Best of LinkedIn: Health Tech CW 02/ 03

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

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.

00:00:16: And looking at the feed from what the second third weeks of twenty twenty six, it really feels like the ground is shifting under our feet.

00:00:23: It

00:00:23: feels so different, doesn't it?

00:00:25: For the last couple of years, I feel like we've just been swimming in hype.

00:00:28: You know, every press release promised AI was going to cure everything by next Tuesday.

00:00:32: Right.

00:00:32: But looking at the sources from these last two weeks, the whole vibe has changed.

00:00:36: It feels like the rubber is finally hitting the road.

00:00:39: We're seeing.

00:00:40: tools actually, you know, hit the market, but we're also seeing where they break.

00:00:44: That's the perfect way to frame it.

00:00:46: I think that's the mission of this deep dive.

00:00:48: If last year was all about possibility, twenty twenty six is.

00:00:52: it's shaping up to be the year of execution.

00:00:55: We're not just looking at cool tech demos anymore.

00:00:57: We're seeing this fascinating split between, say, rapid experimentation in the U.S.

00:01:02: and the heavy structured regulation in Europe.

00:01:06: But mostly, yeah, it's a desperate focus on workflow.

00:01:10: Because a cool algorithm means nothing if it breaks the hospital.

00:01:12: Absolutely.

00:01:13: So let's unpack this.

00:01:14: We've got a lot of ground to cover everything from AI triage systems that are actually beating human standards to a privacy nightmare that I don't think most patients even realize is happening.

00:01:25: It's a dense couple of weeks.

00:01:26: Let's dive in.

00:01:27: Okay.

00:01:27: So I want to start with the technology itself.

00:01:29: We're seeing this move from, you know, novelty to clinical reality.

00:01:34: And the biggest battleground right now seems to be triage.

00:01:38: It is.

00:01:38: And to understand why, you have to look at the insights Piotr Orszaczewski shared about Health Direct Australia.

00:01:43: Triage is basically the pressure valve of the entire health care system.

00:01:47: Right.

00:01:47: It decides who needs the ER right now and who can just, you know, take an aspirin and call in the morning.

00:01:52: Exactly.

00:01:53: But Piotr points out something fascinating about how we do it now, the gold standards we use.

00:01:57: The protocols nurses follow on the phone.

00:01:59: They're basically decision trees that trace their logic back to the fourth century BC.

00:02:04: The

00:02:04: fourth century BC.

00:02:06: You're kidding?

00:02:06: I'm not.

00:02:07: It's a rigid flow chart.

00:02:08: If chest pain go here, if fever go there.

00:02:11: Efficient maybe, but I'm guessing a lot of nuance gets lost.

00:02:14: A ton.

00:02:15: And that lack of nuance is incredibly expensive.

00:02:18: Legacy decision trees are just structurally biased toward safety buffers.

00:02:23: If the flowchart is unsure, it defaults to go to the ER.

00:02:26: Better safe than the story, I guess.

00:02:28: Right.

00:02:28: But Piotr argues that clinically validated AI is the natural successor because it doesn't need that massive safety buffer to be safe.

00:02:36: So it's not just faster, it's more precise.

00:02:39: Oh, significantly.

00:02:40: He noted that AI triage results in fifty percent less over triage.

00:02:45: Think about that number for a second.

00:02:46: Wow,

00:02:47: fifty percent.

00:02:47: That is half the number of people being sent to the ER who don't need to be there.

00:02:51: That unclogs waiting rooms for people who are actually having a heart attack and it reduces burnout for the staff.

00:02:57: That's massive.

00:02:58: But how is it doing that?

00:02:59: Is it just like processing the flow chart faster?

00:03:02: No, and that's the key distinction.

00:03:04: It captures a much richer clinical profile.

00:03:06: It's looking at risk factors, travel history.

00:03:09: It's asking dynamic questions in real time.

00:03:11: It isn't a binary yes, no tree.

00:03:13: It's a probabilistic model that understands context.

00:03:16: And speaking of these more complex models, I saw a post from Yossi Matias at Google that really caught my eye.

00:03:21: They just announced Majema, one point five.

00:03:24: And the big headline here is we're not just talking about text.

00:03:27: anymore.

00:03:28: Right.

00:03:28: Multimodal is the buzzword of the week.

00:03:30: Exactly.

00:03:31: This new model handles high dimensional imaging, CTs, MRIs, three D volumes.

00:03:36: Yes, he mentioned they've seen a twenty two percent improvement in reasoning on medical records.

00:03:41: That feels like a huge jump in a short time.

00:03:43: It is a huge jump.

