Latest / Let's Talk Risk! Podcast / Executive Brief: PCCPs as a Competitive Strategy

Transcript

Executive Brief: PCCPs as a Competitive Strategy · Let's Talk Risk! Podcast

• 0:00
  SPEAKER_01: Listen, PCCPs transform regulatory filings from a reactive bottleneck into
• 0:05
  SPEAKER_01: a pre-authorized commercial iteration engine.
• 0:08
  SPEAKER_00: Absolutely.
• 0:09
  SPEAKER_01: And if you are sitting in the C-suite making decisions about ROI or time to market
• 0:14
  SPEAKER_01: or capital allocation, you need to understand this.
• 0:16
  SPEAKER_00: Yeah, it's a massive shift.
• 0:18
  SPEAKER_01: Right.
• 0:18
  SPEAKER_01: So welcome to today's deep dive.
• 0:21
  SPEAKER_01: Our mission today is to explore exactly how predetermined change control plans,
• 0:25
  SPEAKER_01: or PCCPs,
• 0:27
  SPEAKER_01: create this incredible competitive moat and drive capital efficiency
• 0:30
  SPEAKER_01: for your business.
• 0:31
  SPEAKER_00: It really is a completely new way to operate.
• 0:33
  SPEAKER_01: It is.
• 0:34
  SPEAKER_01: We are going to pull apart a stack of FDA guidance documents,
• 0:36
  SPEAKER_01: some deep industry analyses,
• 0:38
  SPEAKER_01: and really show you the ground truth mechanics of how this works.
• 0:42
  SPEAKER_01: But to really grasp the sheer magnitude of this,
• 0:45
  SPEAKER_01: I think we have to look at why the FDA fundamentally changed their own rules.
• 0:49
  SPEAKER_00: Right, because historically, the FDA operated on a system that was built entirely for hardware.
• 0:54
  SPEAKER_01: Metal and plastic.
• 0:55
  SPEAKER_00: Exactly.
• 0:55
  SPEAKER_00: The traditional 510k pathway required a company
• 0:58
  SPEAKER_00: to prove their new device was substantially equivalent to something already on
• 1:02
  SPEAKER_00: the market.
• 1:02
  SPEAKER_01: Which they call the predicate device, right?
• 1:04
  SPEAKER_00: Right, the predicate.
• 1:05
  SPEAKER_00: And if you changed your physical device,
• 1:07
  SPEAKER_00: you filed the paperwork, and you waited, you know, 60 to 120 days.
• 1:12
  SPEAKER_00: Which was fine for hardware.
• 1:14
  SPEAKER_00: Sure.
• 1:14
  SPEAKER_00: But as medical technology became heavily software dependent,
• 1:17
  SPEAKER_00: that entire framework just broke down.
• 1:20
  SPEAKER_01: Well, because software engineering runs on, what, two-week Agile sprints?
• 1:24
  SPEAKER_00: Yeah, exactly.
• 1:25
  SPEAKER_00: But the FDA was running on a four- to six-month waterfall review cycle.
• 1:29
  SPEAKER_01: Wow.
• 1:29
  SPEAKER_00: So it meant every single time an AI company,
• 1:32
  SPEAKER_00: you know,
• 1:32
  SPEAKER_00: tweaked an algorithm to make it safer or maybe more accurate,
• 1:36
  SPEAKER_00: they had to hit pause on commercialization.
• 1:38
  SPEAKER_01: Just completely stop.
• 1:39
  SPEAKER_00: Right.
• 1:39
  SPEAKER_00: Stop and jump back into the queue.
• 1:41
  SPEAKER_01: And that friction, that exact pain point is what Congress addressed back in December 2022.
• 1:46
  SPEAKER_01: Yeah.
• 1:47
  SPEAKER_00: When they passed the legislation giving the FDA the statutory authority
• 1:50
  SPEAKER_00: to authorize these PCCPs.
• 1:53
  SPEAKER_01: And the mechanism here is what really matters for you, the executive listener.
• 1:57
  SPEAKER_00: It is.
• 1:58
  SPEAKER_00: Because a PCCP basically allows a company to completely skip
• 2:02
  SPEAKER_00: that standard waiting period for very specific pre-planned updates.
• 2:06
  SPEAKER_01: Right.
• 2:07
  SPEAKER_00: You establish the testing plan up front,
• 2:09
  SPEAKER_00: the FDA signs off on the methodology,
• 2:11
  SPEAKER_00: and then from that point forward, you just execute and release commercially.
