AI in Healthcare and Digital Health Update: AI Eye Screening, Apple Watch Hypertension, Smart Inhalers, Mammogram CVD Risk and More
From Norway’s first national rollout of autonomous AI eye screening and FDA-cleared Apple Watch hypertension notifications to Neuralink’s planned brain “speech” trial—this week’s signals are robust. We span diagnostics, digital therapeutics, public health, and consumer tech, with policy shifts and clinical advances across areas, underscoring real-world implementation challenges and the expanding integration of AI into everyday care.
In Today’s Newsletter
📺 Anemia info on YouTube: mixed quality [1] [18 Sep 2025]
https://journals.sagepub.com/doi/10.1177/20552076251380650
Context: 50 English videos (Mar 2024) scored via DISCERN, GQS, JAMA; VPI for popularity.
Key point: Overall quality moderate; healthcare-posted videos were more reliable/higher quality; popularity weakly tied to quality.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
🧠 Vanderbilt’s AI to curb obesity care drop-off [2] [US • 17 Sep 2025]
Context: $1M Eli Lilly grant; EHR analysis + surveys → multi-agent AI; pilot patient app in VUMC clinics.
Key point: Targets continuity gaps and delayed access; human-AI panels to prioritize interventions.
Implication: Signals pipeline investment and modality expansion.
🔗 APEC Seoul: AI + GLP-1 digital convergence drug (Xperix/Hanmi) [3] [KR • 16 Sep 2025]
https://www.mk.co.kr/en/stock/11420797
Context: Xperix showcased AI-DTx platform; combo with Hanmi’s efpeglenatide under Korea’s Digital Medical Products Act.
Key point: Seeks to optimize effectiveness and adherence by pairing GLP-1 with AI-driven behavioral programs.
Implication: Signals pipeline investment and modality expansion.
🧑⚕️ Survey: attitudes on AI triage in mental health [4] [US • 19 Sep 2025]
Context: Iris Telehealth national survey.
Key point: 50% open to AI-enabled triage, but 73% require human final decisions; concerns: false positives, loss of human connection.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
🍽️ Amazon adds insurance-covered dietitians via Fay [5] [US • 16 Sep 2025]
Context: Fay becomes first nutrition care partner on Health Benefits Connector; AI matches users to RDs; 3,600+ RDs network.
Key point: Extends Amazon’s digital health into nutrition services aligned with employer benefits.
Implication: Could streamline initiation and adherence via remote prescribing and logistics.
🧬 Abu Dhabi’s AI MRI tool speeds MS diagnosis [6] [UAE • 18 Sep 2025]
Context: NYUAD + Yas Clinic + ADSCC; FLAIRstar detects central vein sign per McDonald 2024; tested in LAMINATE.
Key point: Automates CVS detection to reduce uncertainty and time to treatment; data remain in UAE.
Implication: May influence prescriber choice and payer reviews pending full data.
🩻 LLMs improve on Japanese radiology board exam [7] [JP • 15 Sep 2025]
Context: 233 questions/477 images; models tested with/without images; legitimacy rated by radiologists.
Key point: OpenAI o3 hit 67% (text) and 72% with images; image input significantly boosted select models.
Implication: May influence prescriber choice and payer reviews pending full data.
💊 Berry Street launches GLP-1 Nutrition Pathway [8] [15 Sep 2025]
Context: Three 12-week phases (explore/optimize/sustain) led by RDs/CNS; EHR integration and payer-aligned workflows.
Key point: Program aims to improve adherence, safety, cost-efficiency across the GLP-1 journey.
Implication: Could streamline initiation and adherence via remote prescribing and logistics.
🧪 Singapore trials ‘AI-free’ periods to prevent deskilling [9] [Singapore • 15 Sep 2025]
Context: NUHS rolling out AI-off time; NHG/SingHealth exploring usage limits and performance tracking.
Key point: Responds to evidence of skill erosion with constant AI assist (e.g., colon polyp detection).
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
🌬️ Remote lung monitoring via camera biomarkers (electronRx) [10] [18 Sep 2025]
Context: pDx app uses device cameras to derive cardiopulmonary digital biomarkers; clinician dashboards in real time.
Key point: Targets asthma/COPD/ILD/PH; aims to reduce clinic burden and personalize therapy.
Implication: Could streamline initiation and adherence via remote prescribing and logistics.
🏥 NHS AI rollout: slower and harder than expected [11] [UK • 15 Sep 2025]
https://www.digitalhealth.net/2025/09/ucl-study-finds-challenges-in-implementing-ai-in-the-nhs/
Context: Lancet eClinicalMedicine evaluation of £21M program across 66 trusts (chest diagnosis AI).
