Best of LinkedIn: Health Tech CW 12/ 13

00:00:00: This episode is provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts of health tech in CW-Twelve and Thirteen.

00:00:08: Frenness equips HealthTech providers with The Market Intelligence to identify which hospitals target.

00:00:24: Imagine a hospital board, right?

00:00:26: They just signed off on this massive I don't know ten million dollar artificial intelligence suite.

00:00:31: Oh

00:00:31: yeah the classic big-budget tech drop

00:00:34: exactly.

00:00:34: and then Six months later they find out that the clinicians On The actual floor are Just absolutely terrified.

00:00:40: to turn the thing on I mean it's just sitting there gathering

00:00:42: digital dust.

00:00:43: basic.

00:00:43: Yeah Exactly.

00:00:44: And That disconnect that's exactly what we're unpacking today.

00:00:46: over the past couple of weeks We've been tracking This this sweeping wave of insights across the health tech landscape.

00:00:53: Right,

00:00:53: mostly looking at calendar weeks twelve and thirteen.

00:00:57: Yeah And our mission for this deep dive is to basically cut straight through The promotional fluff.

00:01:02: We're gonna show you how digital transformation Is actually surviving contact with the real world.

00:01:09: Honestly that survival is defining challenge Of industry right now.

00:01:13: Like if look at curated insights we have A really clear pattern emerges.

00:01:18: It's a vibe shift

00:01:19: Totally.

00:01:19: We are finally moving past that era where a piece of tech could get fronted just by being, you know shiny and novel.

00:01:26: Yeah

00:01:27: the AI for the sake of AI phase.

00:01:28: Exactly!

00:01:29: The market is aggressively demanding.

00:01:31: shift toward workflow reality.

00:01:33: I mean an algorithm has to function inside super chaotic highly regulated structures of actual hospital or well it simply becomes very expensive shelf wear.

00:01:42: Yeah, so let's jump right into our first major theme which hits exactly on that friction.

00:01:47: AI adoption.

00:01:49: it really feels like artificial intelligence and health care has hit this fascinating operational wall.

00:01:54: It definitely has.

00:01:55: Meenal Shah pointed us out brilliantly in a recent post.

00:01:58: she noted that physicians were adopting AI primarily for things like ambient scribing.

00:02:02: Right!

00:02:03: Which is huge right now?

00:02:04: And uh... For anyone listening who isn't familiar with the term Ambient Scribing is essentially when an AI securely listens to the conversation between the doctor and patient in the background.

00:02:15: And it just automatically drafts clinical notes.

00:02:17: It's like having a hyper-efficient, invisible assistant...

00:02:20: Exactly!

00:02:21: And Minol observed that doctors are adopting this eagerly because well..it solves an immediate deeply painful daily need.

00:02:29: Yeah

00:02:29: The charting burden is just brutal

00:02:32: Right.

00:02:32: But what they're not adopting Are these tools built solely for generic hospital revenue capture.

00:02:39: And it's such a perfect example of misaligned incentives between the buyer, you know?

00:02:43: The hospital admin and the actual end user.

00:02:46: Tell me about it

00:02:47: Like administrators often purchase software to optimize what they call an RVU or relative value unit.

00:02:53: Right...the

00:02:54: golden metric?

00:02:54: Yeah

00:02:55: that is the metric hospitals use to measure clinician productivity in figure out billing.

00:03:00: So from the administration perspective An AI that captures more RVUs Is fantastic investment.

00:03:05: Sure looks great on spreadsheet

00:03:07: But the clinician on the floor, they do not care about RVUs when they are you know staring down a two-hour backlog of documentation at seven o'clock at night.

00:03:17: Exactly I love that distinction.

00:03:19: it is um It's like trying to sell a Formula One engine To someone who just needs a reliable commuter car.

00:03:25: That is exactly it.

00:03:26: You can show a doctor an AI model with a billion parameters, but if it doesn't help them with that miserable documentation commute at the end of their shift they will just completely ignore.

00:03:38: Yeah!

00:03:38: They're trying to get home for families.

00:03:40: Right.

00:03:40: Linal's core insight was that developers really have to segment their products by this specific workflow friction they are relieving The closer your solution gets.