• 2:15
  SPEAKER_01: Without ever asking permission again.
• 2:17
  SPEAKER_00: Exactly.
• 2:17
  SPEAKER_01: I want to contextualize this because if you are viewing regulatory strategy
• 2:21
  SPEAKER_01: as purely,
• 2:22
  SPEAKER_01: I don't know,
• 2:23
  SPEAKER_01: defensive compliance,
• 2:25
  SPEAKER_01: that is a mindset that will get a company killed in today's market.
• 2:28
  SPEAKER_00: Oh, 100%.
• 2:30
  SPEAKER_01: Think of the old 510k system like applying for a single entry travel visa.
• 2:34
  SPEAKER_00: Right.
• 2:35
  SPEAKER_01: Every single time you want to cross a border,
• 2:37
  SPEAKER_01: you have to fill out the application, submit it to the embassy, wait in line.
• 2:41
  SPEAKER_00: Endless waiting.
• 2:42
  SPEAKER_01: Yeah.
• 2:43
  SPEAKER_01: And your travel plans are completely paralyzed until they literally stamp your passport.
• 2:47
  SPEAKER_01: But a PCCP operates more like securing a multi-entry fast pass visa
• 2:52
  SPEAKER_01: for your entire product roadmap.
• 2:55
  SPEAKER_00: I like that analogy.
• 2:56
  SPEAKER_01: Thanks.
• 2:56
  SPEAKER_01: I mean, you do the heavy lifting of proving your identity and your security protocols once.
• 3:02
  SPEAKER_01: And after that,
• 3:03
  SPEAKER_01: you have the freedom to move,
• 3:04
  SPEAKER_01: to iterate and cross that commercial border the very moment
• 3:07
  SPEAKER_01: a software update is ready.
• 3:09
  SPEAKER_00: But the tradeoff for that freedom is extreme rigor up front.
• 3:12
  SPEAKER_01: Right.
• 3:13
  SPEAKER_01: They don't just hand these out.
• 3:14
  SPEAKER_00: No, no.
• 3:15
  SPEAKER_00: The FDA guidance is incredibly strict about the anatomy of this fast pass visa.
• 3:19
  SPEAKER_00: You can't just vaguely ask for flexibility.
• 3:22
  SPEAKER_00: You have to engineer this really rigid three part structure.
• 3:26
  SPEAKER_01: Okay, let's break that structure down.
• 3:28
  SPEAKER_00: Sure.
• 3:28
  SPEAKER_00: So every authorized PCCP must contain a description of modifications,
• 3:33
  SPEAKER_00: a modification protocol, and an impact assessment.
• 3:36
  SPEAKER_01: Got it.
• 3:36
  SPEAKER_01: Now, looking at the source documentation,
• 3:39
  SPEAKER_01: the description of modifications seems fairly straightforward.
• 3:43
  SPEAKER_01: It's really just outlining the exact boundaries of what you intend to change.
• 3:47
  SPEAKER_00: Yeah, it's the what.
• 3:48
  SPEAKER_00: But the modification protocol, that's the how.
• 3:51
  SPEAKER_00: That is the actual engine of the filing.
• 3:54
  SPEAKER_01: Because it forces the engineering and regulatory teams to basically merge their thinking?
• 3:59
  SPEAKER_00: Exactly.
• 4:00
  SPEAKER_00: It dictates precisely how the performance of any future change is going to be evaluated.
• 4:05
  SPEAKER_01: Like down to the statistical level?
• 4:07
  SPEAKER_00: Oh, absolutely.
• 4:08
  SPEAKER_00: You have to lock in your verification procedures,
• 4:10
  SPEAKER_00: the exact statistical thresholds your software has to hit,
• 4:14
  SPEAKER_00: your rollout procedures, everything.
• 4:16
  SPEAKER_01: Wow.
• 4:17
  SPEAKER_00: And then you arrive at the impact assessment,
• 4:20
  SPEAKER_00: which honestly is where a lot of companies really stumble.
• 4:22
  SPEAKER_01: Because it forces a company to look at cumulative risk, right?
• 4:26
  SPEAKER_00: Yes.
• 4:26
  SPEAKER_00: The impact assessment asks a very specific mechanical question.
• 4:30
  SPEAKER_00: How does modification A impact modification B?
• 4:34
  SPEAKER_01: Give me a hypothetical here.
• 4:36
  SPEAKER_00: Okay.
• 4:36
  SPEAKER_00: Say your roadmap includes updating a diagnostic algorithm to process data faster.
