Best of LinkedIn: Health Tech CW 18/ 19

00:00:00: This episode is provided by Thomas Allgeier and Frennus.

00:00:03: Based on the most relevant LinkedIn posts, on health tech in CW-Eighteen and Nineteen, Frennis equips HealthTech providers with The Market Intelligence to identify which hospitals target.

00:00:24: Welcome to this deep dive.

00:00:26: We're we're getting into something really massive today.

00:00:28: Yeah,

00:00:29: we really

00:00:29: are because if you work in health tech or You know even it's just a patient navigating the system?

00:00:34: You really need to know that we were officially moving out of The AI height phase

00:00:39: right the theoretical phases over

00:00:40: exactly.

00:00:41: we are diving headfirst Into the deployment phase.

00:00:43: mm-hmm like imagine an artificial intelligence That doesn't just you know passively draft and email for you but acts as a fully autonomous medical coordinator orchestrating your lab results, booking your follow-ups and building clinical workflows while you sleep.

00:00:58: And the shift over the past few weeks is just palpable.

00:01:02: I mean we are no longer debating what technology might theoretically achieve in a decade right?

00:01:08: We're looking at the messy real world integration of artificial intelligence advanced hardware and digital front doors into actual hospital work flows.

00:01:17: which exactly our mission for this deep dive?

00:01:20: yeah!

00:01:20: To understand that it's truly working on ground.

00:01:22: now we've sifted through the most critical insights shared by health tech leaders across LinkedIn during calendar weeks, eighteen and nineteen.

00:01:32: So let's start with a fundamental shift in how artificial intelligence actually operates right now because it sort of sets the stage for everything else.

00:01:40: Absolutely!

00:01:41: For long time AI in healthcare was basically just really smart search engine you know?

00:01:44: You asked that question gave an answer then did all the heavy lifting.

00:01:47: Yeah The human was motor

00:01:49: Right.

00:01:50: But Based on observations from Wolfgang Schleifer and Dr.

00:01:53: Faluppo Katamartiri, the industry is undergoing this massive shift towards something called agentic AI.

00:02:00: Yeah!

00:02:00: And Katomartiri frames this perfectly.

00:02:02: He points out that we're moving from AI as passive software to AI as an active digital workforce.

00:02:08: Active

00:02:09: work force

00:02:09: Right Because historically highly trained humans in healthcare have been forced to act as a biological middleware.

00:02:17: Wait biological middleware, that sounds exhausting.

00:02:20: It

00:02:20: is I mean think about a nurse or doctor manually copying data from an imaging system opening and electronic health record pasting that data And then you know moving the patient to a triage list.

00:02:31: Yeah it's just a massive waste of human intellect.

00:02:34: yeah.

00:02:35: so Agendic AI completely changes.

00:02:39: Well, these are independent digital agents that autonomously orchestrate these workflows in real time.

00:02:44: They create plans they use external software tools and even self-correct when the hit an error.

00:02:50: Oh!

00:02:50: That's fascinating And Yasir Khan shared a very concrete example of this in action right with a system called Agent Forge.

00:02:56: Yes He described was essentially multi agent AI office.

00:02:59: so it isn't just one massive AI trying to do everything Which

00:03:02: usually fails

00:03:03: Exactly.

00:03:04: It's a specialized team of digital workers, you have an agent designated as a patient safety auditor another is a privacy compliance agent and Another has a clinical risk assessor.

00:03:15: right?

00:03:16: And in the traditional setting if a clinical teams sees a pattern of patient deterioration But lacks the IT tools to track it they They just submit an IT ticket and wait like eighteen months on a backlog.

00:03:27: Yeah, and cons example shows how these autonomous agents Just bypass that entire bottleneck.

00:03:32: A clinician can describe what they need in plain English.

00:03:35: Just normal

00:03:36: text?

00:03:37: And the lead AI agent takes that request and delegates it, The coder region writes a software script.