00:03:49: improving the Kleditions Live Daily experience...the warmer reception.

00:03:54: And you know ignoring daily reality is what leads directly that Asim Khan recently described as pilotitis.

00:04:03: Oh,

00:04:03: pilotitis?

00:04:04: That is a great word

00:04:04: isn't it?

00:04:05: he highlighted this really sobering dynamic currently playing out over in the UK specifically within The National Health Service.

00:04:11: What's

00:04:11: happening there...the

00:04:12: issue Is that for A lot of these startups securing public funding For an AI Pilot is actually ironically the beginning Of their downfall.

00:04:22: Wait I need to push back on that a bit.

00:04:23: Sure Surely the whole point.

00:04:27: generate the academic and clinical evidence you need to justify long-term adoption, how does winning a pilot become a downfall?

00:04:36: It's a fair question.

00:04:38: The problem is that the metrics for successful pilots are rarely the same as the metrics of successful large scale enterprise software.

00:04:46: I see...

00:04:46: Right!

00:04:47: A pilot might prove an algorithm can detect disease with high accuracy in super controlled environment but it doesn't prove who will pay out of operational budget

00:04:57: or how it talks to the horrible legacy IT systems a hospital is already stuck with.

00:05:01: Exactly, Asim noted that without clear predefined implementation pathways you just end up with an endless repetition of pilot studies that literally never scale.

00:05:09: Wow!

00:05:10: Just stuck in purgatory

00:05:11: Pretty much In fact he pointed out some companies are abandoning UK market entirely and moving their operations into US

00:05:18: Really?

00:05:19: Leaving completely?

00:05:20: Yeah

00:05:21: They are seeking out environments where the financing mechanisms are a bit clearer, and adoption can actually become the default outcome rather than just this endless loop of testing.

00:05:32: So we have this massive disconnect between what the administrators buy and what the doctors need compounded by an economic disconnect in how these tools scale?

00:05:40: It's amiss

00:05:41: But you know there was another layer to that.

00:05:43: I think is even more concerning.

00:05:44: Oh!

00:05:44: You're talking about systemic blind spots.

00:05:46: Yes exactly

00:05:48: Jan Beger recently visualized this perfectly.

00:05:51: He created this thematic co-occurrence heat map based on an analysis of a hundred and sixty-two recent healthcare AI papers.

00:06:01: Hundred

00:06:01: and sixty two?

00:06:01: That's a lot reading!

00:06:02: Yeah,

00:06:03: he basically clustered the paper by topic just to see where the industries intellectual energy is actually being spent

00:06:08: And I'm guessing that dense zones were exactly what we'd expect

00:06:11: Entirely predictable.

00:06:12: Everyone was researching radiology AI General AI governance and large language models.

00:06:18: I mean there are eighty-two papers focused solely on radiology.

00:06:21: Right

00:06:21: But it's empty spaces in map most revealing

00:06:25: Exactly.

00:06:26: This is the scary part, out of one hundred and sixty-two papers only eight addressed legal liability.

00:06:32: Eight?

00:06:33: Out Of One.

00:06:33: Hundred and Sixty Two?

00:06:34: Yep

00:06:35: And Only Ten Bothered to Look at The Patient's Perspective.

00:06:38: Wow What Yon pointed out as particularly alarming Is that the intersection between quote Workforce & De skilling and Legal Liability is essentially a ghost town in the research.

00:06:50: Okay, let's unpack the mechanics of that intersection for a second because if I am a hospital administrator.

00:06:55: this is the exact scenario that keeps me awake at night.

00:06:57: oh

00:06:57: absolutely.

00:06:58: we know from human factors engineering that when you give a highly trained professional an automated co-pilot they eventually lean on it

00:07:06: like heavily.

00:07:07: It's human nature, right?

00:07:08: They can become de-skilled over time.

00:07:10: so if an AI assisted clinician starts performing worse on the days when they say offline or malfunctioning and a patient is harmed as a result how does the legal system even handle that?

00:07:21: That Is The Million Dollar Question!

00:07:23: Is the liability On The Software Vendor Who Wrote The Code ?

00:07:27: Is it On The Hospital The Potter?

00:07:29: Or Is The Doctor Still Ultimately Responsible Because They Are The Licensed Professional In The Room?