• 4:41
  SPEAKER_01: Okay.
• 4:41
  SPEAKER_01: Modification A.
• 4:42
  SPEAKER_00: Right.
• 4:43
  SPEAKER_00: And modification B is an update to the user interface dashboard for the nurses.
• 4:47
  SPEAKER_01: Got it.
• 4:48
  SPEAKER_00: The impact assessment requires you to prove that the faster algorithm won't,
• 4:52
  SPEAKER_00: you know,
• 4:52
  SPEAKER_00: spit out raw data, that the new dashboard isn't calibrated to the display properly.
• 4:57
  SPEAKER_00: Because if it does, that could lead a nurse to misinterpret a vital sign.
• 5:01
  SPEAKER_00: You have to prove the collision of these updates won't compromise patient safety.
• 5:05
  SPEAKER_01: So it isn't just about testing the new spark plugs in a vacuum.
• 5:08
  SPEAKER_01: It is proving the new spark plugs don't completely overwhelm the engine computer
• 5:12
  SPEAKER_01: and cause a catastrophic failure.
• 5:14
  SPEAKER_00: That captures the mechanical reality perfectly.
• 5:17
  SPEAKER_00: The FDA wants to see that you have architected a closed-loop system
• 5:21
  SPEAKER_00: for evaluating compounding risks.
• 5:24
  SPEAKER_01: What I find fascinating in our source material, though, is the scope of this.
• 5:28
  SPEAKER_01: Because the headlines are always dominated by AI and machine learning algorithms.
• 5:32
  SPEAKER_01: Always.
• 5:33
  SPEAKER_01: But the applications for PCCPs stretch way beyond just AI.
• 5:37
  SPEAKER_00: They do.
• 5:38
  SPEAKER_00: The creativity we're seeing from manufacturers right now is actually remarkable.
• 5:43
  SPEAKER_00: The sources detail this great case with natural cycles.
• 5:46
  SPEAKER_01: The fertility monitoring company?
• 5:48
  SPEAKER_00: Yeah.
• 5:48
  SPEAKER_00: They make a software as a medical device for fertility monitoring.
• 5:52
  SPEAKER_00: And originally,
• 5:53
  SPEAKER_00: their system required users to take their temperature with
• 5:55
  SPEAKER_00: a standard oral thermometer.
• 5:57
  SPEAKER_01: Right.
• 5:58
  SPEAKER_00: But over time,
• 5:59
  SPEAKER_00: they obviously wanted to integrate with wearable devices like the Aura Smart Ring
• 6:04
  SPEAKER_00: and eventually the Apple Watch.
• 6:06
  SPEAKER_01: But adding a new hardware input,
• 6:07
  SPEAKER_01: especially a consumer wearable,
• 6:09
  SPEAKER_01: I mean, traditionally that would trigger a massive regulatory bottleneck.
• 6:13
  SPEAKER_00: In the past, yes, a whole new 510K.
• 6:16
  SPEAKER_00: But they submitted a clearance, it's K241006, that was virtually entirely focused on a PCCP.
• 6:25
  SPEAKER_01: Oh, wow.
• 6:25
  SPEAKER_00: Yeah.
• 6:26
  SPEAKER_00: They weren't adding new software features per se.
• 6:29
  SPEAKER_00: They were seeking authorization to add more wearable temperature measuring devices
• 6:33
  SPEAKER_00: in the future without needing additional 510K submissions.
• 6:36
  SPEAKER_01: Because they already knew how to test it.
• 6:38
  SPEAKER_00: Exactly.
• 6:39
  SPEAKER_00: They had already conducted multiple clinical studies,
• 6:42
  SPEAKER_00: so they knew exactly how to establish the precise key performance indicators needed
• 6:46
  SPEAKER_00: to validate any new wearable.
• 6:48
  SPEAKER_00: They basically pre-cleared the methodology for hardware expansion.
• 6:52
  SPEAKER_01: Which is just a masterstroke in capital allocation.
• 6:55
  SPEAKER_01: I mean, they didn't just clear the Apple Watch.
• 6:58
  SPEAKER_01: They cleared the structural pathway for whatever watch comes out next year.
• 7:02
  SPEAKER_00: Exactly.
• 7:03
  SPEAKER_01: I also noted in the FDA guidance documents they explicitly outline scenarios
• 7:08
  SPEAKER_01: for site-specific tuning, like with ventilator settings.
