Hi Pulse Readers - this week, we’re diving into:

• the NHS’s largest surgical AI evaluation to treat more patients in existing theatres,

• what a study reveals about moving a self-harm AI surveillance tool between hospitals,

• and how to use Heidi’s template syntax to generate notes in your own style.

Subscribe now

Also, congratulations to our GPCME South Conference Giveaway Winner 🎉 !

A big shoutout to Theresa Koen from Turuki Health Care who has won a Heidi Remote by subscribing to The Pulse. Enjoy staying hands-free and focused on your patients!!

There is still one more Heidi Remote to be won by one lucky subscriber. Our September giveaway closes in two weeks, and every new subscriber since August is in the draw.

Now, on to this week’s issue...

2. Case Study – Your Fortnightly Practical

Video Source: Dr Karl Cole (MB ChB, FRNZCGP), Supercharge your GP Practice with Heidi Webinar in collaboration with Health Accelerator and Collaborative Aotearoa

Editing Template Structure with Syntax

Case Presentation: When Dr Harry first started with Heidi, he built a template by uploading a de-identified example note to Heidi.

One of those templates has been producing outputs that are not quite his style. A section he wants as a short list comes back as full paragraphs. Notes sometimes appear under another section when nothing relevant was discussed. He would also like a standard sign-off on every note. He opens the template editor, switches to the Structure tab, and finds brackets and quotation marks he does not fully understand.

He wonders what each part means, and how to customise the template so Heidi writes like him.

Approach: Understand what each part of the template syntax controls, so you can shape the output to match your style of documentation.

1. Open the Structure tab of a template editor

   From a note, select the three dots beside the scribe style, choose View / edit template, then click on the Structure tab.

   Note: You can also change a template by editing the example note or by asking Heidi in plain language, and Heidi updates the structure for you automatically.

2. Write section headings in plain text

   Type each heading as ordinary text, for example HxPc, OE, imp, plan. Text without brackets appears in the note as written, so your headings stay in place every time.

3. Use square brackets to signal what information to include

   A placeholder in square brackets tells Heidi what information to insert, for example [Describe patient's presenting complaint]. Describe the type of content you want rather than a specific example, and Heidi fills it from the consult.

4. Use round brackets for how to handle it

   An instruction in round brackets tells Heidi how to write a section, for example (Write in narrative paragraphs with full sentences) or (List as bullet points using dashes). Place it beside a placeholder to apply it there.

   Important: To stop Heidi filling gaps with assumptions, add a safety instruction such as (Only include if explicitly mentioned in transcript, contextual notes or clinical note, else omit section entirely). Always remember to review the generated note for accuracy before copying it into your PMS or EHR.

5. Use quotation marks for exact wording

   Text in quotation marks appears word for word, every time. Use it for fixed lines such as "Signed Dr John Smith, Cardiologist".

6. Add conditional logic for rules that change

   For content that depends on the situation, write an if / then instruction in round brackets, for example (If no allergies are mentioned, write NKDA), or (If no examination findings are mentioned, write "Examination not formally done"). This keeps routine entries like allergies and examination findings consistent without editing each note by hand.

Outcomes: Now the template has been edited for Heidi to write in Dr Harry’s style. The list stays a list, the unwanted section stays out unless it comes up, and his sign-off appears each time.

Understanding the syntax means he can customise a template for any recurring task - an ADHD review, a rest home progress note, or an ACC report. Heidi is writing notes just like him, and Dr Harry makes fewer edits to them.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

NHS Launches Its Largest Surgical AI Evaluation to Treat More Patients in Existing Theatres

Eight NHS hospitals have begun the largest surgical AI evaluation yet run in the health service, a year-long programme announced just last week. Led by UK health technology company Proximie, with Amazon Web Services (AWS) and Deloitte, it deploys Proximie's Intelligence Suite across 104 operating theatres, endoscopy suites, and catheterisation laboratories at the eight sites, six in London and two elsewhere in England.

Image Source: Proximie

Operating theatres are among a hospital’s most expensive rooms, and a share of their time is lost in the gaps between one operation and the next.

The aim is to treat more patients across more than 20,000 procedures a year using existing theatres and staff, without longer working hours.

Key Features:

• Runs in the background: the tool captures surgical video and procedural data from every room as the list proceeds, without disrupting the surgical team

• Anonymised at source: identifying details are removed at the point of capture, so the system works only from the timing and flow of the list

• Live visibility for the whole team: it turns that data into real-time predictions and alerts, giving nurses, surgeons, anaesthetists, and operational leaders a live view of theatre turnover, utilisation, on-time starts, and variability to act on

• Early results at live sites: where it already runs, at University Hospitals Coventry and Warwickshire and at Barking, Havering and Redbridge, Proximie reports an 8.2% increase in cases, a 30% reduction in turnover time, and the potential for one extra case a day

Implications for the Health System and Clinicians: This is AI aimed at the operational side of care, patient flow through costly, resource-heavy theatres, and it leaves clinical decisions to the team. For Aotearoa, where elective waiting lists and workforce shortages are constant pressures with little room to add theatres or staff, getting more from the theatres already running is worth watching. The caveat is that the headline figures are early and company-reported from a handful of sites. It would be interesting to see whether they hold across eight hospitals over a full year, and translate into shorter waits rather than simply busier theatres.

AI for Public Health Surveillance: What a Study Reveals About Moving It Between Hospitals

Most healthcare AI aims to help treat an individual patient. This tool has a different job: public health surveillance. In a study in PLOS Digital Health, an Australian team tested an AI that reads emergency department triage notes to track self-harm across a population, the strongest risk factor for later suicide.

Image Source: PLOS Digital Health

The team asked whether a tool built at one hospital still works when it moves to another. Developed at the University of Melbourne on notes from the metropolitan Royal Melbourne Hospital, it was tested there over four years and at Latrobe Regional Hospital, 150 km away. The system cleans up the brief, error-filled triage text, then a machine-learning model flags the notes whose wording signals current self-harm. It was validated on 329,655 metropolitan and 316,877 regional notes.

Key Findings:

• Stable at the city hospital: At Royal Melbourne, the tool held its performance, scoring 0.84 where 1.0 is perfect.

• Weaker at the regional hospital: At Latrobe Regional, its ability to pick out self-harm fell to 0.78, with precision of 71% and sensitivity of 73%.

• Local wording and methods drove the gap: The regional notes were written differently, and self-harm there more often involved medication overdose than self-injury.

• Better than codes, not yet for individual care: It catches far more self-harm than diagnostic coding, which misses over 60%, though about 23% of what it flags may be suicidal ideation.

Implications for Healthcare Systems:

For New Zealand, the wider lesson is about AI itself. As health services push AI out from big city hospitals to smaller rural ones, a tool proven in one setting cannot be assumed to work in another, since local wording and patient populations can quietly erode it, a risk for rural and underserved communities. As a validation study rather than a trial, and one that still confuses some self-harm with suicidal ideation, it points to local tuning and prospective testing before clinical use.

Read the full study here.

4. Vitals – Quick Bytes

• AI Early-Warning Tool Linked to Fewer Ward Cardiac Arrests in Resource-Limited Korean Hospitals

  A retrospective study tested DeepCARS, an AI-SaMD that predicts which ward patients are likely to have a cardiac arrest within 24 hours. It uses just four routine vital signs, and flags a risk score inside the usual electronic record alongside existing EWS. Three secondary hospitals in South Korea have adopted it because they cannot afford a rapid-response team. It alerts the bedside team when a patient's score climbs. No extra staff are needed. Across 164,761 ward admissions, DeepCARS’s use was linked to 21% fewer cardiac arrests and 15% fewer in-hospital deaths. The gains were larger in patients with sepsis. However, the study was retrospective, so it shows association rather than cause, and the authors also say that RCTs are needed to confirm it.

