This week: the FDA asks how to examine a generative AI device, a nephrologist turns the week's papers into a podcast, and pharmacists and nurses fill the shipping log.
Top signal
The FDA opens the file on generative AI devices. A discussion paper from the FDA's Digital Health Center of Excellence (18 August) sets out how generative-AI-enabled medical devices might be regulated: risk assessment on two axes, premarket evaluation, postmarket monitoring, and considerations for foundation models and agentic systems. The idea to watch is a competency assessment "inspired at a high level by how physicians are trained and evaluated": non-clinical benchmarking of clinical knowledge, analytical ability, safety behaviour and communication, then clinical confirmation with real patients and clinicians. Clinicians are explicitly invited to comment; docket FDA-2026-N-7874 is open until 19 October. FDA
Shipped this week
- AudioScholar: nephrologist Nikhil Shah (University of Alberta) built, with Lovable, a free weekly podcast of each specialty's most important papers: about ten minutes, on Apple Podcasts, Spotify and YouTube, no account needed. His answer to years of trying to make Twitter work for following the literature. X
- Curovana: Dr Mihir H. Vasudev, part-time doctor and full-time co-founder in Delhi, launched the clinical simulation app on Google Play: AI patients with modelled physiology, an AI tutor that watches you work, exam prep and spaced repetition, free for medical students and doctors. For training only, not a medical device, as the site says itself. X
- ICU AI Assistant: B Gurubasavaraj, a BAMS clinician and healthcare data engineer in Bengaluru, shipped a Streamlit app for explainable arterial blood gas interpretation and ventilator assessment: rule-based acid-base reasoning, ML-assisted classification, oxygenation analytics and trend monitoring. GitHub
- NHS prescribing, queried: Zahra, a UK pharmacist and data analyst, published a SQL (Supabase) and Tableau analysis of NHS prescription data: which regions spend most, which BNF chapters cost most, what gets dispensed most and at what unit cost. GitHub
- Pharmacovigilance, automated: registered pharmacist Shreyash Shriram published a prototype that pulls adverse drug reactions out of social-media posts and codes them to MedDRA preferred terms, orchestrated with Make.com, ChatGPT and Airtable. An academic project running on simulated feeds, and labelled as such in its own README. GitHub
- CLARA: Melinda Yormick, MBA BSN RN, is building an app that shows operating-room teams where people and equipment are in real time, so nobody hunts for the one thing they need; featured by the American Nurses Association. X
- And in the workshop, ER nurse turned developer Anna's Bolus (Issue 2) now has a landing page and waitlist at bolus-app.fr, drug monographs plus dose and flow-rate calculators for French nurses and midwives; and endourologist Jonathan Pavlinec MD, founder of Notafid, reports building his practice's own app for hospital-to-office handoffs. bolus-app.fr
Build safely
- 1,357 cleared, 3 tested on outcomes: a PLOS Digital Health analysis of every FDA-cleared AI device to December 2025 finds 34 linked to registered prospective trials, 12 with peer-reviewed publications, and three evaluated on patient-centred outcomes such as mortality or readmission. The authors' conclusion is that clearance should no longer define readiness; read it alongside the FDA paper above. PLOS Digital Health
- Who bills for the algorithm, and who is liable: the American Medical Association has proposed billing codes for "clinically meaningful algorithmic analyses", AI outputs produced without a physician at the point of care. Nursing organisations object that the algorithm becomes a line item while nursing stays buried in the room charge, and the piece notes that responsibility for a wrong output "remains largely unanswered". Law-firm commentary, so read it as such; comments on the CMS proposed rule close 31 August. National Law Review
Events
- 27 Aug (online): next Thursday, SIIM's webinar "The Future Is Prompted: Vibe Coding for Healthcare Apps" with radiologist Ali Tejani MD. Last call. SIIM
- Aug to Oct (India): Health-a-thon 2026, billed as India's first national health hackathon for doctors and technologists, runs through October; from the Koita Foundation. Koita Foundation
- 3 Oct (Antwerp): Care & Code Clinical Build Day: doctors, nurses, pharmacists and physiotherapists vibe-code a working AI care tool in one day. Disclosure: this is our own event. careandcode.be
- 19 Oct: two dates on one day: applications for Cohort 11 of the NHS Clinical Entrepreneur Programme open (until 20 November), and the FDA docket on generative AI devices closes. NHS CEP
New in the directory
- Wiktoria Milczynska, MD (UK): internal medicine physician turned founder; her chronic skin-condition companion app Moya Health went live on the App Store (Issue 1) and on Product Hunt (Issue 2). moyahealth.app
- The directory now counts 36 clinician-builders with verified, evidence-graded ships. Browse it, forward this to a colleague who ships, and reply if you know someone who belongs in it. cliniciansthatcode.com