00:07:35: The scary thing is the industry does not have a standardized answer to that yet.

00:07:40: Connor Finn weighed in on this exact vulnerability recently.

00:07:43: What did he say?

00:07:44: He warned that health systems are running the risk of simply adding, quote, smarter layer of complexity On top an already overburdened tech

00:07:53: stack A Smarter Layer Of Complexity.

00:07:55: That's a great way To Put It.

00:07:56: Yeah

00:07:57: Without a rigorous governance backbone that defines those liability and fallback procedures, you aren't really transforming healthcare at all.

00:08:05: You are just deploying software and like crossing your fingers.

00:08:08: Looping for the best!

00:08:09: Exactly.

00:08:10: Connor emphasized we need platforms that earn clinical trust through clear operational governance rather than isolated applications completely falling apart when they encounter real world ambiguity.

00:08:22: Right but... you know, if general clinical AI is struggling with trust and liability let's pivot.

00:08:29: And look at a domain where the technology actually seems to be winning the battle for workflow integration.

00:08:34: You're thinking of imaging.

00:08:35: Exactly Identifying anomalies inside the human body Is where we are seeing the most tangible progress right now.

00:08:42: Claire Sadler de Sousa Ibrido shared an incredibly insightful take away from the ECR twenty-twenty six conference over in Vienna.

00:08:49: Oh yeah!

00:08:50: i saw that

00:08:50: Her observation was that radiology doesn't need louder innovation.

00:08:56: What clinicians actually want is dependable precision.

00:08:59: They are practically begging for technology that reduces their cognitive load.

00:09:03: And that phrase, Cognitive Load is just so vital to understand here.

00:09:07: Break

00:09:07: that down for us!

00:09:08: Think about it... A radiologist spends hours in a dark room staring at various shades of gray trying spot millimeter-sized anomalies.

00:09:16: The mental fatigue is profound.

00:09:18: It has to be exhausting.

00:09:19: Christian Wolfram shared really concrete example how to alleviate the fatigue.

00:09:23: He cited a recent CT lung cancer screening pilot at Metro Health in Cleveland.

00:09:28: Okay, and how did that go?

00:09:29: It was a notable success.

00:09:31: it cut the perceived read time for the radiologists by roughly twenty percent.

00:09:36: Wow!

00:09:36: Twenty percent's of massive times safe.

00:09:39: But the underlying reason for that success is what really matters here.

00:09:44: I am guessing they didn't just drop a brand new user interface on The Doctors And expect them to figure out over the weekend.

00:09:49: Precisely

00:09:50: It succeeded because the developers actually went in and shadowed the radiologists.

00:09:55: Oh, they watched them work?

00:09:57: Yeah

00:09:57: They observed their eye movements Their clicking habits The exact sequence of their analysis And then...they fitted the AI seamlessly into those existing habits.

00:10:07: That makes so much sense!

00:10:08: ...They did not force doctors to learn some new convoluted software logic.

00:10:13: The system was molded to the human.

00:10:15: That brings us right back to a commuter car analogy from earlier.

00:10:18: Exactly!

00:10:18: It just works without demanding extra effort from the driver.

00:10:21: But you know, underneath that simple user experience... ...the technical engines powering these tools are becoming incredibly sophisticated.

00:10:29: Oh they're mind-blowing.

00:10:30: Taha Casout shared an update on something called DecipherMR.

00:10:34: This is what is known as a vision language foundation model,

00:10:37: right?

00:10:38: The multimodal stuff.

00:10:39: Yeah

00:10:39: They trained this AI on over two hundred thousand three D MRI series.

00:10:44: But the kicker is they didn't stop at just the images.

00:10:47: we fed it to text

00:10:48: too.

00:10:49: Right exactly.

00:10:50: They also fed them model the corresponding text-based radiology reports written by human doctors.

00:10:56: the methodology there's just fascinating because historically MRI is a notoriously difficult modality for artificial intelligence to master.

00:11:05: Why's

00:11:06: that?

00:11:06: Well, the data is three-dimensional for one.

00:11:08: The anatomical variations are basically endless.

00:11:11: Hospital scanning protocols differ wildly and paying human experts To manually label every single slice of an MRI it's prohibitively expensive.

