AI in Healthcare and Digital Health Video Recap — October 20, 2025

LucidQuest

3 weeks ago

This biweekly video summary highlights global updates in AI in Healthcare and digital health, showing how AI continues to expand its role in shaping patient care and healthcare delivery worldwide.

Watch Our Video Summary Capturing Top AI and Digital Health News from the Last Two Weeks

📶 Wi-Fi beats: contactless pulse via ambient signals [1] [US • 05 Oct 2025]

https://spectrum.ieee.org/wi-fi-signal-heartbeat-detection

Context: UC Santa Cruz’s Pulse-Fi filters amplitude changes in Wi-Fi reflections; AI runs on Raspberry Pi/ESP32. Two experiments, n=7 stationary and >100 mixed-activity participants.
Key point: Reported <1.5 BPM error, robust up to ~10 ft and across postures; low-cost hardware (~$5–10 ESP32; ~$30 Pi). Multiuser pilots starting.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.

🧬 AML digital twins forecast neutropenia on Ven/Aza [2] [06 Oct 2025]

https://www.nature.com/articles/s41746-025-01978-4

Context: ODE-based models trained on VenEx trial longitudinal counts; individualized train-test showed error falls as cycles accrue.
Key point: Best model captured neutrophil trajectories with variable patient-specific accuracy; ~5 cycles needed for good predictions.
Implication: May influence prescriber choice and payer reviews pending full data.

🏪 Amazon meds at the point of care [3] [US • 08 Oct 2025]

https://www.theverge.com/news/796076/amazon-pharmacy-prescription-drug-vending-machine-kiosks

Context: Kiosks at One Medical clinics in Los Angeles starting Dec; antibiotics, inhalers, BP meds; no controlled or refrigerated drugs.
Key point: Prescriptions verified by Amazon pharmacists, pickup in minutes via app QR; expansion planned.
Implication: Could streamline initiation and adherence via remote prescribing and logistics.

🧠 Neuralink scale-up hinges on robot surgeons [4] [US • 10 Oct 2025]

https://uk.pcmag.com/health-fitness/160661/neuralink-has-a-huge-patient-waitlist-bring-on-the-robot-surgeons

Context: Company cites ~10,000 waitlist, 12 trial implants so far, targeting 25 by year-end; no FDA approval for sale.
Key point: Robotic arm inserts ultra-fine threads; latency reportedly ~10× faster than brain-to-muscle output, per Morgan Stanley report.
Implication: Signals pipeline investment and modality expansion.

🤝 Microsoft taps Harvard Health for Copilot [5] [US • 10 Oct 2025]

https://www.digitalhealthnews.com/microsoft-partners-with-harvard-to-enhance-copilot-s-health-ai-capabilities

Context: Licensing Harvard Health Publishing content to answer health queries; part of diversifying beyond OpenAI models.
Key point: Update expected as soon as this month per report; Microsoft declined comment, Harvard not responsive.
Implication: Could streamline initiation and adherence via remote prescribing and logistics.

🤖 All-AI papers and reviews at Agents4Science [6] [14 Oct 2025]

https://www.nature.com/articles/d41586-025-03363-3

Context: Online event 22 Oct; >300 AI-agent submissions, 48 accepted; human attendees, AI or humans present.
Key point: Conference will compare AI vs human reviews, document agent error modes and human-AI interaction levels.
Implication: Could inform practice and payer discussions; interpretation depends on study design and confounding control.

🔥 DeepMind/Yale combo to “heat up” cold tumors [7] [15 Oct 2025]

https://interestingengineering.com/health/google-deepmind-new-ai-cancer-treatment

Context: C2S-Scale 27B screened ~4,000 drugs across contexts; hypothesis validated in human cell models.
Key point: Silmitasertib + low-dose interferon increased antigen presentation ~50% in specific immune contexts. Preclinical stage.
Implication: May influence prescriber choice and payer reviews pending full data.

❤️‍🩹 AI and digital health for China’s CVD burden [8] [China • 16 Oct 2025]

https://bioengineer.org/ais-role-in-preventing-cardiovascular-disease-in-china/

Context: Perspective on prevention gaps across urban–rural settings; calls for privacy safeguards and integration into care.
Key point: mHealth plus AI risk stratification could expand access, but fragmentation and literacy barriers persist.
Implication: Access programs may expand screening, initiation, and follow-up at scale.

🫀 GRACE 3.0: validated and extended across 10 countries [9] [EU • 16 Oct 2025]

https://www.thelancet.com/journals/landig/article/PIIS2589-7500(25)00089-5/fulltext

Context: 609,063 NSTE-ACS patients; external validation across 7 countries; VERDICT RCT used for individualized treatment-effect model.
Key point: In-hospital AUC ~0.90, 1-year AUC ~0.84; model flags subgroups with benefit from early invasive care (HR 0.60, 95% CI 0.41–0.88).
Implication: May influence prescriber choice and payer reviews pending full data.

👶 One Medical adds pediatric telehealth pay-per-visit [10] [US • 17 Oct 2025]

https://www.digitalhealthnews.com/amazon-one-medical-launches-telehealth-platform-for-pediatric-care

Context: Ages 2–11; $29 message visits, $49 video; no One Medical membership required.
Key point: Covers common conditions and renewals, with prescriptions via Amazon Pharmacy or preferred pharmacy.
Implication: DTC/telehealth could streamline initiation and adherence via remote prescribing and logistics.

Why it matters

Point-of-care access and telehealth from Amazon could reduce primary-care friction and improve time-to-therapy.
Contactless vitals via Wi-Fi suggest low-cost monitoring routes beyond wearables.
Validated GRACE 3.0 may refine invasive-strategy decisions in NSTE-ACS.
Preclinical AI-guided combos from DeepMind hint at new immunotherapy avenues.
AI-only science venues will test the reliability and limits of agent-driven research.

FAQ

What hardware powers Pulse-Fi and how accurate is it?

ESP32 microcontrollers and Raspberry Pi devices run Pulse-Fi’s AI; the article reports <1.5 BPM error up to ~10 ft and across postures, with single-user tests to date [1].

How do the AML digital-twin models improve over time?

Accuracy improves as more treatment cycles are used for training; the paper suggests ~5 cycles are needed post-diagnosis for good neutropenia predictions on Ven/Aza [2].

When and where will Amazon’s prescription kiosks appear?

At One Medical clinics in Los Angeles starting December, expanding later. They exclude controlled and refrigerated medications [3].

Is Neuralink’s implant widely available?

No. It is in clinical trials with 12 patients implanted so far and a reported ~10,000 waitlist. No FDA approval for sale; robot-assisted surgery is positioned as the scale lever [4].

What changes are coming to Microsoft Copilot’s health answers?

Per reporting, Microsoft will license Harvard Health Publishing content so Copilot can respond with clinician-aligned information. Timing is as soon as this month, per the outlet, with Microsoft declining comment [5].

How is GRACE 3.0 different from GRACE 2.0?

GRACE 3.0 adds sex-specific NSTE-ACS models for in-hospital and 1-year mortality and an individualized early-invasive benefit model, showing improved discrimination and reclassification on external validation [9].

Entities / Keywords

Pulse-Fi; UC Santa Cruz; ESP32; Raspberry Pi; venetoclax; azacitidine; AML digital twin; Amazon Pharmacy; One Medical; Neuralink N1; Telepathy platform; Microsoft Copilot; Harvard Health Publishing; Agents4Science; DeepMind C2S-Scale 27B; silmitasertib (CX-4945); interferon; China CVD prevention; GRACE 3.0; VERDICT trial; NSTE-ACS.