00:03:43: then the privacy compliance agent reviews.

00:03:45: it realizes it violates data protocol kicks back for rewrite

00:03:50: Oh wow!

00:03:51: Without any human telling you to do

00:03:52: so Exactly.

00:03:54: Then the risk assessor verifies clinical logic.

00:03:56: They hand tasks off one another automatically To build fully functional compliant clinical app in

00:04:02: minutes instead of eighteen months.

00:04:04: Right because the core bottleneck and healthcare IT has never been a lack of clinical insight.

00:04:10: It's always been the slow, expensive translation between that insight and digital capability.

00:04:15: That

00:04:15: makes total sense.

00:04:16: And we're also seeing this level of integration reach the patient directly.

00:04:20: like Robert Sleppen highlighted a really fascinating deployment at Hartford HealthCare called Patient GPT.

00:04:25: Oh yeah Because most of us have interacted with a healthcare chatbot and frankly they usually leave you out dead end.

00:04:31: Yeah They tell ya to call nine one-one or just call your doctor

00:04:33: Right...they give generic advice.

00:04:35: But patient gpt is fundamentally different.

00:04:38: Because it uses an architecture called retrieval augmented generation,

00:04:43: which means what exactly?

00:04:44: Well

00:04:44: instead of just relying on the general knowledge he was trained on patient GPT has secure direct access to this specific patients electronic medical record.

00:04:55: oh wow yeah.

00:04:56: so when a patient asks a question The AI retrieves their actual lab values Their current medication list and they're clinical history before generating an answer.

00:05:06: So it's highly personalized.

00:05:08: Exactly, It anchors its responses in the patient's lived reality and crucially... ...it can seamlessly route to a human clinician with all that context already attached.

00:05:18: Okay so upgrading from smart calculator that waits for our input into fully autonomous medical assistant sounds incredible.

00:05:25: But I do have push back here because polished AI is incredibly dangerous if wrong.

00:05:32: And Sunday, Genoja posted a very strong warning about this exact transition.

00:05:37: He pointed out that AI output fluency does not equal reliability.

00:05:41: like an AI can generate a beautifully written, highly confident summary from a fragmented medical record.

00:05:47: Right!

00:05:48: But if the doctor reading that summary cannot trace the answer back to source documentation aren't we just creating false sense of confidence?

00:05:57: I mean how could hospital trust in Autonomous Agent verify its chain of

00:06:01: evidence?!

00:06:02: And

00:06:02: this is exactly what the deployment phase looks at right now as AI takes on autonomous clinical orchestration.

00:06:08: The danger isn't that robot replaces a doctor.

00:06:11: The imminent danger is that the AI operates with accountability gaps and hallucinations, because medical records are notoriously messy.

00:06:20: They contain conflicting notes from different specialists.

00:06:22: Oh constantly?

00:06:23: Yeah

00:06:24: so if an AI agent smooths over those conflicts to present a clean polished summary it might omit critical contradiction about drug allergy or previous diagnosis

00:06:33: And then the polished output becomes hidden clinical risk.

00:06:37: Exactly AI simply must be embedded into operations with strict traceability.

00:06:43: Every single claim it makes, must link directly to the specific line in The Physician's Note.

00:06:49: and that friction brings us to the most difficult hurdle of this entire digital transformation.

00:06:54: It actually isn't a technology problem at all

00:06:56: No!

00:06:56: Its the humans.

00:06:58: We have all these incredible autonomous technologies ready-to go which leads directly to biggest bottleneck today the clinicians and staff who actually have to use it.

00:07:14: Health

00:07:35: systems are incredibly rigid, highly regulated and financially strapped entities.

00:07:40: They simply cannot integrate train for and finance what already exists fast enough

00:07:45: Right so the biggest gains in health tech right now Are purely organizational

00:07:50: Exactly And Chrissy Hall offered some deeply practical advice For anyone trying to sell technology into these systems.

