As a government agency, FDA is comprised of many people, and not surprisingly some of those people disagree with others from time to time. Oral advice from one or even a handful of FDA employees does not mean that the advice given will remain true. These decisions need to be grounded in more than just oral advice.
  4. 4. That’s not a fair system. It replaces the rule of law with the rule of people. If everything is up to the agency in its discretion, there is very little practical oversight of the agency possible. When the agency commits a view like the classification of the new technology to writing, stakeholders throughout government, patients, providers and industry all have a chance to express their viewpoints. Relegating these decisions to informal conversations means that sort of public oversight is lost.

For all of these reasons, a system based on “just come and ask us what we think” simply doesn’t work.

So What’s The Solution?   

After trying to think through solutions that might help, an idea I floated to some officials at FDA was: let’s pretend this is 1976 again. Instead of making all of these manufacturers pursue reclassification petitions at great expense, could the agency come up with new, more up-to-date classifications at its own initiative? In other words, I proposed that FDA survey the landscape of new medical technology, and come up with some new classifications to accommodate that new technology.

Of course there’d have to be collaboration between FDA and industry in order to figure out what exactly is on the horizon and to get a better understanding of the risk implications of the new technology. But it seems to me if FDA led the charge as a part of a comprehensive refresh of its classification regulations, the agency could tear down at least some of the walls blocking innovative, new software. Further, I wasn’t proposing that FDA classify the really exotic and potentially dangerous stuff, but rather low risk, regulated software that has been developed recently and that we can anticipate seeing in the next couple years.

Based on the reaction I got, you’d think I was the funniest man alive. Like much in life, I guess it all comes down to money. And FDA explains that they don’t have any extra. Congress has been underfunding FDA for decades, and the agency only stays afloat now by virtue of user fees the industry pays to get submissions reviewed. The FDA flat out says they do not have the money to undertake the work that would be involved in creating new classifications. Further, I suspect FDA also has some institutional reluctance to committing to classifications of relatively novel technology.  The agency will, with some justification, always worry about unknown risks.

A Specific Proposal

Funny idea or not, I think FDA needs to help fix this.

On the one hand, there is certainly logic behind having industry drive the classification process. Industry knows what technology lies on the horizon. On the other hand, FDA in some ways has broader vision, seeing all the new technologies that industry brings to the agency.  As the federal regulator, FDA also plays a leadership role, which would allow the agency to serve as a natural coordinator.

In urban planning, experts often develop what they refer to as a “master plan.” The purpose of such a plan is to bring a holistic vision to how a city should develop and solicit public input on that vision. The completed plan then encourages private development in a coordinated fashion.

We need FDA to develop a master plan for the classification of health information technology (HIT) type medical devices. This master plan could lay out the landscape of HIT- type medical devices, give guidance to industry on how the agency views the risks associated with the different categories of software, and then solicit proposals from industry for reclassification. Importantly, this master plan also has to address the existing evidence of safety and effectiveness for the different categories of software, and identify where evidence is already sufficient and where it is lacking. By doing so, this master plan would help industry understand where FDA is in its current thinking on classification, and would facilitate both parties working together to achieve mutually beneficial reclassification of important new technologies efficiently, in the shortest possible time. Geez, it sounds like I ought to be on the campaign trail.

The Bottom Line

So there you have it. If an innovator comes up with something truly novel for use in the diagnosis or treatment of disease, the law places it automatically in class III even if it is patently low risk. That means, under very specific and prescriptive premarket approval regulations, the innovator needs to follow a pathway that involves substantial clinical trials and potentially a long wait for FDA approval. Or the innovator needs to petition FDA to come up with a new classification for the product that, not coincidentally, involves much the same level of evidence as getting a class III product approved. Either way the innovator has to spend lots of money and encounter great delays. The innovator is penalized simply because her idea is new.

Frankly, the issue is acute because the classification system ossified long ago. Over the last couple decades, FDA has created very few new classifications, especially for software. So the system we have now is quite similar to the system we had back in the 1980s.

Of course during that same time, the technology has not stayed the same. I am fortunate to work with many companies that are developing medical software, and I see incredible innovations coming down the pike. It absolutely breaks my heart to see so many exciting technologies that are basically in limbo because the developers don’t know what the required regulatory pathway is for their new technology.

If this approach ever made sense, it doesn’t now. We need a new system.  In part two of this post, I’ll talk about the implications of the system specifically on mHealth and on the public health generally.


The latest news in digital health delivered daily to your inbox.

Thank you for subscribing!
Error! Something went wrong!