• Health New Zealand Introduces a National Framework to Assess AI Tools Before Use

  6 weeks ago, Health NZ introduced a national framework for checking AI tools before they enter the health system. The New Zealand AI Pre-Implementation Evaluation Framework sets one consistent way to judge whether a tool is safe, secure, ethical and equitable. It was built over 18 months as a Health Research Council project, led by Associate Professor Rosie Dobson. It is tailored to a country like NZ, with no specific AI law. The Health NZ Board has endorsed it as the method its National AI and Algorithm Expert Advisory Group will use to assess new tools in the public health sector.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• a world-first live AI-assisted brain tumour surgery to protect a patient's sight,

• what a 1,100-patient RCT found about AI for secondary prevention in heart disease,

• and how to dictate concise, focused referrals with Ask Heidi.

Subscribe now

2. Case Study – Your Fortnightly Practical

Video Source: Dr Karl Cole (MB ChB, FRNZCGP), Supercharge your GP Practice with Heidi Webinar in collaboration with Health Accelerator and Collaborative Aotearoa

Dictating Concise Referrals By Voice with Ask Heidi

Case Presentation: Last time, Dr Harry set up a lasting record for each patient, so their history and documents are there as context each time he sees them.

Today, he is seeing a woman who needs a rheumatology referral. She was assessed in Fiji and started on a medication that helped, but since returning her bloods have all been normal, and he is unsure of the next step.

In the past, he sometimes pasted the whole consult into the referral, and specialist colleagues find it hard to tell what was being asked. His referral system already holds the patient’s name and background, so what the specialist needs is the specific clinical question, kept short.

He wonders whether he can build the referral by voice and keep it that tight.

Approach: With Heidi Dictate, speak your request into Ask Heidi, then refine by voice until only the referral information you need remains.

1. Start from the finished consult note. Review the note as usual, so Ask Heidi works from an accurate source. Because the patient’s record is linked, her earlier notes and the specialist letter from Fiji already come in as context.

2. Dictate your request into the Ask Heidi bar. With the Heidi desktop app open, click into the Ask Heidi bar, hold your Dictate hotkey, and speak what you want. Dictate drops the text wherever your cursor sits, including this bar.

   Prompt example: “Write a referral from primary care to a rheumatologist asking for an opinion on the next steps for this patient. She was assessed in Fiji and started on a medication that seemed to help, but her bloods here have all been normal. Keep it concise.”

3. Refine by voice until it is concise. Read what Ask Heidi returns, then dictate follow-up instructions in the same bar to shape it: “That is too long, cut it to the key points,” or “Focus on the fact that she can now walk no more than 10 metres.” Each round takes seconds.

4. Review, then paste it into your referral. Read the final version in full, then copy it into the referral field in your PMS or EHR.

5. Save a template for the common referrals. For the types of referrals you write regularly, have a template that summarises the structure concisely, in your library. Next time, you dictate only the specifics.

Outcomes: Dr Harry now speaks the request for the letter, tweaks it in a few passes, and pastes a short, clear referral into his system within minutes. Heidi Tasks also captured the referral as an action from the consult, and it is in his task list ready to send.

The referrals themselves are also answered more readily. When the clinical question is easy to find, the specialist can act on it, which matters most for patients like this one, whose care crosses borders and depends on us to pick up the thread quickly.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

London Surgeons Perform the World's First Live AI-Assisted Brain Tumour Surgery to Protect a Patient's Sight

Surgeons at the National Hospital for Neurology and Neurosurgery (NHNN), part of University College London Hospitals (UCLH), have performed the world's first brain tumour operation supported by AI in real time. It removed an 11mm pituitary tumour that was costing a 48-year-old man his sight. The operation was part of an early UCL trial.

Image Source: Maeil Business

The pituitary gland sits at the base of the skull, wedged against the carotid arteries and optic nerves, where an error of a millimetre can cause blindness, stroke or death. Reaching the tumour with an endoscope through the nose, the surgeons operated while the AI read the live video feed, rather than pre-surgery scans, and marked hidden vessels and nerves on a second screen to show the safest zones.

Key Features:

• Trained on hundreds of operations: It learned from hundreds of annotated pituitary surgery videos, more than most surgeons see in a career, to recognise critical anatomy and instruments

• Works in real time: Unlike earlier surgical AI that reviews footage afterwards, it runs live on an NVIDIA Clara IGX platform that processes the video locally in theatre

• Assistive, not autonomous: It marks danger zones and safe areas on a second screen while the surgeon keeps full control, working like facial recognition but for hidden anatomy

• Early feasibility trial: The case is one of the first in a UCL and UCLH trial; the operation removed the tumour and preserved the patient’s vision, and a larger trial is planned

Implications for the Health System and Clinicians: This is an early sign of AI moving into the operating theatre, working with a surgeon, live. For New Zealand, where complex neurosurgery sits in a few major centres, a tool that acts as an expert second pair of eyes during high-risk procedures is worth watching, and its developers hope to extend it to other delicate operations. The caution is that this is a first-in-human case in an early feasibility trial, with regulatory approval pending and no outcomes data at scale. The jury is still out on whether it improves results across many surgeons and centres, without new risks from a mislabelled structure.

AI-Enhanced Secondary Prevention in Coronary Heart Disease: 1,100-Patient RCT

A randomised controlled trial in BMC Medicine tested whether an AI-enhanced smartphone system could improve secondary prevention after coronary heart disease. The single-centre, open-label trial at Fuwai Hospital in Beijing randomised 1,100 adults with confirmed CHD to the AIM-CHD system with usual care or usual care alone, over three months.

Image Source: BMC Medicine

After discharge, secondary prevention often drifts as follow-up fragments. AIM-CHD pulls in hospital and wearable data automatically, and has the patient photograph their follow-up blood-test reports. It then reads the results and checks them against guideline targets across 11 risk areas. When a target is missed, it sends alerts, education, and reminders, with treatment decisions left to the clinician. At baseline, cholesterol was already well controlled (LDL 2.03 mmol/L), leaving little room to improve.

Key Findings:

• A small drop in cholesterol: At three months, LDL was slightly lower, 1.56 versus 1.65 mmol/L, a modest but significant reduction of 0.09 mmol/L (p=0.03).

• More patients hit targets: More reached the LDL goal below 1.81 mmol/L (71% versus 64%; p=0.03) and the blood pressure goal below 130/80 mmHg (45% versus 35%; p=0.002).

• No change elsewhere: The system did not shift blood sugar, weight, smoking, or medication-taking, and made no difference to a combined measure of heart attacks, strokes, readmissions, and deaths (3% versus 2%; p=0.69).

• Low burden, but patchy engagement: Onboarding took about 10 minutes of nursing time, yet fewer than half (44.3%) sent in a follow-up cholesterol result for the app to act on.

Implications for Healthcare Systems:

In NZ, secondary prevention is daily primary-care work, and cardiovascular disease falls hardest on Māori and Pacific communities, so a low-burden phone-based system that nudges more patients to their targets is an appealing post-discharge model. But gains were modest and limited to surrogate measures. The participants are also mostly of Han Chinese ethnicity and the trial was run by the team that built the system. Longer follow-up ongoing and multicentre validation still needed.

Read the full study here.

4. Vitals – Quick Bytes

• RACP Backs Human-in-the-Loop AI to Support More Personalised, Equitable Care

  A recent position statement from the Royal Australasian College of Physicians (RACP) sets out how AI can be adopted safely and ethically in clinical practice. It points to gains already flowing through image-based disciplines such as radiology and pathology, alongside stronger diagnostic and therapeutic decision support and greater patient self-management. The RACP frames AI as a route towards a more personalised, equitable health system. To get there safely, the college backs an agile, human-in-the-loop approach that keeps clinical judgement central while building AI skills across training. This is supported by clinical governance, training, and data sovereignty for indigenous peoples. It cautions though, that AI should not be seen as a complete fix for the structural and funding constraints facing the health system.

• EchoNext Gains First FDA Clearance to Flag Hidden Structural Heart Disease From a Standard ECG

  EchoNext, a screening tool developed by Pathway Labs with NewYork-Presbyterian and Columbia, is the first AI cleared by the FDA to detect six forms of structural heart disease from a standard 12-lead ECG. It reads patterns in an ECG that are invisible to the human eye and flags patients who need a confirmatory echocardiogram. Trained on more than 700,000 ECG and echocardiogram pairs, it was validated across more than 20 hospitals and 500,000 patients in North America. On 3,200 ECGs it identified 77% of structural heart disease against 64% for cardiologists, and among nearly 85,000 people screened it roughly doubled the usual diagnostic yield. The tool triages rather than diagnoses and does not replace echocardiography or specialist judgement. Its performance in broader community settings is still to be proven at scale.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• how a US hospital put the first FDA-authorised sepsis diagnostic into clinical use,

• what a 295-patient RCT reveals about AI and gene diagnosis,

• and how to keep each patient's history and documents together in Heidi.

