Best of LinkedIn: Health Tech CW 04/ 05

00:00:00: This episode is provided by Thomas Allgeier and Frennis based on the most relevant LinkedIn posts on health tech in CW four and five.

00:00:08: Frennis equips product and strategy teams with market and competitive intelligence to navigate the health tech landscape.

00:00:14: It is great to be back and you know looking at the intelligence stack from calendar weeks four and five there's a definite shift in the air.

00:00:22: It feels

00:00:22: heavier doesn't it?

00:00:23: Less about the shiny object and more about well the heavy lifting.

00:00:27: Precisely.

00:00:28: If last year was the year of the demo, where everyone was showing off what was theoretically possible, the start of twenty.

00:00:34: twenty-six is really shaping up to be the era of the execution frontier.

00:00:38: The industry is rolling up its sleeves.

00:00:40: It is.

00:00:41: We're seeing this clear shift from, you know, look what this AI can do to, does this actually work when things get messy in a real hospital?

00:00:48: And that's the mission for this deep dive.

00:00:50: We're stripping away the hype to see what's actually sticking.

00:00:52: We've got what, four main clusters to get through?

00:00:55: Yep.

00:00:55: First, we'll dissect the reality gap in AI diagnostics, specifically why these magic demos are failing in clinical practice.

00:01:02: Then we're scrubbing into the operating room.

00:01:04: Right.

00:01:04: Looking at why robotics is moving from a hardware conversation to a workflow conversation.

00:01:10: We'll also take a hard look at the consumer side wearables, glucose monitoring, and the harsh truth that, you know... data doesn't equal behavior change.

00:01:19: And finally, we have to talk about the energy vampires and strategy.

00:01:23: We'll look at why pilot programs are failing and a massive regulatory blind spot that was frankly exposed this week.

00:01:31: It's a dense stack.

00:01:32: So let's get right into it.

00:01:33: Theme one, AI and care delivery.

00:01:36: And right away, there is this This real tension between marketing promises and clinical reality.

00:01:42: It's the network to product gap.

00:01:45: And nothing showed this better this week than a story shared by Yongcha, the former Apple Health exec.

00:01:50: Oh,

00:01:50: I saw this one.

00:01:51: He highlighted

00:01:51: a really telling experiment and involved a reporter from the Washington Post and Chad GPT Health.

00:01:56: This is the one where the reporter fed ten years of his Apple Watch data into the system, right?

00:02:00: Correct.

00:02:01: Ten years of heart rate, sleep, activity data.

00:02:04: And he asked the AI a simple question.

00:02:06: grade my cardiac health.

00:02:08: The AI came back and gave him F.

00:02:09: An F?

00:02:10: Wow.

00:02:11: That implies imminent failure.

00:02:13: Which

00:02:13: is terrifying.

00:02:14: Especially since his actual human cardiologist had told him he was perfectly fine.

00:02:18: Okay, but here's

00:02:19: the kicker.

00:02:19: Here's the kicker.

00:02:20: The part that really matters for our listeners building these tools.

00:02:24: The reporter asked the AI again.

00:02:27: Same data.

00:02:28: This time.

00:02:28: It gave him a C.

00:02:29: A C?

00:02:30: He asked

00:02:30: a third time.

00:02:31: It gave him a B.

00:02:31: Hold on.

00:02:32: Same inputs.

00:02:33: Three wildly different outputs ranging from you're failing to you're doing great.

00:02:38: That's not a diagnostic tool.

00:02:39: That's that's a random number generator.

00:02:42: That is exactly Young Chae's point.

00:02:44: He argues that general purpose large language models lack what he calls native priors for trend estimation.

00:02:50: Unpack that for us native priors.

00:02:52: Think of it as context or a baseline.

00:02:55: A human doctor has a prior understanding of what's a normal fluctuation versus a dangerous anomaly for you based on your history.

00:03:02: Okay.

00:03:02: The LLM is just pattern matching against a huge generic training set.

00:03:06: Doesn't know you, it's guessing.

00:03:08: So, Charles basically saying we have a reliability crisis.

00:03:12: It's easy to make a demo that gets the right answer once for an investor deck.

00:03:16: But it's incredibly hard to build a product that's reliable enough to trust with the diagnosis.

00:03:21: Reliability is the currency.