00:07:56: Yeah I saw that She pointed out Innovation sessions and pilots usually fail to ship actual products because the tech companies don't frame the solution in the currency of a decision maker.

00:08:06: Oh,

00:08:06: that makes so much sense!

00:08:08: Right... You can pitch the most brilliant sophisticated multi-agent AI workflow tool But if you cannot explain to the hospital CFO or front line nursing director exactly how it impacts their immediate costs, time and liability risk.

00:08:25: You don't have alignment!

00:08:30: And getting executives on board is really only half of battle.

00:08:33: you still have to convince the front-line clinicians to change how they work.

00:08:36: Which

00:08:37: is the hardest part?

00:08:37: It is, and Meenal Shah shared a deeply insightful framework for understanding this resistance.

00:08:44: she used a neuroscience concept called The Scar Off Model which stands for status certainty autonomy relatedness and fairness.

00:08:52: When a hospital deploys a new piece of technology to thousands of doctors and nurses the implementation teams often treat it as you know, A pure software training issue like

00:09:01: UI problem

00:09:02: exactly.

00:09:03: But the barriers are rarely about the softwares user interface.

00:09:07: They're rooted in biological fear.

00:09:09: Yeah Put yourself in the shoes of a seasoned physician for a second.

00:09:13: You have spent twenty years Mastering your craft building a mental model Of how to treat patients safely right?

00:09:20: And suddenly A hospital administrator hands you a new tablet and says we are doing everything differently starting Monday.

00:09:28: Just use this.

00:09:29: Oh, it feels incredibly unfair And it threatens your autonomy because you weren't consulted.

00:09:33: You were commanded

00:09:34: exactly?

00:09:36: It destroys your certainty Because you don't fully understand the new system which means you might make a mistake and harm someone.

00:09:42: And Shaw points out that when clinicians naturally push back against that sudden vulnerability, implementation teams usually respond by just throwing more PowerPoint training decks at them.

00:09:51: Which

00:09:52: doesn't work?

00:09:52: No!

00:09:53: Because you cannot use logic to override an emotional threat response.

00:09:58: A perceived threat to a clinician's status or autonomy triggers the amygdala in their brain.

00:10:03: It initiates a fight-or-flight response.

00:10:05: Exactly So.

00:10:06: labeling this as stubbornness Or resistance to change Completely misses the point.

00:10:11: It is a biological threat response to a sudden loss of control,

00:10:14: right?

00:10:15: And if software designers and hospital leaders do not actively design technology rollouts for human trust Psychological safety an actual workflow integration The best AI in the world will literally just gather dust.

00:10:27: so true.

00:10:28: So on the software side Human friction is a massive complicated barrier.

00:10:34: But this is where the dynamic gets super fascinating to me.

00:10:37: how so

00:10:38: Well, while software adoption requires this delicate change management we're seeing that when you combine AI with advanced hardware specifically in imaging and diagnostics the clinical value becomes so undeniable that hospitals simply cannot afford to ignore it.

00:10:53: Hardware seems to bypass a lot of that human friction.

00:10:56: It really does, because in imaging the technology directly and immediately alters the clinical pathway.

00:11:02: it provides instant proof-of value to the physician.

00:11:04: Dr Daniel Stromer & Anja Dyke shared remarkable trauma case perfectly illustrates this difference.

00:11:10: A patient was brought into the emergency department after severe traffic accident.

00:11:15: The initial conventional CT scan showed no displaced fracture lines.

00:11:21: Under normal circumstances, that patient might be cleared or moved.

00:11:24: Which could be catastrophic if there was a hidden fracture?

00:11:26: Exactly!

00:11:27: But the trauma team used single source Photon Counting CT Or PCCT And That scan revealed A Hidden Non-Displaced Highly Unstable Fracture At The Base of C-II Dens in Neck

00:11:39: Wait...a fracture at the C-ii vertebra?

00:11:41: That is incredibly dangerous Extremely If a patient has moved with an unrecognized unstable neck fracture.