Subscribe now

2. Case Study – Your Fortnightly Practical

Image Source: Heidi Health

Keeping Each Patient's Clinical Context in One Place

Case Presentation: Dr Harry recently learned how to explain to patients what happens to their information inside Heidi.

Running a busy practice, many of his patients return. Some of the longer sessions such as the adult ADHD assessments tend to run across several appointments. At the start of each patient visit, Dr Harry re-enters the same history, medications and allergies into the context window, and hunts for the specialist letter he saved previously.

Dr Harry wonders whether Heidi can hold a lasting record for each patient, so their long term history and documents are simply there each time he sees them.

Approach: Open the Patients feature to add long-term/background information for each patient, then link sessions so the context carries from one visit to the next.

1. Open your Patient list

   In the left-hand navigation menu, select Patients. This opens a searchable list of every patient you have created or linked a session to. Select a patient to see their details on the right: their linked session history, stored details, and any attached files. To add someone new, select New patient and complete their details.

2. Review and complete the stored fields

   A patient record holds fields such as identifier, date of birth, past medical history, current medications, and allergies. Fill in what is useful. These persist across sessions and come in as context for each new session, so you do not re-enter them.

Important: Your PMS or EHR stays the source of truth for the patient record. The Heidi record holds the clinical context that lives within Heidi only.

3. Attach documents to the patient

   In the patient's profile, select Files, then Attach files, to add background such as a referral letter or a specialist report. These stay with the patient and can be selected as context in any future session with them.

4. Start a linked session

   Start a new session from within the patient, or at session start type the identifier and let Heidi match the record. Once the session is linked, the stored fields, the documents you choose, and up to three recent prior sessions automatically come in as context, so the note reflects continuity across visits.

5. Review what Heidi extracts afterwards

   When Heidi detects new details in the transcript, such as a medication, an allergy, or a contact number, it offers to save them to the patient. You review each suggestion and accept or dismiss it. Nothing is saved without your confirmation.

Outcomes: Dr Harry now opens the patient and finds their record already in place at the start of every visit. Their history, documents, and recent sessions are available as context from the first minute.

For his ADHD follow-ups, where continuity across long, multi-visit assessments matters, the record carries the story forward. His notes reflect the full picture across visits, and he spends less time rebuilding context and more time with the patient and their whānau.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

LifeBridge Health Puts the FDA-Authorised Sepsis ImmunoScore Into Clinical Use

Back in Issue 1, our Vitals review of AI in emergency-department sepsis care flagged tools like the Sepsis ImmunoScore. We noted sepsis AI still needed stronger validation and better workflow fit. That tool has now reached the bedside.

Image Source: Prenosis

LifeBridge Health’s Sinai Hospital of Baltimore has started using the Sepsis ImmunoScore, an AI diagnostic from Chicago-based Prenosis. It analyses 22 parameters of the immune response to flag sepsis, or its progression, within 24 hours. Results return in under an hour, inside the electronic medical record. Sinai went live with the tool in June and, by late July, had used it on 75 patients."

Sepsis is hard to diagnose. Uncertainty starts on arrival, blood cultures take days, and both unnecessary antibiotics and missed cases cause harm. In the US it affects at least 1.7 million adults a year.

Key Features:

• A regulatory first: In 2024 the Sepsis ImmunoScore became the first software the FDA authorised to diagnose sepsis, cleared through its De Novo pathway

• Peer-reviewed evidence: A study in NEJM AI found high accuracy for sepsis diagnosis and also predicted in-hospital mortality, length of stay, ICU admission, ventilation, and vasopressor use

• Built to fit the workflow: LifeBridge’s emergency medicine chief says the tool seldom changes care in the ED itself; its value comes downstream, giving hospitalists more diagnostic certainty when they take over the patient

• A precision-medicine approach: Prenosis treats sepsis as a syndrome with many biological subtypes, drawing on a biobank that, by company reports, covers 11 hospitals and more than 35,000 patients

• What is being measured: Prenosis and LifeBridge are assessing the tool’s effect on sepsis-bundle (SEP-1) compliance, length of stay, in-hospital mortality, and 30-day readmission

Implications for the Health System and Clinicians: This is an example of AI moving from research into routine acute care. Sepsis carries high mortality and depends on early recognition. An easy-to-use tool that improves diagnostic certainty matches a familiar pressure in New Zealand's emergency departments. Prenosis plans to extend it to pneumonia, heart failure, and kidney injury, and LifeBridge to its other hospitals. Caveat being that it is still an early, single-site rollout with no outcome data yet. Whether sharper diagnosis improves the outcomes that matter is only now being measured.

Pointing to the Faulty Gene in Inherited Retinal Disease: What a 295-Patient RCT Shows

A randomised controlled trial in Nature Medicine tested whether an AI decision support tool, Retina4IRD, could help specialists predict the genetic cause of inherited retinal disease (IRD) before genetic testing. It compared specialists using the tool against specialists working alone across seven centres in China.

Image Source: Nature Medicine

Retina4IRD reads ordinary retinal photographs and scans alongside basic clinical details, then gives the specialist a shortlist of the genes most likely at fault. Even experienced specialists find these patterns hard to read by eye. It focuses on the treatable subtypes that make up about 60% of IRD cases. In the trial, 300 patients took part, and predictions were scored against genetic sequencing results, available for 295 patients.

Key Findings:

• Found the right gene more often: With the tool, the correct gene was in the specialist’s top five ranking, 88.5% of the time, against 67.3% alone (P<0.001).

• Better at the harder calls: The top pick was right 37.8% with the tool, up from 22.4%, and the top four covered 81.8%, up from 53.1%.

• The lift came from the AI: The same specialists scored 68.2% before seeing the tool’s output and 88.5% after, close to the 67.3% in the control group.

• Better plans, no harm: In an extra, exploratory analysis, management plans scored higher with the tool (37.7 versus 28.5, P<0.001), with no harm to patients.

Implications for Healthcare Systems:

Inherited retinal disease is a leading cause of blindness in children, and gene therapies make an early diagnosis matter. This is rare RCT-grade evidence that AI can lift even expert diagnostic accuracy, with scans most eye clinics already have. For New Zealand, where genetic-ophthalmology expertise sits in a few centres, pointing to the right gene sooner is worth watching. However, it guides testing rather than replacing it; the trial ran only in China, so multinational trials are the next test.

Read the full study here.

4. Vitals – Quick Bytes

• New Zealand's GastroTriage Uses AI to Triage Colonoscopy Referrals

  GastroTriage is an AI-assisted decision-support tool built in New Zealand to help specialists triage colonoscopy referrals, where endoscopy demand is high and free-text referrals must be read against several guidelines. Developed by Dr Jay Gong and Dr Henry Wei with Health New Zealand and the University of Auckland, it uses a large language model to read a referral against Ministry of Health criteria, return a recommended priority, flag anything missing or conflicting, and leave the final decision to the specialist. In an early evaluation of 100 referrals it matched a blinded consultant 70.1% of the time, against 58.0% agreement between clinicians themselves. A larger validation study of 600 referrals, half from Māori and Pacific patients, will build a specialist-consensus standard before any clinical use.

• First Fully AI-Discovered Drug Reaches Phase III, with its Phase 1 Trialled in Aotearoa

  A Phase 1 trial of 78 healthy volunteers in New Zealand was part of the early testing of rentosertib, now the first medicine with both an AI-discovered target and an AI-designed molecule to reach Phase III. Developed on Insilico Medicine's generative AI platform, the oral drug inhibits TNIK, a kinase implicated in fibrosis, for idiopathic pulmonary fibrosis. This is a progressive and often fatal lung-scarring disease. In an earlier Phase IIa trial of 71 patients in China, published in Nature Medicine and focused mainly on safety, the 60 mg daily dose improved forced vital capacity by a mean of 98.4 mL at 12 weeks, against a 20.3 mL decline on placebo. The Phase III trial will enrol 320 patients across 47 centres in China over 52 weeks, and the drug remains investigational and unapproved.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• Tāmaki Health's rollout of NZ's first primary care AI health coach,

• how an AI tool reads a standard pathology slide to flag bowel cancer relapse risk,

• and how to explain Heidi's data journey in plain language.

