Best of LinkedIn: Health Tech CW 42/ 43

00:00:00: This episode is provided by Thomas Allgaier and Freeness, based on the most relevant LinkedIn posts about health tech from CW- Forty Two and Forty Three.

00:00:08: Freenes equips HealthTech providers with the market intelligence to identify which hospitals to target And how to reach decision makers for hospital digitalization as a result of The Krakkenhaus Sukunskisitz.

00:00:18: You can find more info in the description.

00:00:21: Imagine...for a second That you are A doctor In a rural clinic in Uganda Okay?

00:00:30: no stable internet connection.

00:00:32: You definitely don't have massive cloud servers humming in the back room somewhere, and a patient walks-in who might have tuberculosis

00:00:39: which is obviously a highly time sensitive issue.

00:00:42: exactly I mean five years ago you would be totally stuck waiting on slow lab results or struggling to get an expert consult right but today An AI diagnostic model that runs entirely offline just on a basic device could screen that patient right there on the spot.

00:01:01: Wow!

00:01:02: Yeah, we are entering this era where healthcare's biggest bottlenecks were finally breaking but not necessarily what you might think.

00:01:11: That is a wild image to kick us off.

00:01:13: and for your listening...that is exactly why we're bypassing all of our usual AI hype today.

00:01:17: Oh absolutely We have too

00:01:19: Right..we are looking at hard realities in digital transformation in health tech.

00:01:23: Our whole goal today actionable market intelligence on where healthcare is actually scaling.

00:01:30: Which means we need to talk about the gritty stuff, yeah?

00:01:32: The real mechanics of it

00:01:33: workflow integration how data's actually harmonized and How these massive digital platforms are finally connecting with physical things like surgical robotics in the operating

00:01:44: room.

00:01:44: because the conversation among health tech leaders has aggressively shifted recently.

00:01:48: Like nobody is asking what the technology can do anymore.

00:01:51: right that demos are already impressive enough

00:01:54: exactly

00:01:55: The only question that actually matters now is how do we get this to work inside a highly complex, highly regulated hospital system?

00:02:02: Well let's start right there with the reality of the clinical workflow.

00:02:05: There was this post from John Exario-Sillies sharing a recent report from Defined Ventures and it really puts this whole thing into perspective.

00:02:14: Oh yeah!

00:02:14: That report was incredibly telling

00:02:16: Right.

00:02:16: According their data Eighty-two percent of provider leaders now name workflow integration as their absolute biggest AI barrier, which is huge.

00:02:25: But the crazy part is that this number is up fifty one points from just last year.

00:02:30: Yeah a fifty point jump in twelve months tells you very Very clear story.

00:02:35: I mean The honeymoon phase of AI and healthcare Is officially over

00:02:39: Totally Over.

00:02:40: Hospital administrators You know they bought into the promise.

00:02:42: They saw those flashy demos he mentioned.

00:02:44: And Now?

00:02:45: Now they are facing the operational nightmare of actually implementing it.

00:02:49: Let's unpack this for you with a quick analogy, It is basically like a hospital buying state-of the art Formula One engine.

00:02:56: I love this analogy

00:02:57: Right because the tech is incredibly powerful.

00:03:00: its sitting right there in garage but they literally have no transmission to connect into wheels.

00:03:05: Yeah that can't absorb the horsepower

00:03:07: Exactly Because They just don't have internal plumbing.

00:03:09: The

00:03:12: transmission problem is the perfect way to visualize it because if hospital administrators don't build that connection soon, all this capital flowing into AI has gonna hit a massive brick wall.

00:03:22: And that impending bottleneck is exactly why large platform players are moving so aggressively right now?

00:03:27: Yeah they're trying to build the transmission themselves

00:03:30: Exactly like Cormac Waters highlighted this mass of global AI collaboration between Oracle Health and AstraZeneca.

00:03:37: Oh Right!

00:03:37: The clinical trial matching?

00:03:38: yeah.

00:03:39: So They Are Taking AI & Real World Data to identify eligible clinical trial patients sooner.

