Best of LinkedIn: Health Tech CW 28/ 29

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

00:00:08: Frenis equips HealthTech providers with a market intelligence to identify which hospitals to target and how to reach decision makers for hospital digitalization as a result of the Kranchenhausukumske sets.

00:00:20: You can find more info in description.

00:00:24: Welcome To The Deep Dive!

00:00:26: For you, the digital transformation and tech professional?

00:00:28: Well...the conversation in health tech has fundamentally shifted.

00:00:31: Oh

00:00:31: absolutely!

00:00:32: I mean we are definitely no longer asking if these technologies work.

00:00:36: that ship has sailed

00:00:37: right exactly.

00:00:38: so looking at our stack of curated sources today The real question is how we govern them.

00:00:42: You know How they impact human clinical skills And how We actually manage the massive data They generate.

00:00:47: Yeah, and we'll be breaking down this shift across three core areas today.

00:00:51: So we're going to look at the reality of clinical AI governance The hidden safety risks of human-AI interaction And data driven evolution of robotic surgery and digital twins.

00:01:01: Let's just get right into it.

00:01:02: Okay sounds good.

00:01:03: so in enterprise tech We really love talking about polished AI rollouts Perfect APIs Clean interfaces.

00:01:11: Oh

00:01:11: yeah dream scenario.

00:01:13: But the reality in hospitals right now is much messier.

00:01:16: Let's talk about governance and what is actually happening in The Shadows.

00:01:20: Yeah, the shadow IT situation is honestly pretty wild right now.

00:01:24: Sigrid Berge van Roysen shared this example showing that eighty-six percent of healthcare IT leaders are reporting shadow AI in their organizations.

00:01:33: Wow!

00:01:33: Eighty six percent?

00:01:34: Yeah it's massive.

00:01:36: And because clinicians are just so overworked and they're constantly working around slow legacy systems, it's resulting in seventy-one percent of GenAI logins happening on personal accounts.

00:01:45: Wait really?

00:01:46: Personal accounts?

00:01:47: I mean that is not a privacy risk but an active security breach.

00:01:50: Exactly!

00:01:50: It's complete architectural nightmare.

00:01:52: You have patient data potentially going into public models.

00:01:55: Okay let us unpack this Because Will Conway brought up great insight here.

00:01:59: He argues healthcare AI must be treated as leadership capability Not some tech project.

00:02:04: Right, because you can't just deploy the software and look away.

00:02:07: Exactly!

00:02:08: His whole thing is that organizations must govern the shadows before they govern YOU... And to me it's a lot like early days of smartphones at work.

00:02:17: Oh totally when everyone brought their own iPhones into office

00:02:21: IT departments tried completely ban them but people used them anyway.

00:02:25: as corporate alternatives were so clunky The only real solution was providing better secure alternative.

00:02:32: But I mean, how do hospitals actually get ready for that?

00:02:35: Well this is where Dr.

00:02:36: Steve Hodgkinson's concept comes in.

00:02:38: he talks about digital fitness and i think it's a perfect analogy here.

00:02:42: Digital

00:02:42: fitness okay what does that look like in practice?

00:02:45: basically you cannot just buy your way into AI with the new software license much Like You know.

00:02:51: you can't buy physical fitness Just by getting a gym membership right.

00:02:54: you Actually have to Do The work

00:02:55: exactly.

00:02:57: Organizations need modern architecture, they needed discipline data strategy and agile platforms before even attempting what he calls the AI hustle.

00:03:06: The AI Hustle?

00:03:08: I like that yeah but assuming a hospital does all that heavy lifting.

00:03:12: i want to bring in a critical caveat here from christian hind regarding vendor lock-in.

00:03:17: oh yeah this is a massive strategic risk

00:03:20: right.

00:03:21: So what does this all mean for long-term strategy?

00:03:24: if a hospital buys into a massive AI platform today?

00:03:28: I mean they get locked in to that ecosystem, right.

00:03:30: Yeah and Heinz's warning is pretty stark.

00:03:32: he says that In three years swapping out the actual AI model might just take an afternoon because

00:03:39: The models themselves become commoditized.

00:03:41: Right exactly

00:03:42: but leaving the platform That holds your institutional memory.

00:03:45: Well, that will be a board level crisis.

00:03:47: Gufford-level

00:03:47: crisis?

00:03:48: Yeah!

00:03:48: Just from changing platforms...

00:03:49: Yeah because in healthcare under GXP audits good clinical practice standards you have to be able to export the full reasoning trail of why decision was made.

00:03:58: Oh

00:03:58: wow so you need the whole context not just the final output.

00:04:01: Exactly if you leave the vendor You can't leave that reasoning behind.