00:03:44: being able to correlate a text note and a patient file with a single pixel on an MRI scan.

00:03:50: That's the holy grail.

00:03:52: But I want to pause here because this is so crucial.

00:03:55: We have to be really careful about how we interpret words like reasoning and detection.

00:04:00: There's a really important reality check posted by Jan Schmidt-Profer.

00:04:04: Oh, I think I saw this.

00:04:05: Was this about that viral image claiming AI detects breast cancer five years early?

00:04:10: That's the one.

00:04:11: It was everywhere.

00:04:12: AI finds cancer five years before it develops.

00:04:14: It sounds like science fiction.

00:04:16: But Jan broke it down and the reality is a lot more grounded.

00:04:20: The AI isn't detecting a tumor that doesn't exist yet.

00:04:23: So what is it doing?

00:04:24: It's performing a statistical risk assessment.

00:04:27: Okay, walk me through the difference.

00:04:28: Why does that distinction matter so much to a patient?

00:04:32: Because language shapes reality.

00:04:34: If you tell a patient we found cancer, that is a diagnosis.

00:04:38: It implies a physical thing we need to treat.

00:04:40: But if you tell a patient you have a high statistical probability of developing cancer, that is risk calculation.

00:04:47: Confusing the two causes unnecessary panic.

00:04:50: You don't want someone prepping for chemo because an algorithm saw a dense tissue.

00:04:54: So it's a tool for the radiologist to manage risk, not a replacement for their diagnostic eye.

00:04:59: Precisely.

00:05:00: And Simon Phillips had a great take on this.

00:05:02: He cited analysis showing AI might reduce radiologist hours by about what, thirty three percent over the next five years?

00:05:09: See, when I hear reduce hours by thirty three percent, I immediately think.

00:05:14: Layoffs

00:05:15: you'd think so but you have to look at the other side of the equation.

00:05:18: Imaging volume is skyrocketing.

00:05:20: We're skating more people more often with higher resolution.

00:05:23: So the AI doesn't take the job.

00:05:25: It just stops the radiologist from drowning.

00:05:27: It's about redesigning the role.

00:05:29: Yes

00:05:29: from image reviewer to clinical decision-maker

00:05:33: which actually brings us perfectly to our next big theme.

00:05:37: We have these incredible engines, but if you drop that engine into a hospital that runs on fax machines and burnout, does it even matter?

00:05:45: That is the defining question of twenty twenty six.

00:05:48: Workflow.

00:05:49: Governance.

00:05:50: infrastructure.

00:05:51: These are the real scaling constraints.

00:05:53: I have to share my favorite quote of the week on this.

00:05:55: This comes from Joel Buccas.

00:05:56: He said, buying AI for a department with broken workflows is like putting a Ferrari engine in a car with flat tires.

00:06:03: It's a brilliant analogy and painful because it's so true.

00:06:06: He says you aren't going to go faster.

00:06:08: You're just going to vibrate harder.

00:06:09: You're just automating chaos.

00:06:11: And that vibrating harder has real financial consequences.

00:06:15: Omar Kaye shared a stat that actually stopped me in my tracks.

00:06:19: He said, eighty seven percent of hospital AI pilots in Europe failed to turn into contracts.

00:06:24: Eighty seven percent.

00:06:26: That is that's wildly high.

00:06:29: That's a catastrophic failure rate.

00:06:30: Why is it happening?

00:06:31: It's not because the AI doesn't work.

00:06:33: The algorithms are usually fine.

00:06:35: It's because of what Omar calls the hidden gates.

00:06:38: See, startups optimize for accuracy.

00:06:40: Right.

00:06:40: Look, our model is ninety nine percent corrupt.

00:06:42: Exactly.

00:06:43: But hospitals, hospitals optimize for risk transfer.

00:06:46: Risk transfer, break that down for me.

00:06:48: It

00:06:48: means, can we defend this in court?

00:06:51: Is the data governance airtight?

00:06:52: Who is liable if it pushes the wrong button?

00:06:55: He lists four gates, but the big one, the real killer, is workflow integration.

00:07:00: If you can't fit into the hospital IT stack in under ninety days, you die in pilot purgatory.

00:07:05: That makes so much sense.

00:07:07: If I'm a nurse, I don't want another login.

00:07:09: I don't want another separate tablet.

00:07:10: I have to carry around just for this one fancy app.

00:07:13: Exactly.

00:07:14: And Greg S. posted about this for nursing specifically.

00:07:17: He argues, we need to stop calling AI tools.

00:07:20: A tool is something you have to pick up and use, and that adds cognitive load.

00:07:24: So what's the alternative?