00:03:44: But the really critical detail here is how they are doing it, which is directly within the clinical workflow.

00:03:51: Wait break that down for a second.

00:03:52: what does Within The Workflow actually look like?

00:03:54: For the physician who's sitting in the room with a patient?

00:03:57: okay so historically A doctor would have To leave their primary electronic health record screen log into some completely separate research portal and manually run a search to see if their patient matched trial.

00:04:09: Which is incredibly clunky!

00:04:11: It's

00:04:11: so time-consuming, nobody wants do it.

00:04:13: What Oracle and AstraZeneca are doing Is having the system Do all that heavy lifting in the background.

00:04:18: Oh

00:04:19: I see.

00:04:19: Yeah constantly matches the patients realtime data To Trial criteria And then just surfaces A simple notification right on The doctor's main screen Exactly when they need it.

00:04:29: Okay That eliminates the friction entirely And the scale at which this is happening, it's really accelerating.

00:04:34: I mean Rommel Khalif noted that Oracle M-Forty Two Health and AstraZeneca just extended its exact partnership to the UAE

00:04:41: Really?

00:04:41: Just for trial matching?

00:04:43: No actually specifically targeting uncontrolled hypertension.

00:04:46: So they are actively moving from reactive care To proactive AI driven intervention on a national level.

00:04:54: On a National Level.

00:04:55: Yeah That Is Wild.

00:04:57: But We are also seeing this forced integration happen in much smaller rural settings too, which honestly is arguably even more difficult.

00:05:05: Oh for sure!

00:05:07: Rural hospitals are strapped for cash and staff?

00:05:09: Exactly Richard Kitchens shared that quorum.

00:05:12: health is completely overhauling its systems across eleven rural and community hospitals like dropping an Oracle Health EHR clinical AI agents and patient accounting all at the exact same time.

00:05:25: Wait All at once?

00:05:26: Yeah.

00:05:28: Doing an overhaul like that across eleven hospitals simultaneously sounds incredibly risky.

00:05:34: Why wouldn't they just phase it in slowly?

00:05:37: Because, In a rural health system you simply don't have the administrative overhead or IT staff to manage five different vendor platforms that refuse talk with each other.

00:05:46: Ah right They can afford inefficiency.

00:05:49: Exactly Unifying that data layer isn't just some nice tech upgrade for them.

00:05:53: It is literally a survival tactic to keep the doors open.

00:05:57: So they have to force the issue?

00:05:58: Yeah, sometimes you just have to Force clinical revenue and patient workflows onto one unified platform because The alternative his operational collapse.

00:06:08: That makes total sense And building infrastructure where it doesn't even exist yet seems to be a recurring theme right now.

00:06:15: David Labajo Iscierto posted about Siemens Healthineers partnering with the autonomous region of The Azores.

00:06:21: Right, the islands

00:06:22: Yeah.

00:06:22: out in the middle of Atlantic Ocean They are building an interoperable digital health ecosystem To connect these remote islands.

00:06:30: It centralizes all clinical information so every citizen gets equitable care.

00:06:35: You know, when you look at the Azores project next to that quorum health overhaul a much larger picture starts to emerge.

00:06:41: What do you mean?

00:06:42: Well as Robin Goldsmith noted from the healthcare on air summit.

00:06:45: The system itself just isn't designed to receive the innovation due to payment and adoption barriers.

00:06:51: Ah so it's not a tech

00:06:53: problem.

00:06:53: No the technology is self-works perfectly fine.

00:06:56: It's the human in financial structures That have to be completely rewired To actually receive

00:07:01: right.

00:07:01: And if workflow and adoption are the true barriers, it makes total sense that smart money is backing companies trying to just completely bypass those hurdles.

00:07:11: We're seeing this massive shift from simple AI assistance tools that require a human to actively click around and use them... ...to autonomous agents who run in the background.

00:07:21: That transition between assistant and agent?

00:07:23: That's exactly where venture capital flows right now!