00:04:05: you have To be able cleanly extract your organization's entire cognitive framework Which is just incredibly difficult.

00:04:12: That makes total sense.

00:04:14: So we really have to govern the data and platforms, but what happens when these systems are fully integrated?

00:04:21: How does relying on those algorithms actually impact human clinicians who operate them day in or out.

00:04:27: Well that brings us our second theme.

00:04:29: The safety of clinical AI isn't just about code.

00:04:32: it's realy about the human-in-the loop reality

00:04:34: Right!

00:04:35: Human element.

00:04:36: And Jan Becker detailed this fascinating Year-long study on an AI planning tool in radiation oncology.

00:04:43: Oh, this study is so telling

00:04:45: it really is.

00:04:46: initially It was a huge success the planting with fifteen percent faster.

00:04:49: things look great On paper but pretty soon clinicians started suffering from what?

00:04:53: The researchers called intuition rust

00:04:55: Intuition rust.

00:04:57: that has such a vivid way to describe it.

00:04:58: yeah

00:04:58: Its wild the decimatrice who used to build multiple manual plans From scratch.

00:05:03: they just started blindly accepting the single AI plan.

00:05:06: and

00:05:06: that Is exactly we call asymptomatic harm.

00:05:08: its this real, measurable skill loss that is totally invisible on standard hospital performance dashboards.

00:05:15: Because throughput is up so the dashboard looks green

00:05:19: Exactly!

00:05:19: The system looks highly efficient but human operators are quietly losing their edge.

00:05:24: It's a huge operational blind spot.

00:05:27: By the way, if you are finding this deep dive valuable make sure to subscribe so we can catch our future additions where we track these exact kinds of operational shifts.

00:05:34: Definitely and you know that human failure to synthesize algorithmic output gets even more concerning!

00:05:39: Let's look at The No-Harm Study shared by friends MJ Fester.

00:05:43: What did they find?

00:05:44: So,

00:05:44: they evaluated forty-five large language models and over one hundred practicing physicians.

00:05:49: And the AI medical advice carried a potential for severe harm in twenty four point six percent of

00:05:54: cases.

00:05:55: Wow!

00:05:55: Almost twenty five percent.

00:05:56: That sounds like complete deal breaker for rolling this out.

00:05:59: Right.

00:05:59: But here's the kicker Doing nothing Just relying entirely on the unaided.

00:06:04: human doctors Had a thirty seven percent harm rate.

00:06:07: Wait really A Thirty Seven percent Harm Rate For The Humans Yeah.

00:06:10: And what's fascinating here is that over eighty percent of those severe AI errors were actually omissions.

00:06:16: So critical advice, that was just never given!

00:06:19: Ah I see.

00:06:20: so the AI isn't necessarily hallucinating bad treatments it's just withholding information to play its safe.

00:06:26: Exactly The guardrails are too tight Furthermore, while the AI-supported doctors did perform better overall they frequently just ignored their really valuable AI recommendations.

00:06:36: So that human cognitive load is too high to merge them properly?

00:06:40: Right!

00:06:40: Combining human and AI would theoretically yield absolute best results but humans are basically failing to synthesize it properly.

00:06:49: How does industry actually fix this fragmented safety landscape?

00:06:54: Because right now it feels like everyone is just flying blind.

00:06:57: Well, Dr Nick made this powerful comparison to the aviation industry that really maps perfectly to this problem.

00:07:02: Oh I love a good aviation analogy.

00:07:04: What's the connection?

00:07:05: So in nineteen seventy two Eastern Airlines flight four one crashed and The reason it crashed was because airlines simply didn't share near miss data with each other.

00:07:14: oh wow so another airline had the same issue but Didn't tell anyone

00:07:17: exactly.

00:07:18: And right Now healthcare AI Is essentially learning One tragedy at A time

00:07:22: That Isn't terrifying.

00:07:23: Thought

00:07:24: It really is.

00:07:24: A hospital in Chicago might catch an AI hallucination locally, but the rest of the industry remains completely blind to it.

00:07:32: Healthcare AI desperately needs a shared safety intelligence system just like aviation has

00:07:37: today.".

00:07:38: So as standardized data pipeline to broadcast those anomalies across the whole industry?

00:07:43: Exactly!

00:07:43: Okay so if we can master the safety and the governance and that human integration... The actual capabilities of these technologies become incredibly powerful which brings us to the physical and digital frontiers, right?

00:07:58: Robotics and digital trends.

00:07:59: Yeah this is where the massive data generation really starts redefining the sharpest edges of patient care.

00:08:05: Okay tell me about the robotics side.