Subscribe now

Also, congratulations to our July Giveaway Winner 🎉 !

A big shoutout to Victoria Purdy from Workfit Auckland and O Hands Hand Therapy Clinic who has won a Heidi Remote. Use it well across both of your practices!!

Now, on to this week’s issue...

2. Case Study – Your Fortnightly Practical

Video Source: Hendrix Health

Following Your Patient's Information Through Heidi

Case Presentation: Last time, Dr Harry found his everyday note controls in one menu on the note screen.

This fortnight, a patient asks him a question after providing her consent for Heidi to transcribe the consultation. She wants to know where the recording goes, who can listen to it, and whether their conversation will be used to train any AI. Dr Harry answers honestly but thinly: the tool is secure, and he reviews every output himself.

Dr Harry realises he could not explain exactly what he was asking his patients to agree to. He wonders whether he can learn the path a patient's information takes, well enough to explain it simply.

Approach: Understand how information is passed through Heidi, so you can describe it to a patient in plain language.

1. No audio recording is stored

   Heidi transcribes as you speak and as soon as the speech is transcribed into text, the underlying audio disappears. The transcription occurs in real-time - think of it like the microphone button on your phone's keyboard, where your speech becomes text as you go.

   Files you place in the context window, a referral, a specialist letter, or photographed ambulance notes, are also converted to text and treated as part of that session, so add only what the note needs.

2. How identifying information is protected
   Identifying details are automatically removed from the transcript and replaced with placeholders, so a patient's name becomes something like Jane Doe or John Smith. Those identifiers sit in a separate secure vault, protected by their own layer of digital lock and key. The de-identified clinical text is then processed by large language models into a polished output, following the instructions in your template.

Note: At this stage the identifiers are released from the vault, and swapped back into the output. Please review any regenerated or edited note for accuracy before it goes out.

3. Data is not used to train AI
   Consultation data, session transcripts, and generated notes are not used to train AI models. This is often the question patients care most about, and it can be answered directly.

4. Decide how long sessions are kept, and know that deletion is final
   You choose how long Heidi keeps a session, from one day through to indefinitely. The default is to keep it, so set a period that matches your practice’s records policy. Once a session is deleted, it cannot be recovered.

Outcomes: Instead of a general reassurance and a hope that the patient does not press further, Dr Harry can now describe what happens to information from a Heidi transcript in a few sentences. Explaining it plainly upholds the patient’s mana over their own health information.

The same knowledge sits behind every consent conversation his team has. Patients and their whānau who understand where their information goes are better placed to make their own decisions.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

Tāmaki Health Launches New Zealand's First Primary Care AI Health Coach

Tāmaki Health, one of New Zealand's largest privately owned healthcare providers, has become the country's first primary care organisation to bring an AI health coach into routine care, partnering with digital health company Groov. The tool launched on 27 July 2026 and is now piloting across selected clinics.

Image Source: Tāmaki Health

The coach acts as a 24/7 digital front door. Patients can start a coaching conversation even before they book an appointment. It then helps them build personalised action plans for lifestyle and behaviour change, covering conditions such as diabetes, gout, weight, and anxiety.

The AI coach extends Tāmaki Health’s established health coach service. The provider says that service has improved management of chronic conditions and cut unnecessary clinic visits. Tāmaki Health says primary care cannot meet rising demand by adding appointments alone.

Key Features:

• Human in the loop: The coach handles early, lower-intensity support and hands patients to Tāmaki Health’s coaches and clinical teams when symptoms, distress, or requests for more help appear.

• Purpose-built platform: Powered by Groov, maker of Ask Groov (Health New Zealand’s first endorsed AI wellbeing guide), it uses clinically reviewed content and defined escalation pathways, and is security-tested and ISO 27001 certified.

• Built on a proven model: It is modelled on NZ’s Integrated Primary Mental Health and Addiction (IPMHA) programme and co-designed with patients and Tāmaki Health’s clinicians and coaches.

• Early results: Across more than 4,500 coaching conversations, users' self-rated confidence to manage their health rose an average of 32% after one session and 75% by month two. They rated the coach 8.5 out of 10 for helpfulness.

• Equity focus: The providers see earlier, more accessible preventative support as especially relevant for Māori and Pasifika communities, who carry a disproportionate burden of chronic disease.

Implications for the Health System and Clinicians: This is a local example of AI extending a proven human service rather than automating a clinical decision. Support that reaches patients between visits speaks directly to access and equity in Aotearoa. The partners will evaluate engagement, access, and outcomes as the pilot expands. However, these figures are early, self-reported, and come from a company-run pilot. Whether short-term gains in confidence become lasting behaviour change and measurable clinical improvement, across diverse populations, is the real test.

Refining Risk in Stage II Colorectal Cancer: What a 1,220-Patient Australian AI Study Tells Us

An Australian study in Gastroenterology tested an AI tool that reads a standard pathology slide to predict which stage II bowel cancer patients are most likely to relapse. This is the hardest time to make decisions about chemotherapy. Led from La Trobe and Melbourne universities, the AI was trained on a separate patient group with more advanced cancer, then tested on three patient groups, two external.

Image Source: Gastroenterology (AGA Journal)

The tool, called SÉMIL, looks at the edge of the tumour, where it meets healthy tissue, and judges whether that border is jagged (higher-risk) or smooth (lower-risk). Pathologists know this border matters, but it is too subjective to judge it by eye. SÉMIL scores an ordinary slide with no extra lab tests, with scores checked against whether patients stayed cancer-free over five years.

Key Findings:

• Spotted higher-risk patients: In all three groups, patients it flagged higher-risk relapsed about 2-5 times as often as lower-risk patients (hazard ratios 4.73, 2.84, 2.10; all statistically significant).

• Helped with the hardest cases: Even among patients already rated high-risk by current guidelines, where the call is toughest, it still picked out those at greatest risk in every group (hazard ratios 2.96 to 3.50).

• Added something new: After accounting for the risk factors pathologists already check, such as tumour depth and lymph node sampling, the AI tool still predicted relapse on its own, roughly doubling the risk (hazard ratio 1.98).

• Best paired with a pathologist: When tool and pathologist agreed a border was high-risk, those patients did worst of all (hazard ratio 3.96).

Implications for Healthcare Systems:
New Zealand has one of the world's highest bowel cancer rates, so a tool that reads risk from an existing slide, with no extra test, could help decide which stage II patients need chemotherapy after surgery and which can be safely spared. It can also act as support for pathologists. However, this was a retrospective study on stored slides, all from Australia, and the researchers say it must be tested in real time, guiding treatment, before it changes care.

Read the full study here.

4. Vitals – Quick Bytes

• RCT Shows That AI-Trained Physicians Diagnose Better in a LMIC

  A single-blind RCT in Pakistan tested whether large language models improve clinical reasoning in a lower-middle-income setting. After all 58 participating physicians completed a 20-hour AI-literacy programme, they were randomised to review up to six clinical vignettes with either GPT-4o + conventional resources or conventional resources alone. Those with LLM access scored a mean 71.4% on an expert-validated diagnostic reasoning rubric, against 42.6% for the control group, with no meaningful change in time per case. Some caveats: the study used vignettes rather than real patients, tested only GPT-4o across six cases, and did not include an untrained physician group with LLM access, to separate the AI training effect from the effect of merely having access to AI.