Key point: Contracting 4–10 months late; by Jun 2025 23/66 not live; governance, IT, and workload barriers.
Implication: May influence prescriber choice and payer reviews pending full data.
🌬️ iCARE RWE: smart inhalers boost adherence/utilization [12] [US • 15 Sep 2025]
https://sg.finance.yahoo.com/news/data-icare-study-hailie-smartinhaler-040500453.html
Context: Adherium + Intermountain + CareCentra interim results (~850 pts subset; 2-year program up to 2,500).
Key point: Two-week baseline adherence ~62%, ~40% ≥80%; signals fewer admissions/readmissions (preliminary).
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
🩺 Mammograms to predict CVD risk [13] [AU • 18 Sep 2025]
https://www.mobihealthnews.com/news/anz/using-mammograms-predict-stroke-heart-attacks
Context: George Institute + UNSW/Sydney; DL trained/validated on 49k mammograms; features include arterial calcification.
Key point: Comparable accuracy to standard calculators without extensive clinical data; dual-purpose screening potential.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
🔋 Wearable hyperkalemia detection: pilot begins (AK+ Guard) [14] [US • 16 Sep 2025]
Context: AccurKardia AK+ Guard (FDA Breakthrough) multicenter pilot across inpatient/outpatient/home.
Key point: Prior 20,750-pt retrospective showed ~89% accuracy for moderate-to-severe hyperkalemia via Lead I ECG on wearables.
Implication: May influence prescriber choice and payer reviews pending full data.
🧪 Lunit highlights cancer AI at APEC/WBS [15] [KR • 16 Sep 2025]
Context: Dedicated APEC session; portfolio demos; notes 7,000+ sites and reimbursement progress.
Key point: Advocates multimodal AI and APEC-level collaboration (data infra, training).
Implication: Signals pipeline investment and modality expansion.
👁️ Norway rolls out autonomous AI eye screening [16] [Norway • 17 Sep 2025]
Context: EyeArt® deployed in South-Eastern Norway; plan to scale nationally; integrates with DIPS Arena EMR.
Key point: Seeks >95% screening coverage for people with diabetes; addresses specialist shortages.
Implication: Introduces competition that may affect pricing and formulary access.
🧫 HealthTree expands AI trial-matching to all blood cancers [17] [17 Sep 2025]
Context: Enhanced tool matches based on patients’ medical records; nonprofit focus.
Key point: Aims to reduce enrollment barriers with personalized recommendations.
Implication: May expand screening, initiation, and follow-up at scale.
👁️ Eye scans as early signal for blood cancers? (proof-of-principle) [18] [UK • 14 Sep 2025]
Context: European Journal of Cancer study; AI analyzed retinal images (~1,300 UK pts).
Key point: Microvascular retinal changes associated with higher future myeloma/leukemia risk; early research.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
🗣️ Ambient scribe tech in mental health: risks & guardrails [19] [UK • 16 Sep 2025]
https://www.digitalhealth.net/2025/09/the-challenges-of-ambient-scribe-technology-for-mental-health/
Context: Accurx opinion/report for NHS mental health settings.
Key point: Warns of trust/privacy issues, editing fatigue; advocates human-in-loop and authentic consent.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
🚶♀️ Phone video + privacy-preserving AI classifies gait [20] [16 Sep 2025]
https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0001004
Context: PLOS Digital Health; 743 clinician-simulated clips; pose-based features.
Key point: ~86.5% accuracy combining frontal+sagittal views; lower-limb keypoints and frequency/entropy features dominate.
Implication: May influence prescriber choice and payer reviews pending full data.
🧾 Open vs closed LLMs for clinical info extraction (UC) [21] [15 Sep 2025]
https://bmjdigitalhealth.bmj.com/content/1/1/e000014
Context: BMJ Digital Health; extracting Mayo Endoscopic Subscore from colonoscopy notes.
Key point: QLoRA fine-tuning boosts open-source models, but GPT-4o with prompts still best on accuracy/cost.
Implication: May influence prescriber choice and payer reviews pending full data.
❤️ NEJM AI: predicting patient CPR preferences (PPP) [22] [18 Sep 2025]
https://ai.nejm.org/doi/full/10.1056/AIoa2500265
Context: Swiss SHARE data (n=1811); ML models vs partners predicting CPR preferences.
Key point: Models outperformed partners (up to ~70.6% CV accuracy); integration raises ethical questions.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
⌚ FDA clears Apple Watch hypertension notifications [23] [US • 15 Sep 2025]
Context: ML-based Hypertension Notification Feature uses optical signals over ~30 days; supported devices across recent Watch lines.