00:11:47: It Could lead to severe paralysis or even be fatal and the conventional CT completely missed

00:11:52: it.

00:11:52: Completely missed it!

00:11:53: Well,

00:11:54: how does a photon counting CT see something that irregular machine is just blind to?

00:11:59: It comes down to fundamentally changing how we capture X-rays.

00:12:02: In traditional CT scanner The x-ray hits detector gets converted into visible light And then sensor turns this light in an electrical signal.

00:12:11: That middle step converting to light scatters the signal and creates noise which limits small image pixels can be.

00:12:18: A photon counting CT completely eliminates that middle step.

00:12:21: Oh, I see!

00:12:22: It uses specialized semiconductor materials to count individual x-ray photons and convert them directly into an electrical signal.

00:12:29: This direct conversion allows for massively smaller pixel sizes resulting in a monumental jump in spatial resolution

00:12:36: That makes a ton of sense.

00:12:37: Yeah, it decreases visual artifacts increases x-ray efficiency and can even detect subtle bone marrow edema without increasing the radiation dose to.

00:12:48: If a photon counting CT can see life-threatening fracture that regular CT completely misses, doesn't it instantly alter the standard of care?

00:12:57: Like if I'm patient in car crash and want scanner to see broken neck.

00:13:01: You absolutely do!

00:13:02: How long until this technology is just baseline expectation for every single trauma

00:13:06: center?!

00:13:07: Well, hospital systems are wrestling with that exact financial and clinical calculation right now.

00:13:12: We're moving from fragmented traditional image acquisition to integrated high-resolution precision... Right!

00:13:18: ...and we see this automation pushing across other modalities as well.

00:13:21: Puneet Sharma noted the industry pushed toward one click cardiac MRI using foundation models.

00:13:27: Okay let's unpack that for a second.

00:13:29: What exactly is a foundation model in the context of an MRI machine?

00:13:33: And how does it make it one click?

00:13:35: So a Foundation Model is a massive AI architecture trained on vast amounts of unlabeled data.

00:13:42: In this case, millions of anatomical scans.

00:13:46: so it fundamentally understands human geometry.

00:13:49: okay normally an MRI technologist spends fifteen to twenty minutes manually finding the exact correct imaging planes and slices for a beating human heart which was highly complex.

00:14:01: But by embedding a foundation model directly into the scanner, The AI instantly recognizes heart spatial orientation.

00:14:07: It automates positioning and alignment in one click.

00:14:10: Oh wow!

00:14:11: It drastically reduces scan time and guarantees consistency regardless of technology's experience level.

00:14:17: Consistency is huge right?

00:14:18: Especially when we talk about patient equity.

00:14:20: Julie Hamilton raised an important point regarding breast cancer screening.

00:14:23: Yes Because for women with dense breasts tissue traditional mammography isn't enough.

00:14:29: It's not.

00:14:30: Traditional mammograms rely on x-rays and dense breast tissue shows up white, but unfortunately tumors also show up white.

00:14:39: Radiologists often compare it to trying to find a snowball in a blizzard.

00:14:43: the tissue masks cancer.

00:14:46: so Hamilton advocates for widespread adoption of automated breasts ultrasound or ABUS integrated with AI.

00:14:53: Because ultrasound uses sound waves instead of x-rays, it can easily see through dense tissue to find hidden lesions.

00:14:59: That is fantastic!

00:15:01: Yeah

00:15:01: and by automating the ultrasounds sweep using AI to flag abnormalities hospitals can provide scalable consistent screening without exposing the patient extra radiation.

00:15:12: It ensures an equitable care pathway giving women with dense breaths the same diagnostic confidence as anyone else.

00:15:17: it is just incredible to see how AI Is physically supercharging hospital hardware, but you know here's The reality check.

00:15:24: okay we can have the most advanced photon counting CTs and AI driven ultrasounds sitting inside the Hospital But none of that matters if the patient's entry point into the health system is jammed absolutely

00:15:34: true.