00:03:23: And if an AI tells a healthy patient they're dying, that's not just an error, it's actual harm.

00:03:29: It is.

00:03:29: It creates anxiety.

00:03:30: It wastes clinical resources and it erodes trust.

00:03:34: But we did see the flip side of this in the feed.

00:03:36: Taha cast out shared a scenario where AI is working.

00:03:39: This is the stroke case.

00:03:41: Yes.

00:03:41: A patient arrives in the ER, right side paralyzed.

00:03:44: The initial CT scan looks normal to the human eye.

00:03:47: But minutes later, an AI model reviewing that same scan flags a subtle vessel irregularity.

00:03:53: And

00:03:53: that flag got the team to reevaluate and treat.

00:03:56: Exactly.

00:03:57: But notice the difference.

00:03:58: In the chat GPT example, the AI was the judge giving a grade.

00:04:02: Right, on its own.

00:04:03: In the stroke case, the AI was a safety net.

00:04:07: The AI didn't replace the radiologist.

00:04:10: It just caught the subtle signal that a stressed human might miss.

00:04:13: So the takeaway is AI works best as an alarm system, not the judge and jury.

00:04:19: But for that alarm system to work, the data has to be clean.

00:04:23: And Matthias Goyen raised a point that I think flies into the radar for a lot of tech teams.

00:04:27: He talked about sloppy language.

00:04:29: Right, the ethics of it.

00:04:30: Yes, it was a fascinating take.

00:04:31: We usually talk about data hygiene as a technical problem.

00:04:35: Goyen argues it's a moral one.

00:04:36: How so?

00:04:37: Well, AI models are trained on medical reports written by humans.

00:04:42: If doctors use ambiguous language or hedging or just sloppy descriptions in their notes, the AI learns to normalize that ambiguity.

00:04:49: It mirrors our bad habits back at

00:04:52: us.

00:04:52: And then it amplifies them.

00:04:54: Goyne's argument is that if we know these systems will be treating patients in the future, writing precise, clear medical notes today is an ethical obligation.

00:05:02: Wow.

00:05:03: That's a heavy realization.

00:05:04: It means good documentation is no longer just about billing or liability.

00:05:07: It's about the future intelligence of the entire system.

00:05:10: Exactly.

00:05:11: And speaking of precision, I want to touch on one specific win that came up.

00:05:14: Chris Scott or DeVito posted about page predict, which was just acquired by Tempus.

00:05:18: This felt like a very tangible solution to a very real problem.

00:05:22: It addresses the tissue economy in oncology.

00:05:25: This is something people outside of pathology rarely think about, but it's a huge bottleneck.

00:05:30: Explain that.

00:05:31: Sure.

00:05:32: When you do a biopsy, you have a tiny amount of tissue.

00:05:34: You have to slice it up to test for different biomarkers.

00:05:37: Often by the time you get to the advanced tests like next generation sequencing and GS, you've literally run out of tissue.

00:05:44: So the patient misses out on crucial genetic testing because the sample was used up on more basic tests.

00:05:49: That feels like a tragedy of logistics.

00:05:51: It is.

00:05:52: PagePredict uses standard H&E images.

00:05:55: The standard purple and pink slides you see in every lab to identify biomarkers digitally before you even cut the tissue.

00:06:02: So it's a screening tool to save the physical sample.

00:06:05: It helps clinicians decide the best sequence of tests to maximize the chance of an actionable result.

00:06:10: That is what execution looks like.

00:06:12: It's not a chatbot having a conversation.

00:06:15: It's a backend tool ensuring a cancer patient gets the right diagnosis because you didn't waste their biopsy.

00:06:20: That is a perfect transition to our second theme.

00:06:22: Let's move from the lab to the theater.

00:06:25: Digital surgery and robotics.

00:06:27: The narrative here is shifting fast.

00:06:29: For years the conversation was Look at this cool robot.

00:06:32: And

00:06:33: now.

00:06:33: Now the conversation is, how does this fit into my P&L and my workflow?

00:06:37: Right.

00:06:38: The buzzword in the posts from Rajit Kamal and James Boyle at Medtronic wasn't autonomy.

00:06:43: It was choice.

00:06:44: They were talking about the Hugo RIS system.

00:06:46: And notice the pitch.

00:06:48: They aren't saying this robot replaces everything.

00:06:51: They're pitching modularity.

