Skip to main content
RecovrFlow
Back to all issues
Issue #1310 November 20264 min

🏥 RecovrSignal Issue 13: A smartwatch AI cleared for AF detection, and a practice that cut coding costs 80%

WearablesPractice AdminPatient-Facing ToolsClinical Workflow
Issue 13 · 10 November 2026 · Fortnightly | AI in Healthcare · Australia & Beyond

It was a quiet fortnight for Australian news, so this issue looks overseas for things that will reach your clinic. Your clients' smartwatches are getting an AI second opinion on atrial fibrillation, a six-location practice says AI cut the cost of coding each surgical case by 80%, and patients can now hand their own health records to an AI assistant before they walk in. The fourth story is the honest counterweight: a study of 983,154 portal messages found AI drafting sped up the messages it touched and did not reduce after-hours work.

Here's what you need to know:


Summary

Read time: ~4 min

Top Stories

Signals


Top Stories

A Smartwatch AI Cleared for Atrial Fibrillation Cut Inconclusive Readings by 92%

🌏 Global | Wearables

PulseAI's model analyses ECG recordings taken on a wearable and flags arrhythmias. In a prospective study published in Heart Rhythm, it was compared with the Apple Watch's native algorithm on the same recordings from 296 analysable participants at an Irish teaching hospital, collected between November 2022 and February 2023. The company reports higher sensitivity for atrial fibrillation and a 92% reduction in inconclusive interpretations.

The caveats are real. The cohort was people in a cardiac procedure setting, where arrhythmia is more common than in the community, so it does not show how the tool performs on a healthy runner or a client in a falls program. The company's release is the source for the headline claims, a '10x' performance figure is not explained there, and the study has links to the company. This is a US clearance. It says nothing about availability in Australia, and nothing here has been cleared by the TGA.

Why it belongs in an allied health newsletter: wearables already put heart rhythm data in front of your clients, and the common failure is a vague alert with no next step. A tool that turns more 'inconclusive' readings into a clear yes or no changes the conversation a client brings to you. It does not change who diagnoses.

What it means for clinicians:

  • Add one line to your screening routine: ask whether the client's watch or ring has ever flagged an irregular rhythm, and record the answer and date.
  • If a client reports a wearable alert before or during an exercise or rehab session, stop the session and send them to their GP for a proper ECG. A consumer alert, AI-read or not, is never a diagnosis.
  • Do not tell clients a particular wearable is 'cleared' unless it is on the TGA register. US clearance does not carry over to Australia.

A Six-Location Practice Says AI Cut the Cost of Coding Each Surgical Case by 80%

🌏 Global | Practice Admin

The Bone and Joint Institute of Tennessee uses a vendor called Ember AI to code surgical cases. Its chief executive, John Reynolds, told Healthcare IT News that one of the most significant results has been an 80% reduction in cost per coded case, with the vendor reporting accuracy comparable to an expert surgical coder.

Treat the number with care. The 80% figure comes from the practice's chief executive and the vendor, not an independent audit, and surgical coding in the US is a different task from NDIS or Medicare billing in Australia. The practice also says it adopted the tool to improve coding accuracy and unit costs rather than to replace experienced staff.

The useful part for an Australian clinic is the shape of the win. The AI took a repetitive, rule-heavy task that sits between the clinical record and the money, and a human expert stayed in the loop. Billing codes, funding-line mapping and claim checks are the equivalent jobs in your practice, and they are where admin hours go.

What it means for clinicians:

  • List the repetitive, rule-based admin jobs in your week, such as matching sessions to NDIS support items or checking claims before they go out. Those are the tasks a vendor will pitch AI for first.
  • When a vendor quotes a saving, ask whether it was measured by an independent party or by the customer, and what error rate they measured alongside it.
  • Keep a person checking the output. The Tennessee result is described as AI plus experienced staff, not AI instead of them.