Key point: Not diagnostic; prompts users to seek cuff measurement and clinician follow-up.
Implication: Could streamline initiation and adherence via remote prescribing and logistics.
🧠 Neuralink plans brain “speech” trial (IDE) [24] [US • 19 Sep 2025]
Context: US trial aims to translate imagined speech to text for severe impairments; adds to ongoing studies.
Key point: Seeks faster communication; broader consumer visions remain speculative.
Implication: Signals pipeline investment and modality expansion.
🔮 Delphi-2M forecasts 1,000+ diseases from records/lifestyle [25] [UK • 17 Sep 2025]
https://www.nature.com/articles/d41586-025-02993-x
Context: Nature news; modified GPT trained on UK Biobank (400k) to estimate multi-disease risks up to 20 years.
Key point: Multi-disease performance matched/exceeded many single-disease tools; generalizability pending.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
🧩 Perspective: toward a substrate-agnostic consciousness theory [26] [16 Sep 2025]
https://dailyneuron.com/general-theory-of-consciousness-ai/
Context: Summary of academic proposal (ITI + Moment-to-Moment) for informational individuality + continuous updating.
Key point: Frames AI as an experimental partner to test general theories of consciousness.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.
Why it matters
- Public-sector precedent: Norway’s EyeArt program is the first national adoption of autonomous AI screening, setting procurement/integration benchmarks [16].
- Consumer devices formalize screening prompts: FDA-cleared Apple Watch alerts embed pre-diagnostic nudges into daily life [23].
- Implementation beats hype: NHS experience underscores that contracts, IT fit, and capacity drive timelines more than model accuracy [11].
- Dual-use imaging: DL can add CVD risk to routine mammography, improving women’s preventive care without extra visits [13].
- Adherence at scale: Early smart-inhaler data and GLP-1 nutrition pathways highlight behavioral + digital combos to unlock outcomes [12][8].
📢 Stay Ahead in AI in Healthcare and Digital Health Updates!
✅ Contact LucidQuest at info@lqventures.com for strategic guidance on AI in Healthcare and Digital Health breakthroughs.
FAQ
Does EyeArt replace ophthalmologists in Norway?
No. It autonomously grades images and writes to the EMR, but clinicians still manage pathways. It’s CE-marked for DR/AMD/glaucoma; FDA-cleared for DR [16].
Is Apple Watch now measuring blood pressure?
No. The feature analyzes optical signals to notify possible hypertension risk and recommend cuff confirmation. It’s not a diagnostic or BP reading [23].
What slowed NHS AI deployments?
Per UCL, contracting delays, governance approvals, varied/aging IT across trusts, and limited staff capacity. One-third of sites weren’t live by Jun 2025 [11].
How strong is the mammogram-CVD evidence?
Model trained on 49k mammograms showed comparable accuracy to standard CVD tools. It suggests feasible dual screening; further validation planned [13].
Do smart inhalers meaningfully change outcomes?
Interim iCARE RWE shows higher adherence (~62% baseline; ~40% ≥80%) and signals fewer admissions/readmissions. Data are preliminary and need fuller analysis [12].
Can LLMs handle imaging questions now?
In Japan’s radiology board set, adding images raised certain models (e.g., OpenAI o3) to ~72%; legitimacy ratings improved over older baselines [7].
Entities / Keywords
Eyenuk EyeArt; Apple Watch Hypertension Notification Feature; Vanderbilt University Medical Center; Eli Lilly; NUHS/NHG/SingHealth; George Institute for Global Health; AccurKardia AK+ Guard; electronRx pDx/purpleDx; HealthTree Foundation; Berry Street GLP-1 Nutrition Pathway; Xperix/Hanmi efpeglenatide + AI-DTx; YouTube anemia (DISCERN/GQS/JAMA); UCL/Lancet eClinicalMedicine NHS study; OpenAI o3 / Gemini 2.5 Pro radiology; Neuralink BCI; Delphi-2M (UK Biobank).
References
https://journals.sagepub.com/doi/10.1177/20552076251380650
https://www.mk.co.kr/en/stock/11420797
https://www.digitalhealth.net/2025/09/ucl-study-finds-challenges-in-implementing-ai-in-the-nhs/
https://sg.finance.yahoo.com/news/data-icare-study-hailie-smartinhaler-040500453.html
https://www.mobihealthnews.com/news/anz/using-mammograms-predict-stroke-heart-attacks
https://www.digitalhealth.net/2025/09/the-challenges-of-ambient-scribe-technology-for-mental-health/
https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0001004
https://bmjdigitalhealth.bmj.com/content/1/1/e000014
https://ai.nejm.org/doi/full/10.1056/AIoa2500265