00:06:53: Because hospitals are crowded and budgets are tight.

00:06:55: Exactly.

00:06:56: You can't fit a massive monolithic system in every OR.

00:07:00: The Hugo system focuses on flexibility letting surgeons use specific tools, like the Ligashire Maryland jaw or the Rubina visioning system, only when they need them for a procedure.

00:07:10: It sounds like they're admitting the one size fits all era of robotics is over.

00:07:14: It's about being part of an ecosystem, not dominating the room.

00:07:17: But does the modular approach actually deliver?

00:07:21: God will.

00:07:21: Dave shared some numbers from the expanded URL trial that caught my eye.

00:07:24: Those

00:07:25: numbers were incredibly robust.

00:07:27: They were.

00:07:30: They set a target for success at eighty-five percent.

00:07:32: They

00:07:32: hit ninety-eight point five

00:07:34: percent.

00:07:34: Ninety-eight point five percent.

00:07:36: That loops us right back to our reliability theme.

00:07:39: In the early days of robotics, there was always a fear of technical failure or latency.

00:07:44: Hitting nearly ninety-nine percent efficacy in a clinical trial builds the one thing you can't buy.

00:07:50: Surgeon trust.

00:07:51: Because if a surgeon doesn't trust the machine, it just becomes an expensive coat rack in the corner of the OR.

00:07:56: Exactly.

00:07:57: But even with that ninety-nine percent success rate, Gavin Setzen provided a bit of a reality check in the feed.

00:08:03: The adoption lag.

00:08:04: Right.

00:08:04: Setzen warned that we shouldn't expect robots in every local hospital next week.

00:08:08: Broad adoption is measured in years, not months.

00:08:11: Capital cycles, training.

00:08:13: It all takes time.

00:08:13: He emphasized we're in the age of augmentation, not autonomy.

00:08:17: And that connects to the Surgeon of the Future concept that Christian Masi mentioned.

00:08:20: Yeah.

00:08:20: The robot isn't doing the surgery for you while you drink a coffee.

00:08:23: Why from

00:08:23: it?

00:08:24: The surgeon is still accountable at the head of the table.

00:08:26: The robot gives you supervision and super-steady hands, but it doesn't make the decisions.

00:08:32: We are a long, long way from an autonomous droid performing a bypass.

00:08:38: Agreed.

00:08:39: Let's move out of the OR.

00:08:41: and look at the messy world of the consumer.

00:08:44: Theme three, wearables, remote monitoring, and prevention.

00:08:48: This section had some of the most human insights of the week.

00:08:53: There's a growing backlash against the idea that more data equals better health.

00:08:57: Bianca Carbone had the core of the week on this.

00:09:00: She said, AI can give you all the data in the world, but, and I love this, AI can't pick the dumbbells up for you.

00:09:06: It's a sharp critique of the whole quantified self movement.

00:09:09: We confuse information with support.

00:09:11: Right.

00:09:11: Our bonus point is that people don't fail to get healthy because they lack data.

00:09:14: They know they should sleep more.

00:09:16: They fail because they lack accountability, community, and, you know, the actual mechanisms for behavior change.

00:09:21: She basically said, don't spend thousands on health tech if you haven't mastered the basics.

00:09:25: It reminds me of what Mohita Hujo was observing out in Dubai.

00:09:28: The biohacking trend.

00:09:30: He described clinics flooded with people asking for peptides, cryotherapy, all these advanced supplements.

00:09:36: But these are the same people who are sleeping four hours a night and are totally stressed out.

00:09:40: Exactly.

00:09:41: They are trying to tech their way out of a bad lifestyle.

00:09:45: Ahuja's insight is simple but devastating.

00:09:49: Performance doesn't compound.

00:09:50: Foundations do.

00:09:52: You can't use tech to bypass the need for sleep and nutrition.

00:09:55: It's like putting rocket fuel in a car with a flat tire.

00:09:58: Precisely.

00:09:59: That said, for people who are managing a chronic condition, there is some hardware on the horizon that looks promising.

00:10:05: Wonsang called it the holy grail.

00:10:07: non-invasive glucose monitoring.

00:10:10: This has been the next big thing.

00:10:12: for what, a decade?

00:10:13: Are we actually getting closer?

00:10:15: Sang notes the race is getting very real for twenty twenty six.

00:10:18: The impact would be massive.