• 7:11
  SPEAKER_00: That is another incredibly powerful application.
• 7:14
  SPEAKER_00: The guidance describes a manufacturer pre-clearing the ability to retrain
• 7:18
  SPEAKER_00: an AI model to optimize ventilator recommendations for a specific subset of patients
• 7:24
  SPEAKER_00: at a specific hospital.
• 7:26
  SPEAKER_01: So localized AI.
• 7:28
  SPEAKER_00: Right.
• 7:28
  SPEAKER_00: They can adapt to local clinical nuances,
• 7:30
  SPEAKER_00: say a hospital at a high altitude
• 7:32
  SPEAKER_00: that has totally different baseline respiratory data.
• 7:36
  SPEAKER_00: They can do that instantly, as long as they follow that predefined modification protocol.
• 7:41
  SPEAKER_01: Okay, wait.
• 7:41
  SPEAKER_01: I have to step in here and challenge this premise.
• 7:44
  SPEAKER_00: Sure.
• 7:44
  SPEAKER_01: Because if I'm sitting in the C-suite, this all sounds almost too good to be true.
• 7:49
  SPEAKER_00: It's not a magic wand.
• 7:50
  SPEAKER_01: Right.
• 7:51
  SPEAKER_01: If I have to lock in my exact testing protocol,
• 7:53
  SPEAKER_01: my statistical thresholds,
• 7:55
  SPEAKER_01: and my exact data gathering methods years in advance,
• 7:59
  SPEAKER_01: doesn't that completely destroy my R&D agility?
• 8:02
  SPEAKER_00: That's a fair question.
• 8:03
  SPEAKER_01: Like what if the market shifts in 18 months and my engineers need to pivot
• 8:06
  SPEAKER_01: the software roadmap to solve a totally different clinical problem?
• 8:10
  SPEAKER_00: And that is the reality check that every executive team needs to hear.
• 8:14
  SPEAKER_00: A PCCP is not a blank check.
• 8:16
  SPEAKER_01: Okay.
• 8:17
  SPEAKER_00: It absolutely does not allow you to pivot your core business model on the fly.
• 8:21
  SPEAKER_00: The flexibility exists entirely within predefined boundaries,
• 8:25
  SPEAKER_00: specifically around your intended use.
• 8:27
  SPEAKER_01: Intended use.
• 8:29
  SPEAKER_00: Yes.
• 8:30
  SPEAKER_00: So if your device is cleared for adult cardiovascular monitoring,
• 8:34
  SPEAKER_00: you cannot use a pre-cleared update
• 8:36
  SPEAKER_00: to suddenly start diagnosing pediatric respiratory illnesses.
• 8:39
  SPEAKER_01: Right.
• 8:40
  SPEAKER_01: Obviously.
• 8:41
  SPEAKER_00: You also cannot break the substantial equivalence to your predicate device.
• 8:45
  SPEAKER_01: Meaning what exactly?
• 8:46
  SPEAKER_00: Meaning.
• 8:48
  SPEAKER_00: Your underlying technology and safety profile must remain fundamentally tied
• 8:52
  SPEAKER_00: to what was originally cleared.
• 8:53
  SPEAKER_01: So if the market demands a completely new diagnostic feature, the PCCP won't save you.
• 8:58
  SPEAKER_01: You are going back to the traditional 510k queue.
• 9:01
  SPEAKER_00: You are.
• 9:02
  SPEAKER_00: You're starting over.
• 9:04
  SPEAKER_00: And the analytics guidance we reviewed highlights a critical defensive strategy here.
• 9:08
  SPEAKER_00: Modularize your risks.
• 9:10
  SPEAKER_01: Modularize them?
• 9:11
  SPEAKER_00: How so?
• 9:13
  SPEAKER_00: They advise keeping your proposed changes structurally separated in your documentation.
• 9:17
  SPEAKER_00: So if your PCCP covers five different modifications,
• 9:21
  SPEAKER_00: say tweaking an algorithm,
• 9:22
  SPEAKER_00: adding a new language to the UI,
• 9:25
  SPEAKER_00: and integrating a new hardware sensor, keep the protocols for those independent.
• 9:30
  SPEAKER_00: The FDA reviews this line by line.
• 9:33
  SPEAKER_00: If they feel your hardware integration protocol is too risky,
• 9:36
  SPEAKER_00: you want them to line item veto just that one modification.
• 9:39
  SPEAKER_01: Because if your roadmap is all tangled together in one massive protocol...