00:07:26: Stanislaus-Nioch Chateau announced a planned minority investment from NVIDIA into Dr.

00:07:31: Leib.

00:07:32: Oh wow, NVIDia!

00:07:34: Yeah their AI consultation assistant has already documented over twelve million patient visits in Germany alone.

00:07:40: Twelve million visits is a staggering amount of data.

00:07:45: By the way, for you listening if your tracking these kinds investments to see where market headed definitely make sure subscribe.

00:07:50: so catch our future deep dives.

00:07:52: we will absolutely be following this transition from assistance to agents closely.

00:07:57: Oh no doubt.

00:07:59: Tell me more about Dr.

00:08:00: Libbs scale.

00:08:01: Why is NVIDIA interested in consultation assistant?

00:08:04: Because documenting those visit isn't just saving doctors time typing up their notes

00:08:09: Right.

00:08:10: Every single one of those twelve million interactions acts as training data.

00:08:14: It refines the model's understanding of nuanced clinical language, how patients express concerns and outcomes... all of it!

00:08:21: So they are building a massive proprietary dataset?

00:08:25: Exactly Dr.

00:08:26: Lib is leveraging that massive scale to build personal AI health assistant for patients by twenty-twenty seven And NVIDIA recognizes that company with most high quality real world conversational data.

00:08:38: It's going to be the one that builds the most capable autonomous agents.

00:08:41: That is fascinating, but while a company like Dr.

00:08:44: Libb is locking down the European market we are also seeing global players scale open models to really democratize this access

00:08:51: right?

00:08:51: Going back to that Uganda example We started with

00:08:54: exactly Yossi Matias shared that Google's Medjama was just published in Nature Medicine.

00:08:59: and get This.

00:09:01: it has passed ten million downloads.

00:09:04: That's

00:09:04: huge for an open model,

00:09:05: it is and It is powering completely offline triage apps in Uganda and Zambia right now

00:09:11: And Indonesia too actually.

00:09:12: yeah the Ministry of Health there Is using a to build a tuberculosis detection Model.

00:09:16: just green fifty million citizens annually.

00:09:19: but The real key To all this like we mentioned is that offline capability?

00:09:23: Yeah Let me ask you about that.

00:09:25: how

00:09:25: does an advanced AI model run off line?

00:09:28: I thought these systems required, you know massive server farms constantly crunching numbers in the cloud.

00:09:34: Well they do for the initial training.

00:09:36: You need massive compute for that.

00:09:37: Okay.

00:09:38: But once the model is trained Engineers use techniques like quantization to compress the

00:09:43: model.

00:09:44: Quantization?

00:09:46: Yeah

00:09:46: They essentially strip out the unnecessary computational weight without losing the core diagnostic accuracy.

00:09:52: So the model becomes small enough.

00:09:54: just run locally on a processor of basic smartphone.

00:09:57: Oh wow So you don't even need a signal?

00:10:00: Exactly.

00:10:01: When you remove the need for constant cloud connectivity, You instantly solve one of the biggest infrastructure barriers in global health.

00:10:08: That is incredible for accessibility.

00:10:11: But hold on.

00:10:12: let's look at the flip side Of this ambient AI revolution.

00:10:15: For second.

00:10:16: Sure

00:10:16: What happens when we start relying On technology too heavily?

00:10:20: Dr Kashif Mirza brought up A really crucial point about these Ambient AI scribes.

00:10:24: Oh The verification issue.

00:10:25: Yeah

00:10:26: Yes, they save massive amounts of time by listening to the appointment and writing the notes.

00:10:31: But they also drastically raise the verification burden.

00:10:34: like if an AI is drafting everything who's actually checking the work?

00:10:38: That's a big question.

00:10:40: Are we risking clinical skill erosion If doctors just stop writing their own notes And start blindly accepting whatever the AI drafts?

00:10:47: It's very real concern.

00:10:48: It is known as the automation paradox.

00:10:50: The

00:10:50: automation paradox?