00:08:07: So Cristal Croft shared some really interesting reflections on a dataset over thirteen hundred robotic joint replacements

00:08:13: Over a thousand.

00:08:14: that's a solid data set.

00:08:15: yeah

00:08:16: And we usually think of surgical robots purely for their physical precision.

00:08:19: Right like making the perfect

00:08:21: cut The mechanical advantage

00:08:22: Exactly, but the robot's real underdog strength is actually the massive volume of continuous data it generates for post-op insights.

00:08:33: Every micro movement is tracked and quantified.

00:08:35: Oh that makes total sense.

00:08:37: Jim Pache will add a great perspective on this.

00:08:40: regarding the Hugo robotic surgery platform

00:08:42: oh yeah what did he say about?

00:08:44: Well He pointed out its built as modular connected system meaning you aren't just starting over from scratch every single procedure.

00:08:51: the system actually gets smarter with every use.

00:08:53: Because it's feeding all that telemetry data back into its own foundational model?

00:08:57: Exactly, It is a closed loop ecosystem.

00:09:00: And you know I connect this massive data obsession directly to digital twins.

00:09:05: Sajit Katyaar and Mustafa al-Sharif wrote about this.

00:09:08: Digital twins, I hear this buzzword all the time.

00:09:10: but what does it actually mean in this context?

00:09:13: Well In theory digital twins combine electronic health records continuous wearables And AI to simulate a patient's biological state.

00:09:21: Okay To do What exactly?

00:09:22: To simulate risk like predicting a patients risk of sepsis or even modeling hospital bed shortages hours before The harm Actually occurs.

00:09:31: okay here's where he gets really interesting though.

00:09:34: Are these actually digital twins of a patient, or are they just really educated guesses?

00:09:39: That

00:09:40: is the exact critique Edward you brought up.

00:09:42: It's very valid epistemological constraint

00:09:45: Right because biology but our medical data like MRIs or blood tests, they're just discrete snapshots.

00:09:53: Exactly!

00:09:53: It's not a continuous video it is more stop motion photography

00:09:57: Stop Motion Photography?

00:09:58: That the perfect way to describe it.

00:10:00: We have massive gaps of unobserved time between a scan on Tuesday and a blood draw Friday.

00:10:05: Yeah we need to be incredibly careful about what direct observation versus an AI inference filling in those gaps.

00:10:12: Because if the AI infers that patient is crashing based on historical data than real-time telemetry, treating them based on that guess could be a massive liability.

00:10:21: Completely which is why massive foundational data is absolutely required to make these twins even remotely accurate.

00:10:28: You really need true scale to pull this off

00:10:30: Exactly and Christa Cowpey gave a fantastic update on Finland's fine health foundry project.

00:10:37: Oh what are they doing in Finland?

00:10:38: They're basically building the world first nationwide healthcare foundation AI model.

00:10:44: Nationwide, that's incredible!

00:10:45: Yeah they're leveraging decades of high-quality health registry data and running it through super computing infrastructure.

00:10:53: That is what true scale looks like when you want to build predictive models that actually work.

00:10:57: So its not just a siloed hospital database anymore It's an entire nation state level ecosystem.

00:11:02: Precisely Wow.

00:11:04: Okay.

00:11:04: so Summarizing all of this for you, the listener.

00:11:08: The overarching takeaway is that the narrative in health tech has truly matured.

00:11:13: It's no longer about flashy pilot demos or cool localized software.

00:11:18: No we are way past

00:11:19: them.

00:11:19: Right.

00:11:19: it's about securing data to avoid vendor lock-in and preventing human skill erosion And ultimately building continuous learning ecosystems where technology and clinicians operate as a single governed unit.

00:11:31: Exactly, and you know just building on that it leaves me with this really provocative thought about the future of medicine.

00:11:37: Hmm let's hear

00:11:39: If AI makes hospitals faster but quietly dulls human intuition, like we saw with the Intuition Rust in Radiation Oncology.

00:11:48: How do we actually design medical training for next generation of doctors?

00:11:52: That is a fascinating question!

00:11:53: Right...

00:11:54: For practitioners who have literally never practiced without an AI assistant guiding them will eventually need mandatory manual mode simulations just to keep their baseline survival instincts sharp.

00:12:06: Just randomly turn the AI off and say alright figure it out Exactly

00:12:09: like a fire drill for clinical intuition.

00:12:11: I love that.

00:12:12: it really makes you wonder what medical school looks like in ten years.

00:12:15: Well, he enjoyed this episode.

00:12:16: new episodes drop every two weeks.

00:12:18: also check out our other editions on cloud defense tech digital products and services artificial intelligence sustainability And green ict ict and Tech Defense Tech and HealthTech.