• 9:43
  SPEAKER_00: A single FDA objection could tank the entire submission.
• 9:46
  SPEAKER_01: Ah.
• 9:47
  SPEAKER_01: So you have to contain the blast radius of a regulatory rejection.
• 9:50
  SPEAKER_00: Precisely.
• 9:51
  SPEAKER_01: That makes a lot of sense.
• 9:53
  SPEAKER_01: But knowing those boundaries is only half the battle.
• 9:55
  SPEAKER_01: Let's look at how companies actually balance this rigidity with speed in the real world.
• 10:01
  SPEAKER_01: Because Tempest AI provides a total masterclass in this with their ECG MR clearance.
• 10:06
  SPEAKER_00: Yes, Clearitz K254297.
• 10:09
  SPEAKER_00: Tempest AI is an incredible exemplar here.
• 10:12
  SPEAKER_00: Their software analyzes resting,
• 10:15
  SPEAKER_00: non-ambulatory,
• 10:16
  SPEAKER_00: 12-lead ECGs to detect signs of moderate or severe mitral regurgitation.
• 10:21
  SPEAKER_01: And the initial model was trained on, what, over 720,000 ECGs?
• 10:26
  SPEAKER_00: Yeah, a massive data set.
• 10:29
  SPEAKER_00: But the breakthrough is what their PCCP authorized.
• 10:32
  SPEAKER_00: They pre-cleared the ability to retrain their locked machine learning model using new data.
• 10:37
  SPEAKER_00: And they pre-cleared adding support for new ECG machine models.
• 10:41
  SPEAKER_01: And looking at the mechanics of this, the boundaries they set are surgical.
• 10:45
  SPEAKER_01: I mean, it is strictly for patients 65 and older who are non-ambulatory and
• 10:51
  SPEAKER_01: who have not had prior mitral valve interventions.
• 10:54
  SPEAKER_00: Very tight clinical scope.
• 10:56
  SPEAKER_01: Extremely.
• 10:57
  SPEAKER_01: But the statistical floor they negotiated with the FDA is what really caught my eye.
• 11:01
  SPEAKER_00: Yeah, this is where we get into the heavy math of the modification protocol.
• 11:05
  SPEAKER_00: During their original clinical validation study,
• 11:07
  SPEAKER_00: their point estimate, the most likely statistical outcome for sensitivity, was 75%.
• 11:14
  SPEAKER_00: And for specificity, it was 70%.
• 11:16
  SPEAKER_01: Okay.
• 11:17
  SPEAKER_00: But for their PCCP, they had to guarantee safety for future unseen updates.
• 11:23
  SPEAKER_00: So they agreed to a predefined performance floor based on confidence intervals.
• 11:28
  SPEAKER_01: Now, for those of us not living in statistics every day,
• 11:31
  SPEAKER_01: a confidence interval basically measures the reliability of an estimate, right?
• 11:35
  SPEAKER_00: Exactly.
• 11:37
  SPEAKER_00: They stated that for any future model updates,
• 11:40
  SPEAKER_00: the lower bound of the 95% confidence interval for both sensitivity
• 11:44
  SPEAKER_00: and specificity must remain above 65%.
• 11:46
  SPEAKER_01: So in plain English.
• 11:49
  SPEAKER_00: In plain terms,
• 11:50
  SPEAKER_00: they had to guarantee the FDA that even in
• 11:52
  SPEAKER_00: an absolute worst-case statistical scenario,
• 11:56
  SPEAKER_00: the algorithm's accuracy would never drop below 65%.
• 11:59
  SPEAKER_01: And as long as their newly retrained algorithm hits that 65% floor
• 12:04
  SPEAKER_01: on independent validation data,
• 12:06
  SPEAKER_00: they can push the update live to hospitals.
• 12:08
  SPEAKER_01: Wow.
• 12:09
  SPEAKER_01: And GE Healthcare executed a similar strategy with their EchoPak software, right?
• 12:14
  SPEAKER_01: Specifically, Clearance K261623?
• 12:19
  SPEAKER_00: Yes, GE.
• 12:20
  SPEAKER_00: This is a workstation software used for reviewing ultrasound images,
• 12:23
  SPEAKER_00: and they included a PCCP exclusively for their AI cardiac autodoppler feature.
• 12:29
  SPEAKER_01: And they took a slightly different approach than Tempest.
• 12:32
  SPEAKER_00: They did.
• 12:33
  SPEAKER_00: DE went incredibly broad with the technical scope while keeping the clinical scope narrow.