00:10:52: Yeah, basically as a system gets better and more reliable humans naturally pay less attention to it which means when this system inevitably makes a rare hallucinated error...

00:11:03: ...the human completely misses it!

00:11:04: Exactly.

00:11:05: but human has totally checked out making that rare error much more dangerous.

00:11:10: So how do you actually solve that?

00:11:11: because you can't just tell busy doctors to pay more attention.

00:11:15: That never works

00:11:16: No doesn't?

00:11:17: Innovators are trying to solve it structurally rather than just behaviorally.

00:11:21: Sockers Sadiki shared how a company called LiveMed Health is approaching this exact tension.

00:11:26: What're they doing?

00:11:28: They built an AI clinical orchestration layer, but instead of just licensing out at random hospitals... ...they combined with their own multi-specialty physician

00:11:36: group Oh I see!

00:11:38: so they control both the technology and literally employ doctors who use them?

00:11:43: Precisely!

00:11:43: They control entire feedback loop Because they manage the human oversight directly, that aligns incentives perfectly.

00:11:49: That

00:11:50: makes a lot of sense!

00:11:51: Right... The doctors are trained specifically on how to verify AIs output and AI developers get immediate closed-loop feedback from their own clinicians.

00:12:00: By doing this They've been able safely expand specialty care access over a million Americans without constantly worrying about third party hospital misusing tech.

00:12:09: Aligning incentives is incredibly smart But you know To make these AI agents truly capable of orchestrating care like that, they need incredibly dense multimodal data.

00:12:20: Oh definitely You cannot have precision AI without precision...data Which actually brings us to a massive shift in how we handle research data.

00:12:28: Yeah We are moving from simply digitizing old hospital workflows ...to using data to discover new medicine.

00:12:34: Exactly Matthew Tobelman announced this fascinating multi-year partnership between Verily and PacBio

00:12:40: The sequencing project.

00:12:41: Yes They are linking long-read genomic sequencing, which is the patient's genotype with real world clinical records.

00:12:48: The phenotype.

00:12:49: and they're doing this to build a massive AI ready research data sets.

00:12:53: but I really need you to break down the science here for us.

00:12:56: sure what Is long read sequencing?

00:13:00: And why is it such a game changer compared to what we were doing before?

00:13:04: Okay Think of sequencing a human genome like trying to assemble a billion piece jigsaw puzzle of A completely blue sky.

00:13:11: sounds Like a nightmare

00:13:12: It is.

00:13:13: Historically, we used what's called short-read sequencing so we chopped the DNA into tiny little pieces read them and tried to stitch them all back together.

00:13:21: Okay

00:13:22: But because human DNA has so many complex repeating regions like a puzzle with thousands of identical blue pieces it was incredibly hard To assemble accurately.

00:13:31: I get

00:13:31: long reads.

00:13:32: sequencing on the other hand reads much larger continuous strands Of dna.

00:13:37: it basically gives you huge chunks of the puzzle that are already assembled.

00:13:40: Oh okay So you get a much more accurate map of the patient's genetics.

00:13:45: And when you combine that deep genetic map with their lifelong clinical history, what does the AI actually do it?

00:13:51: Well... The AI stops just looking at symptoms!

00:13:54: It doesn't just see that a patient got sick—it can analyze that highly-accurate genetic map alongside the clinical data to pinpoint the exact biological mechanism why they got sick….

00:14:04: …and U.S government is banking heavily on this exact kind of data convergence right now.

00:14:08: Oh absolutely

00:14:09: Barat Rajagopal shared that ARPA-H actually selected Verily to build a shared data platform for their circle program.

00:14:17: The stated goal here is give clinicians real time visibility into the patient's immune system, try and cut ICU stays by twenty five percent

00:14:24: which an incredible target because intensive care unit arguably most data rich and chaotic environment in whole hospital.

00:14:33: right everything constantly beeping.

00:14:35: yeah

00:14:35: patients' immune systems critical care changes hour by hour, sometimes minute-by-minute.