• HealthX and RosterLab Pilot AI-Powered Rostering Across Four New Zealand Sites

  Health New Zealand's digital innovation programme, HealthX, has confirmed a contract with RosterLab, which provides AI-powered healthcare rostering software, for an initial pilot across four sites: Counties Manukau, Waikato, Starship, and Waitematā. From July 2026 the pilot covers medical radiation technologist (MRT) services and other clinical staff. The platform uses optimisation and AI-assisted workforce management to build rosters that balance operational requirements, staff preferences, fatigue, skill mix, and compliance. It is a very tightly controlled expansion from around 500 clinicians rostered to about 900. RosterLab reports that an earlier project with Middlemore MRTs cut a 120-hour monthly rostering task to under 20 hours a month. This pilot is early, and no independent outcomes have been reported yet.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• how a Pacific-led provider halved nurse travel time with AI remote monitoring,

• what an 11,000-patient ANZ registry study shows about mortality risk after hip fracture surgery,

• and how to reach your everyday note controls from one menu in Heidi.

Subscribe now

2. Case Study – Your Fortnightly Practical

Image Source: Heidi Health

Finding Your Everyday Controls On the Spot

Case Presentation: Since last time, Dr Harry has been using Heidi Dictate to speak text into any application on his computer.

Lately he is thinking about the small changes he makes over and over. He switches his default template, adjusts a template that is not quite right, and checks an earlier version of a note when he has second thoughts about an edit. Each change sends him somewhere different: into Settings, across to the Templates tab, or back through several clicks to where he started.

Each step is simple on its own, though the back-and-forth breaks his flow between patients. Dr Harry wonders whether these everyday controls could sit closer to the note he is working on.

Approach: Open the more options menu on the session interface (the three dots at the top of a note) and access multiple controls all in one place.

1. See what the menu offers from where you already are

   On any note, select the three dots beside the template name and scribe style. The template settings, export controls, and version history all sit here, so you can make routine changes without leaving the note screen and taking multiple steps.

2. Make your everyday setup automatic
   Choose Set as default template so your current template becomes pre-selected for every new session. If you regularly need more than one document from a consult, for example a note and a patient letter, choose Queue in all sessions so each is generated as soon as you stop transcribing.

3. Keep or adjust a template on the spot
   When you have edited a note into the way you like, choose Save as new template to reuse it. Choose View / edit template to open the editor and change the headings, wording, or instructions without hunting for the template elsewhere.

4. Regenerate and roll back with confidence
   After editing or changing a template, choose Regenerate output to rewrite the current note using the latest version. Choose Version history to see what changed, who changed it, and roll back to an earlier version in one click.

5. Send or export a finished note

   The menu also includes Copy all text, Email to colleague, Share session, Export as PDF or .docx, and Print, which stay greyed out until a note is generated. Use them to pass a note to a covering colleague or file a copy directly.

Important: Review any regenerated or edited note for accuracy before it goes out. You remain responsible for the final note.

Outcomes: Rather than moving between Settings, the Templates tab, and the note screen for routine changes, Dr Harry now reaches these controls from one menu on the same screen. His default template is set, his common documents are queued, and an edit he is unsure about can be checked or undone in seconds.

For a busy clinician, having these controls in one place removes small interruptions that used to sit between him and the next patient. His attention stays on the note in front of him.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

Pacific Health Services Hutt Valley Halves Nurse Travel Time with AI Remote Monitoring

Pacific Health Services Hutt Valley, a Wellington-based Pacific-led provider serving more than 17,000 Pacific people, has cut nurse travel time in half after deploying an AI-enabled remote monitoring platform. The provider began piloting the tool, Speedback, in December 2024, and presented its results at the Te Tītoki Mataora Forum on 2 July 2026.

Image Source: sisgain.ae

Built by Singapore-based Equitech, the platform uses WhatsApp and SMS rather than a dedicated app. Patients submit blood pressure, blood sugar, wound images, and meal photos for nutritional feedback, which AI interprets to help monitor diabetes and other long-term conditions between 3-month checks. Clinicians review the trends and send personalised nudges.

Before Speedback, a single clinician could actively monitor around 8 clients through in-person visits, sometimes travelling to a home just to take a blood pressure for a family without transport.

Key Features:

• Workforce reach: Within 3 months, the provider reported nurse travel time and administrative workload each fell by 50%, letting monitoring scale from about 8 clients per clinician to 321 without adding staff.

• Runs on WhatsApp: Using a channel Pacific communities already rely on to stay in touch with family in the islands means there is nothing new to download or learn.

• Multilingual by design: Messages are available in Samoan, other Pacific languages, and Te Reo Māori, lowering the language barrier for the Pacific families the service reaches.

• Clinician in the loop: Clinicians monitor the trends and decide when to reach out, keeping clinical judgement with a person rather than the tool.

Implications for the Health System and Clinicians: This shows AI extending the reach of a small workforce rather than making clinical decisions, shifting care from episodic visits towards continuous community monitoring. For Aotearoa, where workforce shortages, transport barriers, and Pasifika health inequities are persistent pressures, a tool that lets one clinician follow more patients speaks directly to access and equity. The provider frames it as a return on investment for reaching more of its community. However, these figures are provider-reported from an early pilot. Whether the gains hold across a full year and other settings, and whether remote monitoring actually improves patient outcomes rather than workforce efficiency alone, remains to be shown.

Ranking Mortality Risk After Hip Fracture Surgery: What an 11,000-Patient ANZ Registry Study Shows

A retrospective study in Anesthesia & Analgesia used machine learning to rank the factors that best predict death after hip fracture surgery. The Australian-led team drew on the Australian and New Zealand Hip Fracture Registry, analysing 11,318 patients treated across 107 hospitals between 2016 and 2019.

Image Source: anesthesia-analgesia.org

Instead of a traditional statistical method, the team used a Random Survival Forest, a machine-learning approach suited to the many interacting factors and changing risks in these patients. It weighed 20 routine demographic, clinical, and perioperative variables and ranked each by how much it helped predict when a patient would die. The cohort had a mean age above 80, with dementia in 38% and an ASA grade of III or higher in 80%. Over a median follow-up of 630 days, 31% died.

Key Findings:

• Model performance: On patients it had not seen, the model correctly ranked who was at higher risk of dying about 73% of the time.

• ASA grade ranked first: Physical status, measured by ASA grade, was the strongest predictor of mortality (importance score 0.051).

• Dementia, age, and mobility followed: Pre-existing dementia (0.036), age (0.019), and pre-admission walking ability (0.016) ranked next, and 40% of those who died had dementia.

• Weaker contributors: Male sex, length of hospital stay, and bone medication at discharge added comparatively little.

Implications for Healthcare Systems:
For an ageing population where hip fracture is common and often fatal, a model built from information hospitals already hold offers a practical way to flag who is most at risk and support goals-of-care discussions. Because the registry includes New Zealand patients, the findings reflect local practice. The caveat is that the model identifies higher-risk patients; it does not explain what causes the risk or how to change it. The predictors (ASA grade, dementia, age, and mobility) are already familiar, so its value lies in ranking and clarity, and the authors say the findings need confirming in other patient groups.

Read the full study here.

4. Vitals – Quick Bytes

• RNZCGP Study Finds Patients and Clinicians Accept Health Data Use for AI, if It Serves the Greater Good

  A New Zealand interview study asked 51 patients and health professionals about using personal health information to build AI tools and use them in care. Divided into three groups, all accepted secondary use of their data only if it serves the greater good (vs commercial gain), and felt the NZ health system should benefit when a company profits from a resulting tool. All valued AI's potential to reduce administrative burden, and tied patient trust to transparency, consent, and strong governance, with Māori representation and data sovereignty raised as important to that trust. Every group agreed AI should support clinicians, not replace them. The authors caution this was a small qualitative study with few Pacific participants, and conclude no single approach to social licence is likely.

• Viz.ai Expands Its Stroke Coordination Model Into Pulmonary Care

  Back in Issue 5, we covered Viz.ai, the US platform across 2,000 hospitals that flags a suspected stroke on a CT scan and cut specialist notification from 45 minutes to 7. On 14 May 2026 the company launched the Viz Pulmonary Suite, applying the same detect-and-coordinate approach to three lung conditions: COPD, lung nodules, and pulmonary embolism (PE). For PE, the tool flags the clot on imaging and mobilises the treating team: in a single-centre study it reduced time to treatment from 1.75 days to 0.56 days and lowered in-hospital mortality among high-risk patients. Viz.ai has since signalled a move into multiple sclerosis. The caution is that these new-area results come from a single site and a company launch rather than independent evaluation. (Viz.ai)

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• how a South Australian home care provider use AI to cut morning rescheduling time,

• what a 9,700-patient Kenyan RCT reveals about generative AI decision support in primary care,

• and how to speak text into any application with Heidi Dictate.