• 12:39
  SPEAKER_00: They pre-cleared updates to multiple core AI algorithms.
• 12:42
  SPEAKER_01: Like the Doppler spectrum cycle detection model?
• 12:45
  SPEAKER_00: Right, a tissue Doppler network and a Doppler trace network.
• 12:49
  SPEAKER_00: They also pre-cleared software optimization.
• 12:51
  SPEAKER_00: But crucially,
• 12:52
  SPEAKER_00: they pre-cleared support for additional transthoracic
• 12:55
  SPEAKER_00: and transesophageal ultrasound probes.
• 12:58
  SPEAKER_01: And just to clarify,
• 13:00
  SPEAKER_01: transthoracic probes,
• 13:01
  SPEAKER_01: or TTEs, are the standard ultrasound wands used on the outside of the chest wall.
• 13:06
  SPEAKER_00: Correct.
• 13:07
  SPEAKER_01: So they are updating software that interacts with entirely new physical hardware
• 13:12
  SPEAKER_01: without breaking the rules.
• 13:13
  SPEAKER_01: How did they manage that?
• 13:15
  SPEAKER_00: By bounding the clinical output.
• 13:17
  SPEAKER_00: The GE protocol explicitly states
• 13:19
  SPEAKER_00: that these modifications cannot introduce new measurement types,
• 13:23
  SPEAKER_00: they cannot add new diagnostic claims,
• 13:25
  SPEAKER_00: and they cannot change the fundamental clinical interpretation of the results.
• 13:29
  SPEAKER_01: So the software simply gets better, faster,
• 13:31
  SPEAKER_01: and more compatible with more ones,
• 13:34
  SPEAKER_01: but it doesn't suddenly claim to diagnose a new disease.
• 13:37
  SPEAKER_00: Exactly.
• 13:38
  SPEAKER_01: I really want to focus on the business impact of what Tempest and GE have achieved here.
• 13:42
  SPEAKER_00: It's huge.
• 13:43
  SPEAKER_01: To use an industrial analogy,
• 13:45
  SPEAKER_01: they didn't just get the FDA to approve the individual software updates coming off
• 13:49
  SPEAKER_01: the assembly line.
• 13:50
  SPEAKER_01: They got the FDA to approve the factory itself.
• 13:53
  SPEAKER_00: That's a great way to put it.
• 13:54
  SPEAKER_01: The testing protocols, the statistical floors.
• 13:57
  SPEAKER_00: Because the factory is approved,
• 13:58
  SPEAKER_00: whatever it produces is instantly commercializable,
• 14:01
  SPEAKER_00: provided it passes the internal quality checks.
• 14:04
  SPEAKER_01: Right.
• 14:04
  SPEAKER_00: GE and Tempest don't have to pause their sales momentum to wait on the government.
• 14:08
  SPEAKER_01: That is a staggering operational advantage.
• 14:10
  SPEAKER_00: It is, but realizing that advantage requires a massive shift in how
• 14:14
  SPEAKER_00: a company operates internally.
• 14:16
  SPEAKER_01: Yeah, this isn't just paperwork.
• 14:17
  SPEAKER_00: No, this isn't a regulatory trick you can just hand off to your compliance team.
• 14:21
  SPEAKER_00: Building and running that factory requires unprecedented synchronization
• 14:26
  SPEAKER_00: between R&D, regulatory, and finance.
• 14:29
  SPEAKER_01: Which brings us to the operational friction.
• 14:31
  SPEAKER_01: The Zalliance analysis in our sources dives into something called the evidence budget.
• 14:35
  SPEAKER_00: The evidence budget, yeah.
• 14:36
  SPEAKER_01: Which really gets to the heart of the tension between,
• 14:39
  SPEAKER_01: say, a chief technology officer and a chief financial officer.
• 14:42
  SPEAKER_00: The evidence budget is the ground truth reality of PCCPs.
• 14:47
  SPEAKER_00: Let's look at the Tempest example again.
• 14:50
  SPEAKER_00: Say your R&D team wants to use the pre-cleared plan to make your software compatible
• 14:54
  SPEAKER_00: with a brand new brand of ECG machine.
• 14:57
  SPEAKER_01: Okay.
• 14:58
  SPEAKER_00: From an engineering standpoint,
• 14:59
  SPEAKER_00: making sure that the new file format integrates into your software is easy.
• 15:03
  SPEAKER_00: It's just coding.
• 15:04
  SPEAKER_00: Sure.