00:15:35: yeah.

00:15:35: If we want to see where the system is bleeding time, We really have to look at the digital front door.

00:15:40: Which brings us out of hospital and into primary care.

00:15:43: Yeah, Primary Care is funnel for entire healthcare systems But as Anna Karen Edstedt Bonomy and Stanislaus Knack-Chateau pointed out recently General practitioners are absolutely exhaust.

00:15:55: Oh

00:15:55: completely burned out.

00:15:56: They're drowning in administrative workloads And fighting with incredibly fragmented outdated systems.

00:16:03: Dr.

00:16:03: Lib which is a major European health tech company, just announced a one hundred million pound investment to launch in the UK.

00:16:11: and they're joining forces with Medicis Health to tackle this exact issue.

00:16:15: And, this is the part that I found genuinely shocking.

00:16:17: Medicis built the first new GP clinical software for The National Health Service in twenty-five years...

00:16:23: That's wild!

00:16:24: We literally have artificial intelligence designing complex protein molecules in weeks but our primary care doctors —the people who actually decide if you need go to hospital— are fighting with software infrastructure build in late nineteen nineties.

00:16:37: It really highlights a severe systemic mismatch in healthcare investment.

00:16:41: We historically overindex on flashy hospital tech and chronically underfund primary care infrastructure, right?

00:16:48: Fixing the digital front door isn't just about creating a sleeker appointment booking app for patients.

00:16:53: it is about fundamentally reducing the crushing administrative fatigue on primary care workers.

00:17:00: Yeah When you upgrade that core clinical software to automate documentation, referrals and lab tracking You return a GP's most precious resource time with the patient.

00:17:10: And we are seeing proof that hospitals can modernize these front doors incredibly fast when their right incentives or penalties is in place.

00:17:18: Annika Kohl shared case study about the Celitinin hospital network.

00:17:22: They successfully rolled out a comprehensive digital patient portal across ten hospitals and eighty departments in just one single year.

00:17:30: That's

00:17:30: incredibly fast!

00:17:31: It is, an major driver for that speed was meeting the KHZG deadlines.

00:17:36: Yeah The KHZ G or the German Hospital Future Act Is perfect example of policy forcing adoption.

00:17:43: It provided billions in funding for hospital digitalization, but it also came with a massive stick.

00:17:49: Exactly!

00:17:50: Hospitals that fail to implement mandatory digital services like patient portals and digital medication management face significant financial penalties up to two percent of the revenue.

00:18:00: When hospital revenue is directly threatened, that eighteen-month IT back law we talked about earlier suddenly gets cleared up.

00:18:07: Funny how it works.

00:18:08: and we are also seeing this digital front door concept expand to a national scale.

00:18:12: Piotr Erzikowski discussed deploying AI powered triage tools for government health channels specifically Health Direct Australia which serves the population of twenty six million people.

00:18:23: right I mean How do you safely triage an entire country with AI?

00:18:27: by moving toward a hybrid pathway, self-direct and deploying conversational AI voice agents capable of handling initial symptom checking.

00:18:34: Instead of patients sitting on hold for hours during a flu surge the AI agent can safely navigate them to their right care setting Whether that is booking a telehealth appointment, visiting a pharmacy or heading to the emergency room.

00:18:47: Well that makes sense!

00:18:48: By automating that initial triage at scale.

00:18:51: it takes the immense pressure off those initial phone lines and reserves the human nurses for complex high-risk cases.

00:18:58: Okay so we've looked at autonomous AI agents The psychology of human adoption Life saving imaging hardware And overhauling primary care front door.

00:19:07: We

00:19:07: have covered alot

00:19:09: To wrap this up, let's look all the way upstream at how medical treatments are created in

00:19:36: which is a vertical AI built specifically for complex biological pathway reasoning.