00:07:25: He suggests we treat them as co-intelligence teams.

00:07:27: He calls it agentic AI.

00:07:29: Agentic AI?

00:07:31: That sounds a bit like buzzword bingo.

00:07:33: What does it actually mean?

00:07:34: In this context, it means autonomy.

00:07:36: Imagine an AI agent named Eddie, or auto.

00:07:39: that just works in the background, listens to the room, handles documentation, fills out billing codes, all without the nurse having to stop and type.

00:07:47: The

00:07:47: goal is to free them up to be actual nurses.

00:07:49: To practice at the top of their license, not be high-paid data entry clerks.

00:07:54: That sounds fantastic, but surely there are barriers.

00:07:57: Sebastian Kassiu pointed out a major blocker in German hospitals.

00:08:01: He called it silo thinking.

00:08:03: Yes, the classic battle between departments.

00:08:06: Sebastian says the problem isn't the algorithm, it's the structure.

00:08:09: Radiology has their budget and their IT systems.

00:08:12: Cardiology has theirs.

00:08:14: If radiology buys a system that IT hasn't vetted, you just can't roll it out.

00:08:18: So we need unified data standards, not just more flashy pilot projects.

00:08:22: We need to fix the chassis of the car before we worry about upgrading the engine again.

00:08:26: Okay, so we have the tech, but the hospital infrastructure is struggling.

00:08:30: But while hospitals fight these internal battles, patients aren't waiting, are they?

00:08:35: No, they are not.

00:08:36: Yeah.

00:08:37: And this is our third theme, and honestly, the most volatile one, the empowered patient versus the privacy gap.

00:08:43: Dr.

00:08:43: Chat GPT is officially seeing patients.

00:08:47: And it is terrifying physicians.

00:08:50: But, and this is interesting, not for the reason you might think, Chris Scato DeVeta posted about this.

00:08:55: He asked providers if they were worried about patients feeding their medical records into these AI models.

00:09:00: And what they say, were they worried about misinformation?

00:09:03: Surprisingly no.

00:09:05: They're worried because the patients are getting answers based on longitudinal data that the doctors don't even have access to.

00:09:11: Oh, wow.

00:09:12: So the patient walks in with a full analysis of their last five years of health data and the doctor is looking at a static PDF from last week.

00:09:19: Exactly.

00:09:20: It completely flips the power dynamic.

00:09:22: It erodes trust because the doctor looks less informed than the AI in the patient's pocket.

00:09:28: In the scale of this is just huge.

00:09:30: Krista Kelpie and Artie Mudaliar were discussing open AI's launch of chat GPT Health.

00:09:35: They said, forty million people are already using it daily for advice.

00:09:39: Forty million.

00:09:40: And Artie made a great point about this.

00:09:42: The general practitioner role isn't dead, but it's transforming.

00:09:46: If AI handles the first pass symptom assessment, explaining lab results, then the doctor is left with the complex cases.

00:09:54: and the human connection.

00:09:56: Which sounds good in theory.

00:09:57: It's efficient, but there is a massive catch.

00:10:00: And this is where the conversation needs to get serious.

00:10:03: Young Cha dropped a legal bombshell about privacy that I don't think people are talking about enough.

00:10:09: This is critical for every single listener to understand.

00:10:12: OK.

00:10:12: Conversations with your doctor are privileged.

00:10:14: They are legally protected.

00:10:15: But conversations with a consumer AI, they are legally treated like emails or text messages.

00:10:20: They're

00:10:20: discoverable in court?

00:10:22: Yes.

00:10:23: Imagine you're going through a custody battle or an employment lawsuit.

00:10:27: If you've been pouring your heart out to a chatbot about your mental health or your private fears, that text exists.

00:10:35: It can be subpoenaed.

00:10:36: But wait, doesn't Apple encrypt everything?

00:10:38: Does an open AI claim it's all private?

00:10:40: It's a technical distinction.

00:10:42: Apple encrypts messages so even Apple can't read them.

00:10:46: But large language models need to read the text to work.

00:10:49: That's how they generate the answer.

00:10:50: The model has to ingest your prompt.

00:10:53: So there's no real end-to-end encryption.

00:10:55: Not in the same way.

00:10:56: The company has the data.

00:10:58: That is a trade-off.

00:10:59: I don't think ninety-nine percent of those forty million users understand.

00:11:02: I agree.

00:11:02: Carolyn Bradner and Jasek put it best.

00:11:05: Buy or beware.

00:11:06: She says patients deserve the full package answers plus IPay protection.

00:11:11: Right now we're asking them to choose between intelligence and privacy.

00:11:14: And most are choosing intelligence because they're desperate for answers.