Subscribe now

2. Case Study – Your Fortnightly Practical

Image Source: Heidi Health

Type Anywhere with Your Voice

Case Presentation: Last time, the practice set up an agent to answer routine patient calls at the front desk.

This fortnight, the friction is back in Dr Harry's own desk: the typing itself. Between consultations, he types in detailed prompts for Heidi Evidence and Ask Heidi, but also replies to a colleague's email, and fills the free-text boxes on an insurer form. None of it is a full consult note, so Heidi Scribe is not the tool, and he is at the keyboard for small jobs all day.

He talks quickly, types slowly, and wonders whether Heidi can let him speak text straight into these other applications, the same way he dictates a note.

Approach: Use Heidi Dictate to speak text into anywhere on your computer, whether it’s Microsoft Word or your email, with one hotkey.

1. Leave the desktop app open and grant permissions

   Dictate types into other applications only when the Heidi desktop app is open. If you use Heidi through a web browser instead, it works inside Heidi only. Install or update the desktop app, then grant microphone access.

2. Turn on Dictate and set your hotkey
   Head over to Settings, under Defaults, scroll down to Dictate. The default hotkey is the right-hand Ctrl key on Windows, or the fn key on a Mac. Change it under Shortcut if it clashes with another key.

Note: Heidi Dictate is built into the desktop app, and Basic Dictate is free for every Heidi user. It needs an internet connection and does not run offline.

3. Then press and hold to speak
   Click into any text field, press and hold the hotkey, speak, then release to drop the text in. For longer pieces, double tap the hotkey to dictate hands free, then tap again to stop.

4. Speak shorthand and commands as you go
   Dictate expands the snippets you already use in Heidi, so a short trigger such as "ros" drops in a full review of systems, and you can even add personalised commands or snippets by pressing the View history button under Dictate. Built for medical-grade use, it handles the drug names and clinical terms that standard voice-to-text often mishears.

5. Use the same hotkey across your day

   The same hotkey works in your PMS, your email, a patient message, or the Ask Heidi bar, and you can also adjust the output tone for each usecase. Dictate supports more than 90 languages, so Dr Harry can set his dictation language to English or Cantonese.

Outcomes: Dr Harry can now speak his text straight into whatever screen he is working in. The sore hands ease, and because Dictate works the same way everywhere, one habit - one button, carries across his whole desktop. He finishes the day with less admin behind him and more time for the people in front of him.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

Home Care Provider ECH Cuts Morning Rescheduling Time with an AI Rostering Agent

ECH, a home care provider in South Australia, has deployed an AI scheduling agent called Schedtris to manage last-minute changes to its rosters. Built by Gadali, an Indigenous-led technology company, the tool went live in March 2026 and now covers a client base of roughly 4,500 people, handling around 20 disruptions a day.

Image Source: iStock.com/hirun

When a care worker takes unplanned leave, Schedtris makes the shift vacant and searches for a replacement. It reads live workforce, skills, and availability data, grades each candidate’s suitability with a percentage score, and presents a shortlist. The final decision stays with the human scheduler.

Rostering in home care is a live optimisation problem. The pressure peaks in the morning rush, when a single absence can unravel a run of visits that must be rebuilt before carers set out.

Key Features:

• Operational impact: Reported morning rescheduling time has fallen by 50%, recovering around 15 hours of scheduling capacity a week. The average time to handle a vacancy dropped from about 8 minutes to under 4, and the disruption window from two hours to under one.

• Skills and continuity checks: It matches the qualifications each visit requires, factors travel time, and checks whether a worker has seen the client before, so an unfamiliar face is not sent to provide personal care.

• Human in the loop: Schedtris recommends and ranks options rather than reassigning staff automatically, with schedulers retaining the final call.

• System architecture: The agent runs inside ECH’s Microsoft Azure and 365 environment with secure read-only access to operational data, and cannot change a roster itself.

• Next phase: ECH is now linking the tool to its HR system and automating staff notifications when shifts change.

Implications for the Health System and Clinicians: For practice managers and health service leaders in Aotearoa, this recalls workforce tools like the NHS's Patchwork Health rostering rollout: some of the most useful AI in healthcare absorbs the operational churn that drains staff time. Last-minute roster changes are a familiar pressure across the health sector. The caveat is that these are self-reported figures coming from one provider's early deployment, and the setting is home care rather than a clinic or hospital. Whether the gains hold in other settings and across a full year remains to be shown.

Generative AI Decision Support in Primary Care: What a 9,700-Patient Kenyan RCT Reveals

A large randomised trial in Nature Medicine tested whether a generative AI decision support tool could improve primary care. Clinicians were randomly assigned to use the tool or not. Called AI Consult and built on GPT-4o, it sat inside the EMR at 16 Penda Health clinics in Kenya, where care is delivered by clinical officers, mid-level clinicians with a three-year diploma.

The tool analysed what the clinician entered and suggested diagnoses and treatments, while the clinician stayed in charge. In total, 103 clinicians and 9,691 patients took part between April and July 2025. The main measure was treatment failure within 14 days: patient returning unresolved, needing urgent care, or harm.

Image Source: nature.com

Key Findings:

• No real change in patient outcomes: Treatment failure occurred in 2.2% of AI-assisted patients and 2.0% of the others, a gap small enough to be chance rather than a real effect (adjusted odds ratio 0.77, P=0.13).

• Better notes and decisions: Clinicians using the tool were significantly more likely to record the right diagnosis, a thorough note, and a sound treatment plan (odds about 1.7 times higher, all P<0.001).

• No sign of harm: An independent review found no serious harm linked to the tool, though the trial was too small to rule out rare problems.

• Lower antibiotic costs: Spending on antibiotics was slightly lower when clinicians used the tool, and running the AI cost about US$0.04 per consultation.

Implications for Healthcare Systems:
The AI improved how clinicians documented and reasoned, but did not reduce treatment failures at 14 days. Outcomes also depend on much beyond a single clinic, such as whether patients take their medicine or other social things. For NZ practices weighing generative AI, better notes and decision support do not automatically mean better patient outcomes. Some other limits: one well-run city network, too few treatment failures to detect a small benefit, a short follow-up, and one LLM model that may be surpassed by newer ones. Whether these early gains reach patients is still an open question. The study was independently funded, but two authors also hold stock options in Penda Health, the operator.

Read the full study here.

4. Vitals – Quick Bytes

• Collaborative Aotearoa Survey Finds AI Already Embedded in Primary Care, Endorsement and Support Gaps Remain

  A survey of health professionals in primary and community care by Collaborative Aotearoa, a national network of PHOs, received 47 responses. Almost half reported using AI daily, and more than two-thirds daily or weekly, with clinical documentation and transcription the most common use, followed by administration and communication support, clinical decision support, and data analysis. Almost all respondents who answered the question supported some form of national endorsement to confirm safety and effectiveness. Awareness of existing pathways, including the National AI and Algorithm Expert Advisory Group (NAIAEAG), remained low, with many unsure whether the tools they use have been assessed. The gap between adoption and endorsement awareness is clear enough to prompt a larger national survey and more active communication of existing AI pathways to health professionals. (HiNZ.org.nz)

• Monash University Launches MAVERIC for Secure AI-Enabled Health Research

  Monash University has deployed MAVERIC, Australia’s largest university-based AI supercomputer, at CDC Data Centres’ Brooklyn campus in Melbourne, at a cost of AU$60 million. Built primarily with NVIDIA and Dell, the system operates as a Trusted Research Environment, keeping sensitive health datasets under Australian jurisdiction and governance through the Five Safes framework, enabling work at a scale and security level previously unavailable in Australian higher education. Health research projects are already underway, including skin cancer detection, epileptic episode prediction from video-EEG data with MRI, biomarker discovery for multiple sclerosis, and mental health AI model development. All current projects are at an early biomedical research stage, with no patient outcomes data published yet.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• a New Zealand-built AI cardiac imaging platform scaling across Australia's largest private cardiology network,

• what a 143,000-admission study reveals about predicting inpatient hypoglycaemia 24 hours ahead,

• and how to go about setting up Heidi Comms to handle routine patient calls.

