Hey HAIFer, welcome back!
This was a quiet week for the US (Thanksgiving will do that). It’s one of those rare weeks where the global AI-healthcare scene picked up the pace while the American side took a breather.
On my side, pulling this together each week takes more work than it looks, even with AI helping along the way. But the feedback over the last two months has been heart-warming: subscribers are using it and finding it genuinely useful. So it’s not slowing down - and I’ll be adding deeper dives into point solutions alongside the weekly roundups soon!
If you’d like to support the work and sponsor an issue - or be featured in one of the deep dives - you can reach me here.
Let’s dive in 👇
If you have 1 minute - here’s the summary
NEJM AI: New trials confirm AI scribes reduce burnout, but meaningful productivity gains require models that access patient history to close the "context gap."
MHRA & NICE (UK): Secured £2M to build a regulatory "sandbox" ensuring digital mental health tools are clinically validated.
Treasury Board (Canada): Published the country's first public AI Register, detailing 400+ systems to mandate transparency.
Singapore Health (Singapore): Achieved 95% sensitivity in flagging rare diseases via automated EHR scanning of 1.27M patients.
Corsano Health (EU): Partnered with Netcare to rollout medical-grade wearables across 6,000 beds nationwide in South Africa.
Doximity (US): Contracted with Scripps Health to deploy GPT-based drafting tools system-wide to reduce administrative burden.
Gosta Labs (Finland): Raised €7.5M to build an "AI operating system" for oncology, moving away from generic LLMs.
BMJ Digital Health: Researchers developed "MedMobile," a 3.8B parameter model that passes MedQA on-device without internet.
Aramco Ventures (Saudi Arabia): Opened a Paris office to deploy capital from its $7 billion fund into European deeptech and Healthcare AI.
If you have more time, read on for the foresight of the week and the full breakdown of what shaped healthcare AI last week👇
Signal of the week
AI Scribing is not enough - studies confirm
Three major papers published last week exposed a critical ceiling for current AI scribes. While two RCT’s confirmed that AI consistently reduces burnout, productivity gains were shockingly variable - ranging from ~22 minutes saved daily (UW Health) to <1 minute per visit (UCLA). The deciding factor wasn't speech recognition, but mainly context: one study showed that "ambient listening" alone achieved a note completeness score of only 40.4, but injecting patient history (labs, past notes) into the model nearly doubled that score to 82.9. [All three papers are cited below]
Foresight:
The "Transcription Era" is ending. Data proves that optimizing for hearing hits a quality ceiling; reading the chart is what actually doubles clinical value. The future belongs to deep integrators, not just good listeners. We are shifting from passive stenographers to context-aware agents that "know" the patient before the visit starts.
Solutions and Launches
Neuraxpharm (EU): Debuted EPISERAS, a wearable device analyzing brain signals to provide advance seizure warnings for epilepsy patients. [Link]
HelloBetter (Germany): Launched Ello, a prescription-grade AI mental health companion with human supervision for crisis management. [Link]
Samsung Medison (South Korea): Released the R20 ultrasound with an AI engine that automates measurements to reduce sonographer fatigue. [Link]
Corsano Health (EU): has partnered with Netcare (a major South African private-hospital group) to roll out medical-grade wearable devices in all general wards nationwide, covering around 6,000 beds.[Link]
Elegant Hoopoe (UAE/UK): Expanded its AI-governed wellness franchise to the UK, using an autonomous engine for compliance and operations. [Link]
Governance, Policy, and Ethics
MHRA & NICE (UK): Secured £2 million from Wellcome to fund a regulatory "sandbox" for digital mental health tools, ensuring new AI therapy apps are clinically validated before NHS rollout. [Link]
UK Regulators (UK): Issued warnings after a survey revealed 24% of patients now consult AI or social media for diagnosis before seeing a clinician. [Link]
Treasury Board (Canada): Published the country's first public AI Register, detailing 400+ systems to ensure compliance with the Artificial Intelligence and Data Act. [Link]
New Zealand Government: Funded an AI-enabled "digital front door" to unify and triage fragmented mental health services. [Link]
Research and AI Advancements
Two new RCT’s confirm that AI scribes consistently reduce clinician burnout, yet productivity gains varied wildly (from 7-22 minutes saved daily). [Link]
Editorial on two RCT’s suggests AI scribes significantly reduce work exhaustion but have no impact on professional fulfillment, suggesting meaningful productivity gains require next-gen agents that automate downstream tasks like orders and billing. [Link]
New data quantifies the "context gap" - giving AI scribes access to patient history (labs/past notes) doubled note completeness scores from 40.4 to 82.9/100 [Link]
A benchmark study showed Vision-Language Models (VLMs) fail significantly when medical images contain common artifacts like noise. This highlights how clinical AI must be trained on "messy" real-world data, not just pristine datasets. [Link]
Researchers developed "MedMobile," a 3.8B parameter model that passed MedQA on-device without internet. [Link]
A study of 1.27M patients in Singapore used EHR scanning to flag rare diseases with 95% sensitivity. [Link]
"GraphRAG" (graph-based retrieval) outperformed standard LLMs in adherence to kidney disease guidelines, revealing, for complex decision support, structural knowledge mapping beats simple text retrieval. [Link]
Partnerships, Collaborations, and Implementations
Doximity (US): Contracted with Scripps Health to deploy GPT-based drafting and scribing tools system-wide to reduce administrative burden. [Link]
RapidAI (US): Expanded its partnership with AWS to scale the Rapid Edge Cloud, a hybrid architecture that runs critical stroke AI locally for zero-latency triage while leveraging the cloud for model development. [Link]
LIVEMED (US): Joined forces with NeuroX to build a nationwide tele-neurology network integrating AI communications with hospital EMRs. [Link]
Avicenna.AI (France): Partnered with Ferrum Health to distribute its emergency radiology algorithms to US health systems via a unified platform. [Link]
Bets on the Next Health System
Funding:
Gosta Labs (Finland): Raised €7.5 million in a Seed round led by Voima Ventures. The funding will build an "AI operating system" specifically for oncology, moving away from generic LLMs to disease-specific models. [Link]
Morphle Labs (India/US): Raised $5 million in Series A funding led by Inflexor Ventures. The capital will scale their AI-powered robotic pathology scanners ("RoboTome") to automate cancer diagnostics in remote labs. [Link]
Aramco Ventures (Saudi Arabia/EU): Opened a Paris office to deploy capital from its $7 billion fund into European deeptech and health AI startups. [Link]
That’s a wrap for Edition #8 of Health AI Foresight.
My goal with this newsletter is simple - to connect the present, the emerging, and the future of healthcare AI.
See you next week.
- Dr. Aboufandi
📩 Something stand out? Or have a suggestion? Hit reply - I read every message.