00:11:20: So how does adding text into the mix actually solve the image problem?

00:11:23: It's all about joint image text pre training.

00:11:26: by doing that The algorithm learns to correlate the visual pixel data with the semantic meaning of words.

00:11:32: Ah, I see!

00:11:33: Yeah it completely bypasses need for manual labeling by basically learning relationships between the three D shapes and doctors written diagnosis.

00:11:42: That's incredibly clever.

00:11:43: It really is.

00:11:45: as a result Taha noted that decipher MR achieved a ninety nine point nine percent accuracy rate in simply recognizing the type of MRI scan being viewed.

00:11:53: Ninety-nine

00:11:54: point nine, that's practically perfect.

00:11:57: And it is also achieving over eighty five percent accuracy and distinguishing major forms of heart disease.

00:12:03: That is a staggering technical achievement.

00:12:05: But here's what I don't get.

00:12:07: What's that?

00:12:08: A foundation model processing hundreds of thousands of high-resolution, three D scans simultaneously.

00:12:15: that requires an astronomical amount of competing

00:12:18: power.

00:12:18: Oh absolutely!

00:12:19: You cannot run a system like DecipherMR on some ten year old server sitting in a hospital basement?

00:12:24: No you absolutely can not which is why the infrastructure discussions happening behind the scenes right now are just as critical as the algorithmic breakthroughs themselves.

00:12:32: Right...is

00:12:32: the plumbing makes it all work?

00:12:34: Exactly Simon Philip Ross and Nina Ikerli Lindsay Both shared some commentary recently on GE Healthcare Acquiring in Tellarad.

00:12:42: Oh, that was a big move!

00:12:43: Huge.

00:12:44: and the explicit strategic goal of that acquisition is to build a connected cloud-first AI enabled enterprise imaging platform

00:12:53: because the old systems just can't handle data.

00:12:56: Exactly, as volume of imaging data explodes across these hospital networks The old on-premise data silos simply buckle under weight.

00:13:05: You need an interoperable cloud native ecosystem to handle bandwidth.

00:13:09: If infrastructure chokes workflow grinds into a halt before AI delivers its insight.

00:13:15: We've talked about friction in diagnosing patient But finding the tumor is really only half of battle.

00:13:21: True Once algorithm identifies anomaly, The physical execution removing it where next bottleneck lies Which brings us to transformation of procedure

00:13:30: room itself.

00:13:30: The operating theater changing so

00:13:32: fast It's rapidly turning into a connected digital and robotic ecosystem.

00:13:39: David Larson at Mayo clinic shared some data recently regarding robotic colorectal surgery that honestly genuinely changes calculus for patient care.

00:13:49: The Mayo Clinic data provides such a highly objective lens on this transition.

00:13:54: What were the specific outcomes they observed?

00:13:57: So, they looked specifically at emergency colorectal procedures which are historically very complex and high risk.

00:14:04: Oh yeah emergencies are always tougher.

00:14:05: Exactly

00:14:07: When surgical teams utilize robotic assistance, the conversion rate dropped from twenty-four point two percent down to seven point eight percent.

00:14:13: Wow!

00:14:13: Seven point eight?

00:14:15: Yes and for anyone who hasn't spent time in an operating room a conversion is the absolute last thing a surgeon wants.

00:14:21: today.

00:14:21: It's

00:14:21: the worst case scenario.

00:14:22: Yeah

00:14:23: it means they start with a minimally invasive keyhole approach but then run into complication have convert to full open surgery

00:14:29: which mean drastically longer recovery times much higher risks of patient.

00:14:34: Exactly So dropping that conversion rate by that massive margin is just a profound win for clinical outcomes.

00:14:41: It really is!

00:14:42: On top of that, they saw a fifty-one percent reduction in the overall odds of surgical complications compared to traditional laparoscopic approaches.

00:14:50: Fifty

00:14:51: one percent?

00:14:51: That's game changing right.

00:14:53: The projections now suggest that robotic utilization and these types of emergencies will hit twenty percent By the year.

00:14:59: twenty twenty five.

00:15:00: it's coming

00:15:00: up fast

00:15:02: But this data brings up a really sticky question for me.

00:15:05: Let's hear it!