Subscribe now

2. Case Study – Your Fortnightly Practical

Image Source: Heidi Health

Setting Up an AI Agent to Handle Routine Patient Calls

Case Presentation: This fortnight, the pressure is at the front desk rather than in Dr Harry’s clinic room.

Reception is dealing with a constant stream of calls: bookings and reschedules, results, and repeat prescriptions. Calls are missed over lunch and after hours, and patients tell Dr Harry they cannot get through. The practice manager raises it at the practice meeting.

They wonder whether an AI Voice Agent can handle routine calls, so reception can focus on patients in front of them.

Approach: Set up Heidi Comms - Heidi’s patient communication agent, to answer routine calls, texts and messages, book appointments, and pass anything clinical to your team.

1. Decide which patient enquiries Heidi will handle first

   Start with high-volume, low-complexity calls and messages: new patient enquiries, booking and rescheduling, opening hours and common questions, appointment reminders, and follow-ups for population health campaigns. Agree what must always reach a person, such as anything clinical or urgent.

2. Prepare your practice information and escalation rules
   Gather what Heidi needs to answer accurately: your services, hours, booking rules, and frequently asked questions. Heidi Comms does not give medical advice, so set clear rules for when a caller is transferred to reception.

3. Configure it with the Hendrix Health and Heidi team
   Heidi Comms is set up with you using call templates that define the goal of each call, the questions it asks, and how it confirms a caller's identity before going further, with the greeting and tone tailored to your practice. In New Zealand, it integrates with a variety of commonly used PMS tools and it can run without an integration, though some features depend on one.

4. Test it before going live
   Run practice calls through Heidi. Confirm that bookings land correctly in your PMS, that the handover to reception works, and that the written summaries read accurately. Adjust the flow until it fits how your practice works.

5. Review the call logs and keep a person in the loop

   Every call is transcribed and automatically summarised in Heidi, so the team keeps full visibility of what was discussed. Review these summaries regularly and follow up on anything flagged for follow-up.

Note: Heidi Comms is currently available to early adopters in New Zealand, with a small number of integrations live and more in progress. It currently operates in English. Practices interested in joining can arrange access through the Hendrix Health team.

Important: Heidi Comms handles communication and leaves clinical decisions to your team.

Outcomes: Instead of a front desk that drops calls when busy and falls silent after hours, the practice now has a first point of contact that answers every call, books appointments, and passes anything clinical to the team. Reception spends less time on repeat questions and more with patients.

For patients, it means reaching the practice when they need to, including outside opening hours. For Dr Harry, fewer interruptions reach his room, and patients arrive booked into the right appointment.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

NZ's HeartLab Rolls Out AI Cardiac Imaging Platform Across Australia's Largest Private Cardiology Network

HeartLab, an Auckland health-technology company, has begun a multi-year rollout of its cloud-based cardiac imaging platform across Advara HeartCare, Australia’s largest private cardiology network. Announced last week, the rollout will reach Advara’s full network of more than 100 clinics, 130 cardiologists, and 160 cardiac sonographers, across five states by the end of the year.

Image Source: HeartLab

When a patient presents with possible cardiac symptoms, they are often referred first for an echocardiogram. HeartLab’s platform stores and shares the study, supports measurement and reporting, and uses AI to speed analysis. Because it runs in a browser, clinicians can review imaging and report from any location, returning results to referrers and patients more quickly.

As its network grew, Advara needed a single platform to standardise reporting across metropolitan, rural, and remote sites.

Key Features:

• Cloud-based access: Clinicians can view studies and report from anywhere through a browser, with no software to download, supporting mobile testing and live review in remote areas.

• Imaging modalities: The platform stores, views, and shares echocardiography, CT, MRI, and angiography studies, replacing legacy systems at each site.

• Selection process: Advara chose HeartLab after a six-month evaluation against around 70 criteria, covering clinical workflow, usability, and reporting consistency.

• Scale of use: By company reports, the platform serves around 500,000 cardiac patients a year across New Zealand, Australia, the UK, and the US, with more than 1,000 studies processed daily.

• New Zealand footprint: Health NZ Te Whatu Ora has deployed HeartLab in one region so far, with the company expecting broader public-system uptake.

Implications for the Health System and Clinicians: This is an example of AI adding value through operational infrastructure rather than autonomous diagnosis, with gains from faster analysis, standardised reporting, and the ability to review imaging from any location. A New Zealand-built platform can meet the demands of Australia’s largest private cardiology network is also a notable signal for local health technology and cardiac services

The caveat is that reach and patient numbers are company-reported. The claims about reduced risk and mortality rest on faster image access and reporting turnaround, and on provider testimony, rather than independent clinical outcome data. Whether faster reporting translates to better patient outcomes, and whether New Zealand’s public system expands beyond one region, remains to be seen.

Predicting Inpatient Hypoglycaemia 24 Hours Ahead: What a 143,000-Admission AI Study Shows

A study in npj Digital Medicine built and prospectively tested an AI model that predicts inpatient hypoglycaemia up to 24 hours before it happens. A US team at three Cedars-Sinai hospitals in Los Angeles used 143,124 adult admissions from 2014 to 2025, all on a glucose-lowering medication.

The model, a long short-term memory network, learns from how each patient’s data changes over time. Every four hours it reads medications, lab results, diet orders, and meals eaten over the previous five days, then estimates the risk of blood glucose dropping below 3.9 mmol/L (70 mg/dL) within 24 hours. It uses ordinary fingerstick readings, not continuous monitors. Hypoglycaemia occurred in 19% of admissions.

Key Findings:

• Predictive accuracy: It flagged about 44% of the hypoglycaemic episodes that followed, and about one in four of its alerts were correct, better than the standard models tested.

• Held up live: That same performance held when the model ran in real time on daily records, a step most prediction tools never reach.

• Workable alert load: Across the three hospitals, this meant about 16 alerts a day, flagging around 3.6 of the 8 daily hypoglycaemia cases.

• Clear reasoning: It showed the factors behind each alert, often modifiable ones like recent insulin or prior lows, and worked evenly across most patient groups.

Implications for Healthcare Systems:
This shows AI making inpatient glucose care proactive rather than reactive, spotting at-risk patients a day ahead. For New Zealand wards, where insulin is a high-alert medicine and inpatient hypoglycaemia a known, preventable harm, a tool using data hospitals already collect is appealing. However, performance is still modest: it misses more than half of events, and most alerts are false. It was also built and tested at one Los Angeles health system over just two and a half weeks, with no trial yet showing it improves outcomes.

Read the full study here.

4. Vitals – Quick Bytes

• New Zealand to Lead 24,000-Patient ICU Trial of AI-Guided Oxygen Therapy

  A New Zealand-led randomised trial will test whether machine learning can improve survival for patients on life support, backed by a $5 million Health Research Council Programme Grant. The REVOLUTION trial will run across 50 ICUs in New Zealand and Australia and recruit more than 24,000 patients, billed as the first of its kind worldwide to evaluate AI-guided treatment on ICU survival. The team will use machine learning to personalise oxygen delivery, the most widely used ICU therapy, comparing AI-assisted oxygen targeting against standard clinician-led care. The model will first be refined on data from the completed Mega-ROX trial, which enrolled 40,003 patients across 137 ICUs in 14 countries. For now, the trial has only been funded and is yet to begin, so any survival benefit remains unproven.