Patients Can Now Hand Their Health Records to an AI Assistant Before They Walk Into Your Clinic

🌏 Global | Patient-Facing Tools

The tool is called athenaPatient MCP. Patients sign in with their patient portal credentials, choose what information the assistant can see, and can revoke access at any time. The data it can draw on includes lab results, medications, vitals, conditions, diagnostic reports, immunisations, visit history, allergies, procedures and care plans. At launch it is available to US subscribers of one AI assistant, with other assistants planned, and athenahealth says it will limit access to assistants that do not train on patient information.

Fierce Healthcare notes that patients are already turning to generative AI to interpret lab results and medication instructions. This simply makes it easier and gives the assistant their actual record rather than a pasted guess. Some reporting also flags that AI tools connected to medical records are not without risk.

Nothing like this is available to Australian clients today, but the behaviour is already here. Clients ask AI about their conditions and arrive with a summary, a list of questions, or a confident wrong idea. Your job is to be the person who can tell which is which.

What it means for clinicians:

  • Ask new clients whether they have used an AI tool to look into their condition or their records, and note what they were told. You are checking for useful questions and for misunderstandings.
  • Prepare a short plain-English explanation of why you recommend what you recommend. A client with an AI summary in hand will compare it with yours.
  • Update your privacy and consent wording now. Clients who paste your reports or their notes into an AI tool are sharing health information, and you should be able to say what you advise.

AI Drafted Replies Sped Up the Messages They Touched, and Did Not Cut After-Hours Work

🌏 Global | Clinical Workflow

The study compared each clinician before and after they began using an AI response tool. Drafts were used on about 5% of messages. Where a draft was used, the reply was 29.5% faster than a message the same clinician answered without one, and 35.9% faster for lab result messages. Overall, though, clinicians answered portal messages 8.6% slower after the tool arrived.

The after-hours finding is the one that matters. Adoption and draft use were not linked to a drop in the share of replies sent outside working hours, and coverage reports more after-hours work for messages that arrived after hours. Clinicians chose message by message whether to use a draft, which favours simple, routine questions, so the speed-up is likely flattering.

For a practice owner, this is a useful corrective to a pitch you will hear. An AI drafting tool can make the easy messages faster. It does not by itself give anyone their evenings back, because the work that fills the evening is not the easy messages.

What it means for clinicians:

  • If a vendor promises less after-hours work, ask for after-hours data from a customer like you, not a time-per-message figure.
  • If you use AI to draft client emails or messages, read every word before it goes out and keep your own tone. Drafts suit routine replies like appointment logistics. They are a poor fit for anything clinical.
  • Measure before you buy. Note how many minutes a week you spend on messages and admin now, so you can tell whether a tool really changes it.

Signals

🌏 Global | Drug Discovery: The prize recognises work that combined deep learning, bioengineering and lab validation to find new antibacterial candidates against drug-resistant pathogens, starting with the 2020 discovery of halicin. Barzilay says AI is shifting from a discovery tool to a design tool, and she cautions that it automates parts of the drug pipeline rather than transforming all of it. For clinicians in wound care, post-surgical and aged care settings, antibiotic resistance is the problem behind the headline, and this is a sign that AI is now pointed at it. Read →

🌏 Global | Diagnostics: The CORE 300 lets a clinician switch between analogue and digital listening, and its AI analyses recordings for suspected structural heart murmurs. It went on sale in the US on 22 October at US$199, with AI analysis requiring a subscription of US$119.99 a year after a 14 day trial. It is a US launch, but the pattern of cheap hardware plus a subscription for the AI is the one allied health clinicians will be offered next. Read →

🌏 Global | AI Policy: Senator Bill Cassidy released the PATIENT AI Act as a discussion draft on 29 October, with public comments due by 1 December. It would have Medicare trial payment for AI tools that work without a physician interpreting results in real time, and would require patients to be told when AI is part of their care. It is a draft and not law, but the notice-to-patients idea is one Australian regulators and clients are likely to want too. Read →

🌏 Global | Imaging: AVIEW Aorta Metrics segments the aorta from chest CT images, classifies aortic regions and measures diameters, and can produce results for the radiology report. It is the company's 13th FDA clearance, announced on 23 October. For allied health the relevance is indirect: more incidental findings from scans your clients already have are likely to reach reports, so expect questions about them. Read →

Enjoyed this?

Get RecovrSignal free, every fortnight.