00:11:18: Okay, let's pivot a bit.

00:11:19: We've talked a lot about software, but there's a physical side to health tech, too.

00:11:23: Remote monitoring, med tech, specialized care.

00:11:26: Right, because ultimately we need to extend care beyond the hospital walls.

00:11:30: Mark D. Martini had a post about wearables, and he made a point that really stuck with me.

00:11:35: Patient deterioration happens between the spot checks.

00:11:38: That is the visibility gap.

00:11:40: In a hospital, a nurse checks on you every few hours.

00:11:44: But if you crash between those checks, no one knows.

00:11:47: Continuous monitoring isn't about having more data.

00:11:50: It's about being present when a clinician can't be.

00:11:53: But

00:11:53: getting people to use these remote tools is not easy.

00:11:56: A CMCon shared some really tough numbers on virtual care for COPD.

00:12:01: The enrollment numbers were sobering.

00:12:03: Only, what, seven and a half percent enrollment?

00:12:05: That signals massive friction.

00:12:07: And he mentioned the language barrier,

00:12:09: right?

00:12:09: He did.

00:12:10: Many of these apps are English only or require the latest iPhone.

00:12:14: And this is the equity piece.

00:12:16: If your health app requires a thousand-dollar phone and perfect English, you are widening the health gap, not closing it.

00:12:22: We need to design for everyone.

00:12:24: Austin calls it equity forward design.

00:12:26: We need to design for the grandmother who doesn't speak English and has a five-year-old Android phone.

00:12:31: Speaking of huge initiatives to close these gaps, we have to talk about access ad.

00:12:35: This was mentioned by Ignacio Valenz, Amira Romani, and Toby Cober.

00:12:40: This seems huge for Europe.

00:12:42: It's massive.

00:12:43: It's a thirty-eight million euro EU project led by Siemens Healthineers.

00:12:47: The goal is to standardize Alzheimer's diagnosis.

00:12:51: Why is standardization such a big deal there?

00:12:53: Can't you just get an MRI anywhere?

00:12:55: You can, but the data speaks different languages.

00:12:58: An MRI protocol in a clinic in Munich might be totally different from one in Madrid.

00:13:03: The data isn't comparable.

00:13:04: Axizide wants to standardize MRI protocols and blood-based biomarkers across the entire continent.

00:13:11: It's that infrastructure theme again.

00:13:12: If we want to treat Alzheimer's effectively, we all need to be speaking the same data language.

00:13:16: Exactly.

00:13:17: Before we wrap up, I want to hit just a few quick industry moves that popped up in the feet just to keep a finger on the pulse.

00:13:22: Let's hear them.

00:13:23: Logan White posted about the Omni-Secure Defibrillation Lead Launch, huge for patient safety and cardiac care.

00:13:30: Then we had Burt Van Meers discussing Phillips acquiring spectral waves.

00:13:35: A big move in cardiac imaging.

00:13:37: Consolidation is the trend there.

00:13:39: big players buying up specific tech.

00:13:41: And finally, Jill Haley mentioned the Hugo RAS system got FDA clearance for urology.

00:13:47: So robotic surgery just keeps marching forward.

00:13:50: It really paints a picture of a vibrant physical ecosystem happening right alongside all the AI software discussion.

00:13:58: So bringing it all back together, we started with this idea that twenty twenty six is an execution year.

00:14:04: What does that actually mean for the listener?

00:14:06: I think we circle right back to that car analogy.

00:14:09: The technology, the Ferrari engine is ready.

00:14:11: We have AI that can triage better than humans, models that can reason through medical records.

00:14:16: The engine is there.

00:14:16: But the challenge for twenty twenty-six isn't building a faster engine.

00:14:20: It's fixing the chassis.

00:14:21: It's the workflow, the regulation, the privacy framework.

00:14:24: We're

00:14:24: fixing the flat tires.

00:14:25: Exactly.

00:14:26: If we don't fix the broken hospital IT systems, the lack of legal protection for patient chats, and the silo departments, then that Ferrari engine is just going to shake the whole car apart.

00:14:36: We need to stop vibrating and start moving forward.

00:14:39: That is the work to be done.

00:14:40: Well, that's a wrap for this deep dive.

00:14:42: If you enjoyed this episode, new episodes drop every two weeks.

00:14:46: Also, check out our other editions on RCT and Tech Insights, Defense Tech, Cloud, Digital Products and Services, Artificial Intelligence, and Sustainability in Green ICT.

00:14:55: Thanks for listening.

00:14:57: And remember, subscribe so you don't miss the next one.

00:14:59: Catch you next time.