00:14:40: So by harmonizing data across different performing teams and using AI powered discovery tools researchers can stop wrestling with fragmented raw data queries...and start looking at real time immune

00:14:52: responses.".

00:14:53: And this data heavy approach is fundamentally changing treatment planning as well isn't it?

00:14:58: Burnt Montag highlighted that Siemens Healthineers in NVIDIA are digital twins.

00:15:06: Such a cool concept!

00:15:07: I love it, is literally like building a virtual crash test dummy that you can run scenarios on before you put a real human in the car?

00:15:14: Yeah and this specific instance they are using the Digital Twin to simulate radiotherapy for cancer patients.

00:15:21: oh wow

00:15:22: because radiation is very delicate.

00:15:24: balance right?

00:15:25: You want to destroy the tumor without damaging the surrounding healthy tissue?

00:15:28: of course

00:15:29: so but using a digital twin dozens of different radiation treatment scenarios on the virtual model of a patient.

00:15:36: It allows for agentic AI treatment planning and real-time adaptive

00:15:40: computing.".

00:15:41: So, they optimize the exact angle in dosage before physical treatments ever even starts?

00:15:46: Exactly!

00:15:47: But you know every time I hear phrases like massive multimodal data sets genomic sequencing digital twins alarm bells go off my head regarding privacy.

00:15:56: Naturally

00:15:57: How do you train these massive AI models without violating patient confidentiality?

00:16:01: I mean, you certainly can't just dump thousands of people's DNA and medical records into a public cloud server somewhere.

00:16:07: No!

00:16:07: You absolutely cannot.

00:16:09: The solution is that actually don't move the data at all... ...you decentralize the compute process.

00:16:14: Okay how does it work?

00:16:16: Well, Raghavan Ashok presented the Umbrella Stroke Research Project which was textbook example on how to solve this.

00:16:23: They built a decentralized data infrastructure So the clinical data never actually leaves the participating hospitals?

00:16:29: It just stays local.

00:16:30: Yep, it stays on-site fully protected by the local hospital's firewalls and their own governance protocols.

00:16:36: Wait but if the data never leaves the local Hospital Firewall how does the central AI model learn anything?

00:16:41: Its essentially blind to patient files!

00:16:43: It learns through a process called Federated Learning Right.

00:16:48: Instead of bringing this sensitive data up into the AI model in cloud you send a blank copy down onto the local hospitals.

00:16:55: Oh, that's clever.

00:16:56: Yeah

00:16:56: the bottle trains locally on hospital secure servers get smarter by looking at data.

00:17:01: and then here is really brilliant part.

00:17:04: only the mathematical updates.

00:17:06: you know learned weights patterns are sent back up to central server.

00:17:10: so sending them math not medical records

00:17:12: exactly.

00:17:13: The central server averages out all the mathematical updates from dozens of different hospitals, creating this master model that has learned from thousands of patients without ever actually seeing a single private medical record.

00:17:25: That is privacy by design!

00:17:27: There really is.

00:17:27: it proves that regulation and data protection don't have to limit innovation.

00:17:32: they just force the industry to build smarter infrastructure.

00:17:35: okay so we've covered the software plumbing the shift AI agents the genetic data fueling the models.

00:17:43: But all of this digital planning has to eventually hit the physical world.

00:17:48: Where does digital transformation actually meet hardware in The Operating Room?

00:17:52: This is the final crucial step, Yeah!

00:17:54: ...the physical execution Of All That Digital Intelligence.

00:17:58: Because if the surgeon's physical tools can't execute a digital plan...all that data Is completely useless.

00:18:04: And Medtronic really seems to be leading the charge here.

00:18:07: Jeff Martha shared that their stealth access surgical system was just named one of Time's best inventions of twenty-twenty six.

00:18:13: Yeah, this system is incredible.

00:18:15: This system basically unifies planning navigation and robotic assisted surgery into one single platform.

00:18:22: We are seeing real world adoption right now.