00:19:41: You have AWS Biodiscovery, an infrastructure platform that recently helped Memorial Sloan Kettering generate three hundred thousand antibody candidates in a matter of weeks instead of taking year and you have Anthrochus Clawed For Life Sciences Which connects directly to scientific databases to synthesize research.

00:20:02: Okay, narrowing down three hundred thousand antibody candidates in weeks is staggering.

00:20:07: But the application that really blew my mind was shared by Maria Luisa Sanchez.

00:20:11: Oh yeah A research team at Georgia Tech took a standard e-commerce algorithm and applied it to chemotherapy regimens for children.

00:20:18: I need you to explain this because taking the tech that sells me phone accessories And using it for pediatric leukemia sounds wild.

00:20:25: How does that actually work?

00:20:26: Well, the underlying mechanism is a concept called collaborative filtering.

00:20:30: In e-commerce The algorithm looks at a basket of goods if it sees that thousands of people who buy specific smartphone also tend to by a specific charger.

00:20:38: It learns that link and recommends it to the next buyer.

00:20:41: right customers Who bought x?

00:20:43: Also thought why

00:20:44: exactly.

00:20:45: Georgia Tech realized that a patient's biological data is essentially just a complex basket of goods.

00:20:51: If patients who present with symptom A and biomarker B during chemotherapy almost always develop infection C, the algorithm flags that hidden pattern.

00:21:00: Oh

00:21:00: wow!

00:21:01: They took consumer retail math And applied it to cellular biology To predict infection risk in children treated for leukemia.

00:21:10: and they achieve seventy-nine percent accuracy on very small patient groups, which is notoriously difficult for AI models because usually require massive data sets to find patterns.

00:21:20: That translation of technology between completely different sectors it's just brilliant!

00:21:24: And we are seeing a similar leveling up effect physically inside the operating room.

00:21:28: Michael Orman highlighted recent findings on robotic surgery showing that robotic assistance is allowing low volume surgeons, those who don't perform a specific procedure very often to achieve clinical outcomes comparable to high-volume seasoned non-robotic surgeons.

00:21:44: Yeah

00:21:44: this is the democratization of excellence.

00:21:46: yeah for decades surgical outcomes were heavily dependent on the specific hands and experience of the individual surgeon.

00:21:52: right robotic systems provide advanced and they allow for hyperprecise micro movements, whether it's an e-commerce algorithm finding a hidden infection pattern in rare pediatric disease or robotic arm raising the baseline skill level of newer surgeon to match veteran.

00:22:09: Technology is actively standardizing high quality care.

00:22:13: It reduces human variability.

00:22:14: that has historically plagued

00:22:16: medicine.".

00:22:17: We have moved entirely out of the hype of what AI might do.

00:22:24: we are solidly in a deployment phase where AI is reading photon counting CTs, restructuring primary care and designing life-saving

00:22:32: antibodies.".

00:22:33: But

00:22:34: you know as we integrate these autonomous AI agents deeper into clinical care it leaves us with profound regulatory blind spot that everyone listening needs to think about.

00:22:44: What's that?

00:22:45: Caroline Bradner Jasek posed a lingering question... and they start performing complex clinical work autonomously without a human double-checking every single click.

00:22:56: who ultimately gets to license that AI, to practice medicine.

00:23:00: Oh wow!

00:23:01: Is it the state medical board?

00:23:02: is at the FDA through their software regulations or is in hospitals internal credentialing committee?

00:23:09: We

00:23:09: literally have figure out how give digital worker of medical license.

00:23:13: That's massive regulatory gray area.

00:23:15: an incredible thought.

00:23:16: leave on If you enjoyed this episode, new episodes drop every two weeks.

00:23:20: Also check out our other editions on ICT and Tech Insights, DefenseTech Cloud Digital Products & Services Artificial Intelligence And Sustainability And Green ICT.

00:23:29: Thank You so much for joining us On This Deep Dive.

00:23:31: Don't forget to hit subscribe!