Andy Rogers talked with Mike Acosta, EVP/Head of Compliance at Coagusense, and later recapped some of the lessons learned with Senior Electrical Engineer Jake Cowperthwaite. Andy and Jake have an informative discussion about how to define performance requirements when you’re aiming for FDA approval.
Speed-to-data determines go-to-market success for medical devices. You need to inform critical decisions with user data, technical demonstration data, and clinical data. We interview med tech leaders about the critical data-driven decisions they make during their product development projects.
Hey, everybody, welcome to MedTech Speed to Data.
I'm your host, Andy Rogers from Key Tech.
Thanks for joining once again.
Episode 18 here.
We're going to be talking
about defining performance requirements.
We have Jake Cowperthwaite here on the line.
Jake, welcome back to the show once again.
Hey, Andy, it's nice to be back on the podcast.
Great. Yeah.
So we're going to talk also about the episode
we just recorded, episode 17,
Mike Acosta from CoaguSense,
where he described his PT/INR
at-home diagnostic test
for those those are with patients
who are taking Warfarin.
He talked about his next generation platform,
the Gen3 platform that they're developing.
It was pretty interesting with that product
how they're actually making it simpler
than their Gen2 product.
And by eliminating
most of the connectivity features
that they had in their Gen2
for their Gen3,
listening to their customer,
the elderly patients
who really just want a simple device to use.
The other highlight from that episode,
I will point out,
was kind of the
the impetus for this episode
is some of the challenges
they're having with getting approval
from FDA related to
verifying their performance requirements.
This is, you know, making sure the product is
equivalent to their on market predicate product,
their own product.
They've had to go back
and forth a few times with FDA
presenting more bench data.
So we thought it'd be a good time
to step back and talk with you, Jake.
You've managed many projects
where we've taken products through regulatory
approval through one way or the other,
either an internal Key Tech product or,
you know, externally with clients.
And it always boils down to what is the product?
What is the product doing?
What are the performance requirements?
And so today, Jake, we want to walk through
what we concluded
to be kind of the primary considerations
when defining these performance requirements.
As you're developing
your product requirements document
and the considerations we have, there's three.
Step one, make sure the performance requirement
is objectively verifiable,
straightforward, clean cut.
The second,
you know, what statistical data do
you need to collect to prove
that the requirement is met?
You know, how many prototypes will you need?
What does the data actually look like?
And then the third consideration
when you're defining
your performance requirements
is related to regulatory approval.
What what considerations do you need to make
as you're
defining these requirements
so that you'll get seamless regulatory approval?
So those are the three considerations.
Number one, objectively verifiable.
Number two, what statistical data.
How many prototypes will
you need to test and prove
that the product actually meets
those requirements?
And the third,
you know what regulatory considerations
are there as you're drafting
this very critical requirement,
your performance requirements.
So let's get going, Jake.
So walk us through,
what does it mean
to have
a requirement, be objectively verifiable?
You know, this is the first consideration.
So when you think about requirements
that are objectively verifiable,
there's four different types
of verification that we use at Key Tech.
There's testing where you run a test
and take measurements.
There's demonstration
where you have the device, perform
a function and, you know, show that it can occur
as expected.
There's analysis
where you,
maybe do calculations or simulations
and you analyze whether the requirement is met.
And then there's inspection
where you're just doing like a visual inspection
to make sure that something is
exists as it should.
So if you can't use one of those four methods
to prove a requirement,
that's probably not objectively verifiable
and you need to rethink how it's written.
Does that make sense, Andy?
It does. It does. And I would say,
in my experience,
aren't most performance requirements
verified through testing?
Yeah.
Most performance
requirements are going to require testing.
Just Jake, for our audience,
what's an example of a well-written requirement
that's objectively verifiable versus one
that's poorly written?
I guess I'll start with the poorly written one.
We see it all the time.
It's a requirement
that says the device shall be easy to use.
That may work well in a higher level requirement
set like a user needs.
But in product requirements,
you want to be more specific than that
because you can't really can't verify.
Something is easy to use.
You can validate it.
You can do user studies
and obtain feedback from users.
But you really can't,
you know, run a benchtop test
and decide, okay, this is easy to use.
So a well-written requirement
is going to be very specific
and it's going to have clear
acceptance criteria.
So to just pick an easy one.
You know, Key Tech,
we deal with pumps a lot and flow rates
and things like that. So a requirement might be,
you know, an aspiration pump shall have,
you know, X flow rate
within, you know, plus or minus Y percent.
It's very clear.
So you have, you know, a flow rate
and you have acceptance bounds.
So you run a test,
you prove it within that range. Great.
You move on.
But again, if you don't have
let's say you don't have
performance bands, let's say
the flow rate shall be X.
You may run the test.
It may not be, you know, exactly as stated.
You know what sort of limits
are allowed to consider it passing.