• 15:04
  SPEAKER_00: But proving to the FDA that your AI still analyzes that new,
• 15:09
  SPEAKER_00: slightly different file correctly, that requires independent clinical data.
• 15:14
  SPEAKER_01: So think about this like building a high-speed railway.
• 15:17
  SPEAKER_00: Okay.
• 15:17
  SPEAKER_01: The R&D team coding the new software feature is just building the train.
• 15:21
  SPEAKER_01: But the PCCP dictates that before that train is legally allowed to carry passengers,
• 15:27
  SPEAKER_01: it has to run on a very specific set of tracks.
• 15:29
  SPEAKER_00: Ah, the independent clinical validation data.
• 15:31
  SPEAKER_01: Exactly.
• 15:32
  SPEAKER_01: And getting a hospital to gather new patient data,
• 15:35
  SPEAKER_01: anonymize it,
• 15:36
  SPEAKER_01: blind it from the engineering team,
• 15:37
  SPEAKER_01: and run it through a statistical analysis,
• 15:40
  SPEAKER_01: that is incredibly expensive and time-consuming.
• 15:43
  SPEAKER_00: And that right there is the finance bottleneck.
• 15:46
  SPEAKER_00: The CFO has to pay for the tracks, not just the train.
• 15:50
  SPEAKER_01: Right.
• 15:50
  SPEAKER_00: Access to more training data for the AI does not automatically establish access to
• 15:55
  SPEAKER_00: the independent evidence needed to validate the update.
• 15:58
  SPEAKER_01: That's a crucial distinction.
• 16:00
  SPEAKER_00: If R&D builds a brilliant new feature,
• 16:02
  SPEAKER_00: but finance didn't budget the,
• 16:04
  SPEAKER_00: I don't know,
• 16:04
  SPEAKER_00: half a million dollars required to run the independent clinical study
• 16:07
  SPEAKER_00: to prove it meets the PCCP's 65% floor.
• 16:12
  SPEAKER_01: Then what?
• 16:13
  SPEAKER_00: That software update is completely useless.
• 16:15
  SPEAKER_00: It cannot be deployed.
• 16:16
  SPEAKER_01: So the commercial sequence flips upside down.
• 16:18
  SPEAKER_00: Totally.
• 16:19
  SPEAKER_01: Under the old 510K model,
• 16:21
  SPEAKER_01: a company might build a feature,
• 16:22
  SPEAKER_01: test it internally,
• 16:23
  SPEAKER_01: see if it works,
• 16:25
  SPEAKER_01: and then figure out how to scrape together the budget for a regulatory submission.
• 16:29
  SPEAKER_00: But with a PCCP?
• 16:30
  SPEAKER_01: You are pre-funding your innovation runway.
• 16:33
  SPEAKER_01: You have to define the testing protocol and the statistical thresholds years in advance.
• 16:38
  SPEAKER_01: The implication is staggering.
• 16:40
  SPEAKER_01: You can only pre-clear what you can actually afford to validate.
• 16:44
  SPEAKER_00: Exactly.
• 16:45
  SPEAKER_00: Treating clinical validation as an afterthought under
• 16:48
  SPEAKER_00: a PCCP leaves your entire multi-year product roadmap dependent on evidence
• 16:52
  SPEAKER_00: that nobody bothered to procure.
• 16:53
  SPEAKER_00: Wow.
• 16:54
  SPEAKER_00: It forces a level of financial discipline that many software as
• 16:57
  SPEAKER_00: a medical device companies,
• 16:58
  SPEAKER_00: especially those accustomed to a move fast and break things Silicon Valley mindset,
• 17:03
  SPEAKER_00: simply are not used to.
• 17:04
  SPEAKER_01: Because they've signed a contract.
• 17:06
  SPEAKER_00: You have signed a binding legal agreement with the federal government
• 17:09
  SPEAKER_00: about exactly how you will measure the things you build.
• 17:13
  SPEAKER_01: We have covered so much ground today.
• 17:14
  SPEAKER_01: We've mapped the strategic shift, the technical boundaries, the severe operational realities.
• 17:20
  SPEAKER_01: Now we need to distill this into an action plan.
• 17:24
  SPEAKER_00: Let's do it.
• 17:25
  SPEAKER_01: If you are an executive stepping into a boardroom for
• 17:27
  SPEAKER_01: a strategic planning session tomorrow morning,
• 17:30
  SPEAKER_01: here is how you operationalize this intelligence.
• 17:32
  SPEAKER_00: The first directive is to modularize the strategy.