00:10:20: But she also points out the physics problem.

00:10:22: It's incredibly difficult to get an accurate reading through the skin using light or radio waves.

00:10:26: Too many variables.

00:10:27: Hydration, skin tone,

00:10:29: skin temperature.

00:10:30: All of it.

00:10:31: It goes back to the young chaw point about reliability.

00:10:34: If your watch says your blood sugar is low, but it's actually high because you're dehydrated, that's dangerous.

00:10:40: So the first company to crack that reliability barrier wins.

00:10:44: Wins the whole market.

00:10:45: But there is another barrier besides physics.

00:10:48: Cost.

00:10:49: Simon Neve highlighted this with a post about Leap Health.

00:10:52: Yeah, I saw that stat.

00:10:53: Fifty-four percent of people cite cost as the number one barrier to sleep tracking.

00:10:59: The big, brand smart rings are expensive.

00:11:01: Prohibitive for the average person.

00:11:03: So Leap Health launched a smart ring specifically focused on affordability.

00:11:07: It's a feature-focused device, doesn't do everything, but it tracks sleep at a price point normal people can actually afford.

00:11:13: It's democratizing the data, because if only the wealthy can afford to track their sleep, we aren't solving public health problems.

00:11:19: We're just optimizing the elites.

00:11:21: And speaking of democratization, we have to talk about the story from Atul Gupta.

00:11:25: This was my favorite example of execution this week because of where it happened.

00:11:29: The Wood Green Community Diagnostic Center.

00:11:31: Yes.

00:11:32: Imagine walking into a shopping mall in north London.

00:11:36: Between the stores and the food court, there is a Phillips blue seal MRI scanner.

00:11:41: a helium-free MRI in a mall.

00:11:43: That's what lets them do it, right?

00:11:44: It eliminates the need for the massive venting

00:11:47: pipes.

00:11:47: Exactly.

00:11:48: But look at the demographic data Gupta shared.

00:11:51: Seventy-five percent of the patients using that scanner are from disadvantaged households.

00:11:55: This is bringing the highest tech AI smart speed imaging directly to the people who usually face the longest wait times.

00:12:03: That is massive.

00:12:04: It challenges the assumption that high tech always means exclusive.

00:12:07: It proves that execution isn't just about the machine, it's about the logistics.

00:12:12: Putting the machine where the people are, that is true innovation.

00:12:15: Okay, so we've covered the AI reliability gap, the robotic workflow, the consumer reality check, but none of this works if the companies are dysfunctional.

00:12:23: Let's talk about our final theme, strategy, leadership, and these energy vampires.

00:12:28: This is a crucial warning for any health tech leader listening.

00:12:31: Meenal Shah introduced this concept of pilot churn as an energy vampire.

00:12:36: It's an evocative term.

00:12:37: What does she mean by it?

00:12:38: We've all seen it.

00:12:39: A hospital runs a pilot program for a new tool.

00:12:42: It runs for six months, shows some decent results, but then...

00:12:46: Nothing.

00:12:47: The budget isn't approved or a priority shift.

00:12:49: Right.

00:12:49: The pilot just dies.

00:12:50: Right.

00:12:51: Then, three months later, management introduces another pilot for a different tool.

00:12:56: And the clinicians just roll their eyes.

00:12:58: They're

00:12:58: burned out.

00:12:59: Exactly.

00:13:00: Shaw argues the cost isn't just the money, it's the erosion of clinician trust.

00:13:05: If you tell a doctor, here's a new tool that will change your life and then it disappears.

00:13:10: They are not going to believe you next time.

00:13:12: You've exhausted their capacity for change.

00:13:14: So when the real innovation finally arrives, the staff is too cynical to adopt it.

00:13:19: That is the vampire.

00:13:20: It sucks the energy out of the organization.

00:13:22: And to prevent that, you need better leadership.

00:13:25: Ian Robertson had a very strong take on who needs to be in charge.

00:13:28: He talked about tandem success coming from having clinicians in leadership, not just as advisors.

00:13:34: He used this brilliant analogy, he said, put a referral letter in front of an engineer.

00:13:38: and they see data transfer, a file moving from point A to point B.

00:13:42: But put that same letter in front of a doctor.

00:13:44: And they see their professional reputation.

00:13:47: That letter represents their judgment, their relationship with the specialist, their care for the patient.

