Hi HAIFers,
Today is a big milestone. My personal LinkedIn account has officially reached 13,000 followers and, surprisingly, our HAIF LinkedIn page just hit 1,000 followers on the same day. I’m genuinely grateful to have you with me on this journey.
I write this breakdown for you, so don’t hesitate to share your feedback!
Here’s a little teaser for what’s coming next. This week, I’ll be doing a deep dive on a platform used by 80% of doctors in the U.S. Can you guess which one it is?
Your shortcut for this week:
Read on for the foresight of the week and the full breakdown of what shaped healthcare AI 👇
Signal of the week
The Hidden Risk Of LLM’s Revealed By The New NOHARM Benchmark
A new benchmark, NOHARM, shows that even top LLMs that outperform generalist physicians still produce severe harm in 22.2% of cases. Most of the harm isn’t hallucination but omission: 76.6% of errors are missing essential actions. The study also confirms that safety and knowledge aren’t the same thing, with only a ~0.6 correlation to standard benchmarks. [Link]
Foresight:
The “single-model” approach won’t hold. When models reviewed each other’s output, harm fell by 8%, hinting at a shift toward multi-agent checks as a default.
Solutions and Launches
OpenEvidence (US): added a HIPAA-secure Dialer to its AI-CDS platform, letting clinicians call patients with hospital caller ID and optionally convert the call into an auto-generated clinical note. [Link]
Aidoc (US): expanded its platform with QP-Prostate for prostate MRI analysis and launched a MONAI integration that lets hospitals deploy their own in-house imaging AI models instead of relying only on closed systems. [Link]
DeepHealth (US): Unveiled new AI-powered imaging informatics and SmartTechnology solutions at RSNA 2024. The tools support disease detection, remote scanning, and workflow optimisation. [Link]
Lexacom (UK): AI Scribe company added free patient messaging (SMS/email) and expanded ambient AI capabilities including call summarisation. [Link]
CogStack (UK): spun out from King’s College London and NHS partners to scale its NLP platform commercially, already driving major operational gains across UK hospitals, from faster trial recruitment and missed-follow-up detection to improved coding accuracy and multi-million-pound savings. [Link]
NHS (UK): Rolled out AI tools to review stroke scans in real time, speeding diagnosis and treatment. This helps patients get timely care and improves recovery chances.[Link]
Governance, Policy, and Ethics
HealthEdge (US): Payers are increasingly adopting AI to cut costs and meet regulatory requirements, with 51% using AI to automate approvals in response to OBBBA mandates. [Link]
FDA (US): Launching the TEMPO pilot in 2026 with CMS’s ACCESS model to expand access to digital health devices for chronic care. Up to 10 manufacturers in each of four clinical areas will be selected. [Link]
University Of Liverpool (UK): AI deepfakes of real doctors, including Professor Taylor-Robinson, were used on social media to promote unproven supplements, highlighting the need for platforms to verify medical credentials.[Link]
AHA (US): called on the FDA to create clearer post-deployment standards for AI-enabled medical devices, citing gaps in monitoring drift, bias, and real-world performance; also urged risk-based oversight and stronger vendor accountability. [link]
Research and AI Advancements
The Lancet Digital Health: Published results from the Brainomix study showing AI-assisted stroke care doubled thrombectomy rates (2.3% to 4.6%) across NHS sites. [Link]
Google Cloud (US): released PubMed as a BigQuery public dataset with Vertex AI vector search, enabling semantic literature search and turning PubMed into an AI-ready resource for researchers and clinicians. [Link]
NOHARM study (US): introduced a new benchmark evaluating LLM clinical safety across 100 real consultation cases; found severe harm in up to 22% of model-generated recommendations - mostly omissions. Amboss’s LiSA ranked first in terms of safety.[Link]
UK GPs (UK): AI for Burnout, Not Access A new study finds 28% of UK GPs use AI, primarily to reduce burnout rather than increase patient volume. The report also warns of a "digital divide," with significantly lower adoption in deprived areas. [Link]
Partnerships & Adoption
Avo & Marathon Health (US): Marathon Health will integrate Avo’s AI-CDS and scribe suite across its 1,600+ providers, embedding into athenahealth. [Link]
GE HealthCare (US): Launched GEMINI-RT with Mayo Clinic to use imaging, AI, and monitoring to personalize radiation therapy. It includes work in predictive oncology and aims to identify side effects earlier. [Link]
IQVIA & AWS (US): Expanded their collaboration to deploy "Agentic AI" across clinical trial workflows. The agents will automate site selection and patient matching to accelerate trial timelines. [Link]
UCHealth (US): Partnered with Doximity to deploy DoxGPT and Scribe across hospitals and outpatient centres. [Link]
Bets on the Next Health System
Investments:
Radical Health (US): Backed by Khosla Ventures, the startup emerged from stealth with $5M to build an “AI oncologist.” Its platform uses data from 11M patients to personalize cancer care. [Link]
Sunflower (US): Raised a total of $6.5M, reaching a $60M valuation, to expand its sobriety‑focused AI platform. [Link]
Aledade (US): Secured a $500M credit facility from Ares to support growth. Supports 20,000+ clinicians across 2,400 practices, achieving $1B in Medicare savings in 2024 with 93% of ACOs realizing shared savings. [Link]
SciNeuro (US): Raised $53M to advance neurodegeneration therapies, backed by LAV and ARCH Venture Partners. [Link]
AITRICS (South Korea): Raised $24M Series C to enhance its VitalCare patient deterioration AI and develop its clinician copilot, V.Doc Pro. The company also plans expansion into Vietnam and Hong Kong. [Link]
M&A:
Lightbeam Health (US): acquired Syntax Health, adding its VBC contract-modeling and incentive-design platform into Lightbeam’s AI-enabled population-health OS. [Link]
That’s a wrap for Edition #9 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.