00:15:06: If the robotic console is, you know stabilizing the surgeon's tremor scaling their movements and basically doing the heavy lifting.

00:15:14: how do we train the next generation of residents?

00:15:16: Like to they just become supervisors watching a screen...

00:15:19: It's a great point.

00:15:20: And your pointing right back to the de-scaling paradox.

00:15:22: We discussed earlier with Jan Beger heat map Right

00:15:25: but applied to the operating

00:15:26: room.

00:15:27: Exactly!

00:15:28: However, good news is that the surgical field is actively adapting its training models... ...to prevent this exact scenario.

00:15:35: Okay how so?

00:15:35: Marc-Alevié Sauvain shared a really great milestone from his hospital, RH&E.

00:15:41: They recently celebrated their one hundredth robotic case

00:15:45: Nice

00:15:46: And key part of it was having chief resident train on the Hugo RAS console.

00:15:52: The training methodology they use now essentially mirrors the aviation industry.

00:15:56: Okay,

00:15:56: the Aviation Model makes a lot of sense.

00:15:58: walk me through how that actually works in practice.

00:16:00: So before our resident ever touches a patient with a robotic instrument They undergo extensive hours of simulation

00:16:07: like a flight simulator

00:16:08: Exactly!

00:16:09: They master the console mechanics build the necessary muscle memory and learn How this system responds to different inputs In a purely virtual environment.

00:16:18: so there's zero risk to a real human

00:16:19: right?

00:16:20: And only after mastering the simulator do they move on to a supervised, quote-unquote assisted flight in the actual operating room.

00:16:27: I love that concept!

00:16:28: It is an ethos built entirely around progressive autonomy.

00:16:32: The goal is not to replace the surgeon's fundamental anatomical knowledge... ...the goal was to structure the pedagogy so they deeply understand the limits of technology before relying it.

00:16:42: That makes perfect sense.

00:16:43: Adopting tech is important but transmitting the underlying surgical judgment is absolutely essential.

00:16:48: And speaking of that specific hardware, The Hugo RAS system is officially gaining serious momentum in the U S market right now?

00:16:56: It definitely is.

00:16:57: Geotin Agrawal and Gregory Hake both highlighted that this system recently received FDA clearance.

00:17:03: it has already performed its first commercial US cases at the Cleveland Clinic.

00:17:08: That's a huge milestone, but you know we also have to expand our definition of what robotics entails in this setting.

00:17:14: What

00:17:14: do you mean?

00:17:15: Well it does not just mean physical mechanical arms cutting tissue.

00:17:19: We are also seeing the rise of digital algorithmic co-pilots inside The procedure room

00:17:25: like software guiding the hardware

00:17:26: exactly A tool.

00:17:27: Gupta shared a recent update that Phillips received FDA clearance for a tool called device guide.

00:17:34: It's an AI application designed to provide real-time visualization and tracking of mitral valve repair devices while they are inside a beating human heart.

00:17:43: Wait, explain the mechanism there because tracking a tiny device

00:17:53: It's super complex.

00:17:54: In a standard procedure, surgeons rely on imaging feeds like fluoroscopy or ultrasound to see what they're doing right?

00:18:01: But those feeds can be really visually noisy and as you said the heart is constantly moving.

00:18:06: so it was hard to see that tiny tool

00:18:08: exactly.

00:18:09: So device guide uses AI to analyze that live Imaging feed Locate this specific structural repair device And artificially enhance its visibility on the screen.

00:18:18: Oh

00:18:19: wow!

00:18:19: It highlights Yeah...It

00:18:21: dynamically tracks the moving tissue, allowing the surgical team to navigate with absolute precision.

00:18:27: It is all about taking an incredibly chaotic visual environment and using algorithms to turn that complexity into

00:18:33: clarity.".

00:18:34: That's wild!

00:18:35: So let us take a step back and look at this sheer volume of data being generated by everything we have just discussed today... It

00:18:41: s a literal mountain

00:18:47: Foundation models analyzing hundreds of thousands three-D MRI scans, robotic surgery consoles logging every millimeter movement and AI enhancing live cardiac video feeds.

00:18:58: It never stops!

00:18:59: This brings us to our final theme health data infrastructure.

00:19:03: who is actually connecting organizing protecting this absolute tsunami information?