00:13:52: That

00:13:52: nuance is everything.

00:13:53: If you build a tool that just transfers data... but ignores the professional tone of a referral, doctors won't use it.

00:14:00: Precisely.

00:14:00: You need that clinical DNA in the C-suite to understand those unspoken rules.

00:14:05: Otherwise, you're just building efficient products that no one actually wants.

00:14:08: But even with great leadership, you're operating in a minefield of regulation.

00:14:13: And Rudolph Ragnar sounded a massive alarm bell this week regarding SAMD software as a medical device.

00:14:19: Ragnar's analysis was shocking.

00:14:21: He claims SAMD regulation has essentially collapsed due to a total failure of enforcement.

00:14:26: Collapsed

00:14:27: is a strong word.

00:14:28: What's the evidence?

00:14:28: He analyzed ninety six thousand health apps on the app stores.

00:14:32: He found that ninety five percent of them, ninety five percent, had no verifiable CE mark or FDA authorization.

00:14:39: Wait, ninety five percent of health apps are unregulated, so the app store is basically the Wild West.

00:14:44: It's worse.

00:14:45: It's an unregulated pharmacy.

00:14:47: You have apps making diagnostic claims, recommending treatment, scoring risk, all without any regulatory oversight.

00:14:53: Wagner argues the app stores have effectively become distributors of illegal medical devices, and the regulators are silent.

00:15:00: That circles right back to the start of our conversation about trust.

00:15:03: If ninety-five percent of these tools are unverified, how can any patient possibly know what is safe?

00:15:08: They can't.

00:15:09: And that is a ticking time bomb for the whole industry.

00:15:12: However, we shouldn't take this as a sign to just give up on regulation.

00:15:15: No.

00:15:16: Sigurd Burs van Roigen offered a necessary counter perspective on the EU AI Act.

00:15:21: She was defending the new regulations.

00:15:23: In a

00:15:23: way, she noted that a lot of companies use the EU AI Act as an excuse to stop innovating.

00:15:28: They say, oh, the compliance is too hard.

00:15:30: We can't build here.

00:15:31: She argues that's lazy thinking.

00:15:34: Regulation should be viewed as a necessary baseline for safety.

00:15:37: It's not a roadblock.

00:15:38: It's the guardrails.

00:15:39: Exactly.

00:15:40: If we want AI to be standard care, it has to be regulated.

00:15:43: Innovation and safety aren't opposites.

00:15:45: They are partners.

00:15:46: If you can't build your product within safety regulations, you probably shouldn't be in health care.

00:15:51: So let's zoom out.

00:15:53: We've gone from an AI giving a reporter a fake heart condition to robots hitting ninety eight percent reliability to an MRI in a shopping mall in a regulatory crisis.

00:16:03: What is the through line here?

00:16:05: If I connect these docs, the message for CW four and five is clear.

00:16:09: The demo era is over.

00:16:11: It is no longer enough to show something cool.

00:16:14: You have to show something that is reliable that respects the clinicians workflow and that can survive scrutiny.

00:16:20: It feels like the industry is maturing.

00:16:21: We're moving from.

00:16:22: can we build it to?

00:16:24: should we build it and does it actually

00:16:25: work?

00:16:26: And is it sustainable?

00:16:27: That is a much harder question to answer.

00:16:29: It requires patience, clinical leadership, and a willingness to do the boring work of execution rather than just the fun work of invention.

00:16:36: And that leaves us with the provocative question for you, the listener, to chew on this week.

00:16:41: Indeed.

00:16:42: Look at your own organization.

00:16:44: Are you building for the demo, for the flash, the VC deck, the hype?

00:16:48: Or are you building for the product, the reliability, the workflow, the trust?

00:16:53: Are you feeding the energy vampire with endless pilots?

00:16:56: Or are you building the foundations for long-term execution?

00:16:59: That is the question.

00:17:01: The execution frontier is where the winners will be determined this year.

00:17:04: Absolutely.

00:17:05: If you enjoyed this episode, new episodes drop every two weeks.

00:17:07: Also, check out our other editions on ICT and tech insights, defense tech, cloud, digital products and services, artificial intelligence and sustainability and green ICT.

00:17:16: Thank you for listening.

00:17:17: Don't forget to subscribe for the next deep dive into the execution frontier.

00:17:21: See you then!