00:18:25: Roman Copat posted about Hadassah Medical Center doing Israel first Stealth Access

00:18:30: case.

00:18:30: Oh right, the scoliosis surgery.

00:18:32: Yeah it was a highly complex pediatric scoliosis correction

00:18:36: which is incredibly demanding.

00:18:38: Pediatric spine surgery you are dealing with tiny developing anatomy.

00:18:43: to perform robotic assisted pedicle screw placement on child requires level of precision that's hard.

00:18:49: even overstate.

00:18:51: I can't imagine pressure.

00:18:52: The fact that the surgeon can now use three-D planning, real time navigation and robotic trajectory alignment in one unified workflow means they could execute a highly complex digital plan with total mechanical confidence.

00:19:06: And we are seeing this shift to advanced hardware execution across multiple specialties and geographies, not just spine surgery.

00:19:13: Where else?

00:19:14: Well Amy Guthiel noted that the London Health Sciences Centre completed Ontario's first pediatric scoliosis surgery using the Mazor X robotic guidance system... ...and Ryan Sanford highlighted The Truman Memorial Veterans Hospital completing their first Boston Scientific Faripulse-Pulstfield ablation case.

00:19:31: Ah PFA!

00:19:32: Yeah let's dig into that one.

00:19:34: What makes pulsed field ablation different from how we used to treat heart arrhythmias?

00:19:39: Pulsed Field Ablation, or a PFA is just a phenomenal example of moving from theory... ...to clinical reality.

00:19:46: Traditionally To ablate-or destroy problematic heart tissue that causes those irregular heartbeat Surgeons use extreme heat or extreme cold.

00:19:55: The problem with thermal ablation Is the heater cold can spread and literally damage surrounding critical structures like the esophagus

00:20:02: which obviously carries massive risks for the patient.

00:20:05: Huge risks!

00:20:06: PFA changes the paradigm entirely, instead of temperature it uses highly specific extremely fast electrical pulses

00:20:14: Really?

00:20:15: Yeah Just electricity.

00:20:18: And these pulses are calibrated to only target cell membranes in problematic myocardial tissue so leaves surrounding nerve and esophageal tissue completely unharmed

00:20:28: So totally isolates problem without collateral damage.

00:20:31: Exactly

00:20:32: What's fascinating here is how the physical hardware is finally catching up to capabilities of this software.

00:20:37: As Subramanian Hariharan pointed out, we are witnessing a fundamental evolution in healthcare technology.

00:20:44: How so?

00:20:44: We're moving from AI that just analyzes data To AI predicts outcomes To AI actually experiments And physical hardware physically intervenes.

00:20:54: Analyze Predict Intervene.

00:20:57: It's The Full Loop.

00:20:58: We are finally taking the breaks off digital transformation because we're essentially building the transmissions capable of handling engines.

00:21:06: It has been a deeply insightful look at where this industry is moving past hype and really putting its money in effort into scaling real solutions!

00:21:15: If you enjoyed our episode, new episodes drop every two weeks.

00:21:18: Also check out other editions on cloud insights & sovereignty Digital products & services AI & agentic systems Green ICT & sustainable AI ICT and Tech Insights, and DefenseTech.

00:21:31: But before we let you go...I want to leave with one final fascinating nugget from our sources!

00:21:38: We have spent this entire deep dive talking about multi-million dollar operating rooms complex federated learning networks massive hospital infrastructure.

00:21:47: but hardware innovation isn't just happening in the surgical suite.

00:21:52: Right, Gia Tomar noted that Apple and Samsung earbuds just received FDA clearance as hearing aids.

00:21:58: Consumer

00:21:58: Earbuds?

00:21:59: Yes!

00:22:00: Apple secured it for the AirPods Pro in Samsung for Galaxy Buds.

00:22:04: The line between consumer tech and medical devices is officially gone.

00:22:08: So something to mull over if you go about your day.

00:22:11: If everyday ear buds are now an FDA cleared device The hospital of the future might not be a building you go to, it just may be clothes and devices already sitting in your pocket.