• 17:36
  SPEAKER_00: When your team is developing a PCCP, Right.
• 17:39
  SPEAKER_00: Right.
• 17:40
  SPEAKER_00: Right.
• 18:01
  SPEAKER_00: Protect your core product clearance by giving the FDA the ability to line item veto
• 18:06
  SPEAKER_00: a single risky feature without bringing down your entire submission.
• 18:28
  SPEAKER_01: If your protocol demands an independent dataset of 1,000 diverse patients
• 18:32
  SPEAKER_01: to validate a model retrain,
• 18:34
  SPEAKER_01: the procurement and blinding of that data must be budgeted before R&D writes
• 18:38
  SPEAKER_01: a single line of code.
• 18:40
  SPEAKER_01: A PCCP without an evidence budget is a hollow promise.
• 18:44
  SPEAKER_00: The third directive is to respect the intended use boundary.
• 18:48
  SPEAKER_00: PCCPs are precision tools designed for iterative performance improvements,
• 18:52
  SPEAKER_00: statistical algorithm tuning, and defined hardware expansions.
• 18:56
  SPEAKER_01: They are not pivot mechanisms.
• 18:57
  SPEAKER_00: No.
• 18:58
  SPEAKER_00: The moment an executive team tries to use a pre-authorized update to sneak in
• 19:03
  SPEAKER_00: a new clinical diagnostic claim,
• 19:05
  SPEAKER_00: they will trigger a regulatory crisis and require a brand new 510K.
• 19:10
  SPEAKER_00: The clinical purpose of the device must remain anchored to its original clearance.
• 19:13
  SPEAKER_01: And the final directive, the ultimate takeaway for the C-suite, weaponize cycle time as a moat.
• 19:20
  SPEAKER_01: We have to stop viewing regulatory strategy as a series of compliance hurdles
• 19:24
  SPEAKER_01: and start treating it as a tool for capital efficiency.
• 19:27
  SPEAKER_00: I couldn't agree more.
• 19:28
  SPEAKER_01: Look at the mechanics of the market.
• 19:30
  SPEAKER_01: While your competitor finishes an algorithm update,
• 19:33
  SPEAKER_01: they have to halt their commercial momentum,
• 19:35
  SPEAKER_01: file a 510K, and sit on their hands for 120 days waiting on a government reviewer.
• 19:40
  SPEAKER_00: Just burning cash.
• 19:41
  SPEAKER_01: Exactly.
• 19:42
  SPEAKER_01: But if you have a PCCP,
• 19:45
  SPEAKER_01: your pre-authorized updates hit the market
• 19:46
  SPEAKER_01: the instant they pass your internal factory tests.
• 19:49
  SPEAKER_01: You can respond to physician feedback,
• 19:51
  SPEAKER_01: integrate new wearables,
• 19:53
  SPEAKER_01: and deploy site-specific AI tuning months before your competition even gets
• 19:56
  SPEAKER_01: an email back from the FDA.
• 19:58
  SPEAKER_00: That cycle time compression is a devastating advantage.
• 20:01
  SPEAKER_01: It really is.
• 20:02
  SPEAKER_00: It allows a medical technology company to finally operate at the speed
• 20:05
  SPEAKER_00: of modern software development while maintaining total documented adherence
• 20:09
  SPEAKER_00: to patient safety standards.
• 20:12
  SPEAKER_00: Understanding the granular details of the modification protocol and
• 20:15
  SPEAKER_00: the impact assessment isn't just regulatory trivia.
• 20:18
  SPEAKER_00: It is the blueprint for a company's commercial velocity.
• 20:21
  SPEAKER_01: It is the blueprint for the factory.
• 20:23
  SPEAKER_01: As you head back to your teams, I want you to sit with this final thought.
• 20:27
  SPEAKER_01: Think about the landscape we've just unpacked.
• 20:29
  SPEAKER_01: Think about market leaders like Tempest AI and GE Healthcare operating
• 20:33
  SPEAKER_01: with these pre-cleared, high-speed iteration engines.
• 20:36
  SPEAKER_01: If your competitor's diagnostic AI can be fine-tuned to
• 20:39
  SPEAKER_01: a specific hospital's unique patient population in real time,
• 20:43
  SPEAKER_01: responding to clinical data in days,
• 20:46
  SPEAKER_01: while your device is stuck waiting four months in an FDA queue just to push
• 20:50
  SPEAKER_01: a minor software patch, how long until your market share completely evaporates?