Late last year, the FDA released its list of ten guiding principles for good machine learning practice (GLMP) in medical devices. That’s the good news. Right?
Bad News:The guidelines are still a federal document. You know, the type we all struggle through only when we have copious amounts of time, patience, and caffeine.
Good News:It is much, much shorter than documents past, perhaps to appeal tl;dr (“too long; didn’t read”) generations. We imagine that the brief was something like: “All ten guidelines must fit on one page”. And they did it.
Bad News:They cheated.
Short… by FDA StandardsThe font has been shrunk to cram over 700 words onto a single page, leaving no room for an introduction. Leaping through the obvious loophole (“You never said the intro had to fit on the same page!”) readers have been gifted an additional 750-word preamble. The first page does have a nifty table outlining the guiding principles in shorter form. Unfortunately, to make that fit they compressed the font to 5 or 6pts, barely readable to the squinting eye. |
Fun Fact For your audiences to understand at least 90% of what you read, sentences should be kept to 14 words or less.
 Once a sentence hits 43 words or more, 10% comprehension drops to less than 10%.
In a bow to irony, the principle that advocates clarity in messaging begins with a 65-word sentence. |
So here is our gift to you (and to the FDA):
Assemble a multi-disciplinary team to ensure device safety and effectiveness.
Observe good engineering and cybersecurity practices.
Use test data and clinical study participants who represent your target population.
Prepare independent datasets for training and testing/validation.
Employ best practices to attain the most useful, clinically relevant reference datasets.
Use ML findings to create a safe and effective model design.
Supplement ML findings with human experience and insight.
Verify that the device performs as expected for intended patients and environments.
Provide easy access to clear information, instructions, and updates.
Continue gathering performance data on deployed devices.
And we get it. It’s the great paradox of having to be both concise and precise that governments struggle with and lawyers flat out ignore.
However, in medical device development and manufacture, every miscommunication can snowball into a big, costly problem. Therefore we take every step to ensure that our clients clearly understand every conversation, document and report.
In the highly competitive industry of medical devices, outsourcing is almost required to stay ahead. Its benefits are evident — multidisciplinary...
A good contract manufacturer fills the gaps in their partners’ skills, capacity or capabilities. In other words, they deliver what you pay for. A...
Microelectromechanical systems (MEMS) sensors are the backbone of connected (IoT) medical devices. Their mechanical workings detect physical stimuli...