00:19:09: It is arguably the defining strategic question for health tech over the next decade.

00:19:13: And interestingly in Europe, the regulatory rules governing that data have just been completely rewritten.

00:19:18: No really?

00:19:19: Yeah Fernanda Sousa highlighted a profound legal shift recently.

00:19:23: The European Health Data Space or EHDS has officially become law.

00:19:28: Okay, we hear about new European tech regulations constantly.

00:19:31: Is this just you know another set of compliance guidelines that companies will eventually work around?

00:19:35: or does this fundamentally change how the technology operates?

00:19:38: It fundamentally changes the architecture of the entire market.

00:19:40: Wow

00:19:41: okay

00:19:41: The EHDS mandates that interoperability is no longer a feature You can charge extra for Or some optional pilot program.

00:19:49: it is now a mandatory cross-border legal requirement.

00:19:53: So you have to play nice with everyone else's data.

00:19:55: Yes

00:19:56: The law creates a common technical framework for electronic health data across the entire European Union.

00:20:03: If your HealthTech solution, whether it is an AI scribe or imaging platform cannot securely integrate exchange and log-data according to these specific common rules you can not participate in the market

00:20:15: period.".

00:20:16: But wait if startups even establish vendors are suddenly forced rapidly rebuild their backend architecture Aren't they simultaneously opening themselves up to significant security vulnerabilities?

00:20:30: That's

00:20:30: the catch.

00:20:31: I mean, when engineering teams are forced to move that fast... ...to meet a compliance deadline things inevitably break

00:20:36: And that is the hidden danger of scaling infrastructure too quickly.

00:20:39: Larry Trotter brought up a highly relevant point regarding this exact vulnerability.

00:20:43: What did Larry say?

00:20:44: Well!

00:20:45: When the public hears about a health tech company suffering a HEPA failure or devastating data breach our minds immediately go to like elite State-sponsored hackers deploying zero day

00:20:58: exploits.

00:20:58: right guys and hoodies in dark rooms

00:21:00: exactly.

00:21:01: But Larry points out that the reality is

00:21:03: far more mundane.

00:21:04: if it isn't elite hackers.

00:21:06: What is actually causing?

00:21:07: The breaches.

00:21:08: unfinished foundational security work just

00:21:11: sloppy coding.

00:21:12: essentially

00:21:12: yeah, It Is the routine unsexy administrative tasks they get left behind as a company scales its operations too fast.

00:21:20: Health tech vendors especially on their growth stage often skip over the boring maturity steps like what.

00:21:25: They delay implementing strict data governance policies, they don't properly configure their access logs and skip routine permission audits.

00:21:48: That's a perfect analogy.

00:21:50: and in a highly regulated environment like healthcare, skipping those fundamental security steps will always catch up with you.

00:21:57: It has to!

00:21:57: it will either result in a crippling regulatory fine or a breach that destroys your clinical trust entirely.

00:22:05: Absolutely

00:22:05: so bringing this all together for you the listener we are looking at a landscape where artificial intelligence must prove its worth by eliminating the daily administrative friction.

00:22:17: We are seeing medical imaging and robotic surgery achieve unprecedented levels of precision by adapting to the human operator.

00:22:25: And we're watching the underlying data infrastructure be forced into maturity, by strict new European

00:22:32: laws.".

00:22:32: It's an incredible time of transition!

00:22:35: As we accelerate in this era of algorithmic efficiency.

00:22:38: I actually want you with a completely different angle.

00:22:42: We talked extensively about the liability of AI and potential for clinical de-skilling, right?

00:22:49: Exactly.

00:22:49: But think how this shifts the economics of medical malpractice insurance... If robotic consoles and digital co-pilots drastically reduce emergency complications as that Mayo Clinic data suggests… The baseline expectation for human error is going to drop.

00:23:07: Because the robots make them so much safer,

00:23:09: right?

00:23:10: So will we eventually reach a tipping point where insurance underwriters refuse to cover a hospital that doesn't use robotic assistance simply because the unassisted human error rate is deemed too high of financial risk?

00:23:23: This is a fascinating thought!

00:23:24: We might see a future where adoption isn't driven by clinical preference at all.

00:23:29: but

00:23:32: What a wild concept to end on.