So you need to include that as well
with your requirement.
So just real quick on that point
with limits,
is it a fair statement to say
that the limits of a requirement
are usually driven by the risk
of going outside of that limit?
So you mentioned, you know,
volumetric accuracy, for example.
There's really
three different values to think about.
And I guess this is another trap
that sometimes folks fall into with requirements.
So going back to the aspiration case,
let's say that for physiological purposes,
you need to be 5% accurate.
So as long as you're within plus or -5%,
you're going to get the
same physiological effect.
But maybe for safety purposes,
plus or -10% is fine.
So that kind of gives you,
you know, wider bounds.
So what we see sometimes is
a client might find a vendor
who's advertising a pump
that, let's say is plus or -1% accurate.
So it's even better than they need.
And the client might think, great,
I'm going to use this pump.
I'm going to set my requirement at plus -1%.
So that's even better than they needed
to perform for physiological reasons,
you know, not to mention safety.
But so what they've done is
they've set a narrower requirement.
They're using this pump.
And, you know,
maybe the vendor spec
is based on specific testing,
you know, on a batch at a certain temperature.
And, you know, maybe it can meet
that performance criteria.
But then you get later on in the project
and you learn, you know,
for that application, it's not great.
And all of a sudden
you've kind of trapped yourself
because you've over-spec’d your requirements.
It would really you need to be testing to 5%.
You're trying to beat something
that's tighter than that.
So that's something we see a lot.
Yeah, that’d be an example of market
requirements, maybe
pushing down to product requirements
that are unnecessary.
Exactly. Yep.
Gotcha. Okay, cool. Great.
So now going forward,
we'll make our requirements
objectively verifiable,
particularly on the performance side. Right.
Which is a good segway
to the next consideration, which is
before you locked in your performance requirements,
make sure you understand what
how many prototypes you need
and what sort of statistical confidence
you need in meeting
and demonstrating that performance requirement.
So can you talk a little bit
about your experience there?
So performance testing is really time consuming
and you're kind of making your luck upfront.
So I guess what we just talked about
was how you write a requirement
and if you write it
in a nice, clean manner,
it might make it easier to test
than otherwise would be.
But that’s the testing part.
The other part is the sample size.
So, you know,
you might have a really easy test to run,
but if you have to have a high sample size
it could be extremely time consuming.
And what we see in terms of sample size
is it's usually risk based.
So what sample do you need to know within,
you know, some confidence
that you're going to meet
some reliability bounds or acceptance
criteria.
And that usually drives everything
if it's a simple test
that's kind of black and white, you know,
that the pumps are going to turn on or it's not.
Then, you know, one sample might be fine,
but if it's a,
really sensitive test
where you're trying to determine accuracy
across, you know, different
consumables, maybe different lots,
then you have to set up a much larger study
and use a much higher sample size.
So there was one project
I'm aware of here at Key Tech,
where we actually hired.
So in lieu of
tens or potentially hundreds of prototypes,
in lieu of that,
we developed this sophisticate weighted
design of experiments plan
and got the confidence that we needed
with less prototypes.
When would that make sense Jake?
So what we've done is
relatively early in development.
So maybe a late alpha or a beta device.
And what a do you do that
that same experiment will do.
Is it will help you understand sensitivities.
So at the time, I think there were
seven critical
variables.
And we want to understand,
you know, the impact of each.
So we hired a consultant, he came to Key Tech.
I think we met for a couple of days
and he helped us design this experiment where
if you take I don't know, you know,
five or ten devices
and test them in different environments,
with the variables set differently,
you can accumulate this huge pile of data
and process it and tease out
what the various sensitivities are.
And the great thing about that
is you're getting it all at once.
And so instead of setting up,
seven independent studies
to look at each variable specifically,
we could just run this, you know,
kind of one massive study
and obtain everything at once.
And that was really helpful
because at the end of it,
we were able to understand, you know, okay,
you know, this variable barely changed.
So, you know, we we ran it at the extreme
extreme of the range for that variable
maybe in a cold environment,
maybe in a warm environment,
maybe with a different person
operating the instrument.
And we could see, okay, you know, it
basically remained flat,
whereas some of the other variables
we could see, you know, very significantly.
And that told us that those variables
that had more variation needed to be studied
more extensively during the project.
Gotcha, and there's real value to be had there.
I mean, you're not building
multiple prototypes
that for some of these complex platforms
can be 50, 75K each.
So I can see the value there
and also just the general value of understanding
which variables are of interest
to then beat on with those small number of prototypes
you've already built.
So it made sense in that in that case.
Absolutely.
And I guess the the other factors
we could automate everything.
So, you know, with automated testing,
not only have you reduce the number of tests
you need to run,
you're also more efficient
because you can set it up over, you know, a week
and then come back and look at the data.
So it was extremely efficient.
This is also another,
I think, example of where marketing might drive
product requirements.
