The NYTimes published an article recently talking about how to get safer medical devices. It raises some good points but I think it ultimately misses the mark.

According to a report from the Brookings Institute, medical devices in the United States are responsible for over 3,000 deaths per year. The CDC more or less corroborates this – Table 10 lists complications of medical and surgical care at 2,768 people/year. To put these numbers in context, the CDC report also lists deaths from motor vehicle accidents at 35,369 people/year and all firearms deaths at 33,636 people/year, which means this article is off to a depressing start. So, morbidly relatively speaking, these numbers are low compared with already highly regulated industries. However, people associated with the NHS (which treats a fraction of the patients compared to the US healthcare system, roughly 156 Million vs. 1.2 Billion) are claiming that there are more than 10,000 deaths per year due to “problems in care”, so right off the bat there’s potentially an order of magnitude discrepancy (yes, I know we’d need to unpack the inclusion criteria behind these statistics) in how severe this problem is.

So what we need then, the Author argues, is a more capable system that utilizes unique device identifiers (UDIs) in combination with a better postmarket adverse events reporting database (I’ll agree that MAUDE is a garbage fire), so not only will we better understand the epidemiology of medical device adverse events, we’ll be better able to recognize problematic devices and issue recalls prior to further adverse events occurring.

“Do for Devices what we do for Drugs.” the author argues. You can currently go online and mine the OpenFDA database for all adverse events associated with pharmaceutical products in the United States but you can’t do this for Devices “because unique device identifiers are not required to be included in standardized medical claims data.” Instead you get to go to MAUDE and try and unpack what went wrong when someone reports something like:

Picture1

Which, to an engineer like myself is about as useful as the “check engine” light in my old VW. Yes, something has gone wrong. No, I don’t know what happened and I’ll probably never figure it out. There are any number of failures that could have occurred in this specific instance and the specifics are forever lost.

So I completely agree with the Author when they argue that a stronger postmarket surveillance system is needed, and the impending European Medical Device Directive agrees too, to the extent that they don’t actually have the infrastructure to meet the demands of the surveillance system, have no mechanism for building that infrastructure before they want adoption in 2017, and argue that no medical device currently on the market in Europe complies with their expectations. Consultants have dollar signs in their eyes.

It’s possible to pretend, just for a moment, that we have a useful medical device adverse events reporting database. Combination products, those that combine a medical device and a pharmaceutical product, are often captured by adverse drug event reports, and you can view up to date information using a nifty webtool. Let’s use a common example – Epinephrine, aka the EpiPen:

Capture

Renal Injury. Device Failure. Unevaluable Event. This looks bleak.

Let’s try something else. Say, the #2 best-selling drug in the world, Humira, a treatment for rheumatoid arthritis.Capture2

Device Malfunction. Incorrect Dose. Injection Site Pain. Drug Ineffective. Wrong Technique in Drug Usage Process. Oh dear, this isn’t much better.

I selected these two drugs to prove a point – They both use an injector pen to deliver the medication and they’re subjected to the pharmaceutical products surveillance that the author alleges is needed for medical devices. You don’t have to be a medical devices engineer to have heard stories of people injecting their thumbs with an Epipen, or to think that several thousand “Device Malfunction[s]” suggests that there’s an inherent problem with the devices being used to deliver these drugs. People evidently aren’t able to use them as the manufacturer intended.

Enhanced surveillance will only let you know when there’s a problem – it won’t prevent harm from occurring in the first place and as of now, it’s not resulting in as many recalls as you’d think. In theory, ISO 14971-based risk management should minimize device-based risks but the above two tables illustrate that we’re not currently able to design away all of our problems. Or we’re ignoring them. Quite often.

So while the Sentinel Initiative has managed to identify and investigate 137 drugs, I worry that it falls down when it comes to discerning between drug side effects, device failures, and device use-events, three items that are visible in the openFDA charts above. Biologics, a class of pharmaceutical products that relies on parenteral delivery aka an injection device, are currently the darling of many big pharmaceutical company’s pipelines. Failing to discern between drug side effects and device problems, particularly as the intended delivery site of the drug (SubQ, IM, etc.) becomes more safety-critical, is going to be boon to the quality of the reported data.

What needs to be done is to establish a better understanding of the intended user population, beyond the FDA’s world-leading human factors expectations, so medical device designers can design safer and more effective drug delivery devices. Currently, as part of your Human Factors Engineering activities, you’re expected to define the Device Users prior to trying to anticipate how they can use the theoretical device incorrectly.

Capture3

If we’re failing to develop and implement appropriate risk mitigation measures, as illustrated by the number of device problems shown above, then chances are something is wrong with our source data – the definition of the intended users and an understanding of their capabilities.

There’s no shortage of users that will be on a certain medication for life, for instance Diabetics and Asthmatics. What I want to see are instrumented devices that learn how these users are actually using these devices. What habits they have, what workarounds for clunky device features they’ve found, what angle they’re holding the device at during use. Like the author wants data on adverse events, I want data on what happens before one occurs. I want medical device black boxes.

Is it possible to outfit every medical device with sensors and recording devices/transmitters? No. Nor is it cost effective when the device in question has a BOM cost of $4. But what it can do is enable designers, engineers, and researchers the ability to run pseudo-formative long-term epidemiological studies on device that are already on the market in order to better inform their next-generation of devices. It’s not until we make a better effort to close the feedback loop on postmarket surveillance that we’re going to truly realise medical devices that are safe and effective for their intended population.