But I guess in your experience, where have you
seen these confidence levels come from?
Like you need 99% confidence,
you're going to hit this requirement.
Where does that come from?
Some of our global clients
will have their own internal procedures
and it will be risk based.
So, you know, minor, moderate, major.
And if it's a major level of concern
or major risk,
then they're going to require high confidence,
you know, high confidence that they're meeting,
you know, their specified requirement.
If it's, let's say, lower,
minor concern,
maybe you don't need that high confidence
because if it falls a little bit outside the range,
it’s not going to have any clinical impact
or, you know, affect the patient in any way.
And then the other end of the spectrum,
I guess, would be the startups
and smaller companies
who don't have those internal policies.
And for that, they're going to be,
you know, hopefully looking to
what other other companies have done.
They're going be talking to the FDA
to try to get some input
and establish their sample sizes.
Yeah, Jake, you just gave away the third
consideration, which is making sure you have FDA
or general regulatory
buy-in on your performance requirements
before you go submit your design history file.
So talk a little bit more
about getting this, you know, FDA
and regulatory buy-in
on these critical performance requirements.
So the first thing you should do is
internally use your own judgment
of what you think the sample size should be.
And that's a combination of
engineering expertize.
So understanding sensitivities
as well as clinical expertize.
So assessing the risk.
So you should come up with a plan internally.
You know, what sample size do you plan to have?
What testing do you intend to perform?
And then you should get in touch with the FDA.
And there's a thing called pre-sub meetings
or also called queue-sub meetings.
That’s where you would present your plan to the FDA
and have them provide feedback.
You know, do they agree with your sample size
and your test approach
or whether they'd like to see you
do something different?
And in that meeting, I'm
assuming you want to have data in hand, right?
The preliminary prototype
data showing that, okay, look,
we're already meeting the requirement.
This is our plan as we go in the formal V&V.
You'd like to have some preliminary data.
Yeah, especially if you can show.
Well,
you know, going back to the same experiment,
if you can show studies you've done
that have established
where the device is sensitive
and where it's not,
I think that's, you know, really
helpful in justifying your
sample size decisions to FDA.
Going into a pre-sub meeting
Jake with the FDA,
what are they looking for it at a high level
with your product,
you're trying to get approved.
So the goal is safety and effectiveness.
And in a PRD,
you're going to have all sorts of requirements.
You're going to have some related to safety.
You know,
to pick an example, complying with IEC 6601
for electrical safety, that would be an example
safety requirement.
You might have effectiveness request.
Well, you will have effectiveness requirements
and those might be based on a predicate device.
So if you're going to go
the 510K route,
you're going to look at a similar product.
What sort of performance does it have?
You know, what are the accuracy criteria,
what's the resolution, those sort of things?
You're going to want your requirements
to be as good or better. So those are critical.
But you're also going to have other requirements
that may come from marketing.
They may help sell the product.
They're not really in the safety and effectiveness,
but they're going to differentiate the product
and make it more successful on the market.
So so when you go into the meeting
with the FDA,
the focus really will be on safety and efficacy.
And you want to make sure
you have a good plan for
the sample size to test those
as well as the procedure to test those.
And then the other ones the more
marketing related requirements
that don't really fall
into those two categories.
They're going to have less emphasis.
But, you know,
because they're part of your design endpoint,
you still need to meet them.
It just may not be as rigorous a testing.
Yeah, I mean, I've never actually been
in one of those pre-sub meetings,
but the only
feedback I've heard is go into
those meetings prepared and
communicating your plan and not asking
kind of high level abstract questions
because they'll see right through that
I suppose.
Yeah, I think you can waste a lot of time
if you're not presenting a plan to begin with.
It's not a brainstorming session,
you know, where FDA is helping you
figure out how to test your product.
It's more
you've done your homework
or you're proposing a plan
and you're getting feedback on that.
All right, Jake,
thanks for your words of wisdom here today.
You know, related
to defining performance requirements
and one of the key requirements
in a broader product requirements document,
which is linked in our bio here
down in the episode here
where we have a whole
flowchart for defining requirements
as well as a template that you can download
for a product requirements document
but related to performance requirements,
which are the most critical.
Jake, you outlined the three main considerations.
The first, absolutely.
The product or product requirements
need to be objectively verifiable.
Secondly, you need to have a plan
for the number of prototypes to build
so that you can get the appropriate
statistical confidence
that the product will meet
those performance requirements in the market.
And lastly,
arrange for a pre-sub meeting with FDA
where you outline your requirements
and they're going to focus primarily on the
the performance requirements
related to safety
and efficacy of the product.
So that's it here from Key Tech.
Again, check out the link below to download
the Product Requirements Document Template
or look at the flowchart
for how you actually define
various requirements.
Until next time.
Thanks everybody for tuning in. Appreciate it.
Thanks Andy, take care.