• AI Chest X-ray Reading to Reach Every NHS Trust in England by 2029

  AI tools that act as a second pair of eyes for radiologists, flagging possible lung cancer on chest X-rays and pushing urgent scans up the queue, are now live in around half of England's NHS trusts and have supported faster diagnosis or all-clear for more than 4 million patients. Early NHS data shows scans now read in around four days on average, against eight previously for the most complex cases. The government has committed £20 million to extend the tools to every trust by 2029. These figures are government-reported, and the gains so far are in reporting speed - commentators caution that faster reading may not shorten overall diagnosis if downstream CT, biopsy and clinic capacity stays the bottleneck, with radiologists central to the read. (gov.uk)

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• how an AI tool now triages suspected skin cancer referrals across 26 NHS trusts,

• what an Australian trial reveals about predicting future heart failure from a routine echo with AI,

• and how you can find ready-made ACC report templates in Heidi’s Community Template Library.

Subscribe now

2. Case Study – Your Fortnightly Practical

Image Source: Heidi Health

Streamlining ACC Documentation with Heidi Templates

Case Presentation: Last time, Dr Harry helped his practice manager put a shared AI policy in place: a clear register of approved tools, rules about what patient information can go where, and a named owner to keep it current.

This fortnight, he is back to his own workflow. A patient with a work-related shoulder injury needs an ACC back-to-work assessment report before returning to a labouring role. He has completed a few of these this year, but not often enough to have the structure memorised: which headings ACC expects, which functional categories apply, how detailed the narrative needs to be.

Before starting another one from scratch, Dr Harry opens Heidi's Community Template Library to check whether a suitable starting point already exists.

Approach: Search the Heidi Community Template Library for pre-built ACC templates, save the relevant ones to your personal library, and adjust them before use.

1. Open the Community Template Library and filter for ACC templates

   Navigate to the Templates section in Heidi and select the Community Template Library. Apply both the “allied health professional” and “new zealand” filters to narrow results to ACC-relevant templates for Aotearoa. The filter returns 8 templates covering the ACC reports that carry the heaviest documentation load in allied health practice: back-to-work reports, stay at work assessments, pain management plans, and psychological services documentation.

2. Preview and save to your personal library
   Click into each template to review its structure and format before saving. Saved templates appear alongside your existing Heidi templates. You can save more than one and label them by form type for easy access.

Note: Templates in the Community Library are available on action credits in the free version of Heidi, or unlimited on any paid plan.

3. Adjust the template to your documentation style
   Community templates are starting points. Open the template settings to edit headings, adjust output instructions, or add practice-specific sections. If your ACC contract requires particular headings or response formats, build those in now. This avoids manual edits each time.

4. Use the template in your next ACC session
   Select the template at the start of your ACC-related consultation. As with all Heidi-generated notes, review the output before copying it into your PMS or filing with ACC.

Important: ACC documentation carries legal and contractual weight. Review every generated note before submission. You remain responsible for the accuracy and completeness of all forms filed under your name.

Outcomes: Instead of rebuilding report structure from scratch, Dr Harry now has a template already shaped around New Zealand's ACC documentation requirements. The layout is handled. He focuses on the clinical content.

For physiotherapists, psychologists, and OTs in ACC-contracted practices, the same library holds templates covering the forms that generate the most consistent overhead week to week. The 8 templates were built by Dr Sarah Manig, with more to follow as the library expands.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

AI Skin Cancer Triage Now Used Across 26 NHS Trusts

Bradford Teaching Hospitals NHS Foundation Trust joined 25 other trusts across the UK to use an AI tool named DERM (Deep Ensemble for Recognition of Malignancy), to triage suspected skin cancer referrals at St Luke's Hospital. Bradford is the first in West Yorkshire to adopt it.

Image Source: Med-Tech Insights

Patients referred by their GP enter a new teledermatology service, where staff photograph suspicious lesions with a dermatoscope attached to their phone lens, and DERM’s AI software returns a triage recommendation in seconds. Flagged lesions go to a consultant dermatologist in an adjoining one-stop clinic; benign cases are reassured and discharged.

Bradford receives around 5,000 skin cancer referrals a year but only 8% prove malignant, so DERM clears the benign majority off the two-week pathway to protect specialist time.

Key Features:

• Scale of adoption: DERM is currently deployed across 26 NHS trusts, including multiple teaching hospitals.

• Regulatory standing: NICE conditionally recommended DERM in May 2025 for three years while evidence is gathered, and an NHS England report found it could be used autonomously, with a negative predictive value of 99.8% versus 98.9% for dermatologist assessment.

• Efficiency and cost: Estimated savings of around £86 per case and a £1.70-2.30 return for every £1 spent, mainly from fewer face-to-face reviews and biopsies.

• One-stop pathway: A flagged lesion can be excised at the same visit, sparing many patients weeks of repeat biopsies.

• Built-in safeguards: Risk-based logic routes each lesion to the safest pathway, and Bradford’s first-year data will be independently evaluated by the regional Cancer Alliance.

Implications for the Health System and Clinicians: Since a 2022 pilot in London, DERM has seen 260,000+ patients and found close to 20,000 cancers. For Aotearoa, where melanoma rates are among the world’s highest, triaging the benign majority off urgent lists is the model to watch, and the MedTech company has signalled expansion into Australia.

However, DERM is still in an evidence-gathering phase: the NICE recommendation is conditional, with company-reported headline figures, and less certain performance for black and brown skin requiring extra clinician review. Whether autonomous triage holds up safely across diverse populations, and whether the promised community and GP rollout arrives, will decide if it becomes standard or stays a supervised pilot at scale.

Predicting for Future Heart Failure from a Routine Echo: What an Australian AI Trial Shows

A blinded, retrospective study in JACC: Heart Failure tested whether AI could turn a routine echocardiogram into a screen for future heart failure. The Australian-led team applied an AI risk algorithm built on the National Echo Database of Australia, to baseline echoes of 453 patients discharged in pre-symptomatic heart failure (stage A or B), then tracked who progressed to symptoms.

Prevention means finding at-risk patients before symptoms take hold, yet predicting who progresses is difficult. The algorithm scores risk from the measurements in an existing echo report, so it needs no new test.

Image Source: Journal of the American College of Cardiology (JACC Journals)

Key Findings:

• Accurate risk stratification: From echoes taken three or more years earlier, it predicted who would develop symptomatic (stage C) heart failure with an AUROC of 0.81 (1.0 is perfect), rated excellent.

• Predicted who would be admitted: It was nearly as accurate at predicting future heart failure admissions (0.79), and most accurate (0.83) at predicting either event, new heart failure or an admission.

• Beat the standard risk tool: Today’s standard calculator (PREVENT-HF) worked for only 54% of patients, needing inputs many lacked, and was less accurate (0.73); the AI scored nearly all and predicted better.

• Progression was common: Though all began symptom-free, within three years 1 in 5 (21.6%) developed symptomatic heart failure, and nearly 1 in 3 (29.8%) developed it or were hospitalised, despite specialist care.

Implications for Healthcare Systems:
For prevention, the appeal is clear: a score from echoes already being performed could help cardiology and primary care flag patients for earlier guideline-directed therapy, before symptoms appear. On the other hand, this was a small, single-centre, retrospective proof-of-concept not powered for these predictions, and funded by Echo IQ, the algorithm's developer, with several authors paid as consultants. Survival bias and advances in heart failure therapy since the original trial temper the results. A risk score is only the beginning; whether acting on it prevents progression cost-effectively is still unanswered.

Read the full study here.

4. Vitals – Quick Bytes

• Monash Health and Heidi Trial Real-Time AI Translation in Hospitals and Clinics

  Monash Health, one of Australia's largest health services, has partnered with Heidi to co-develop and trial a clinical-grade, real-time AI translation tool across its hospitals and clinics. The tool supports communication between patients, families and clinicians, initially covering 10 commonly spoken languages, in a network where 45% of patients speak a language other than English at home. It is built to hospital standards for privacy, security and clinical safety. Professional interpreters will continue to be used for complex care or sensitive conversations, with the trial testing where AI can safely support everyday communication rather than replace existing services.

• Sydney's START-AI Model Predicts Inpatient Admission in a 163,000-Case ED Study

  A study in Emergency Medicine Australasia used ensemble machine learning to extend START, an admission risk tool already used in New South Wales patient flow systems, into START-AI. Drawing on 162,915 adult ED presentations from a single Sydney tertiary hospital (27% admitted), the model combined the original START score with vital signs, blood tests, CT orders and free text from triage and clinical notes to predict who would be admitted. The full model predicted admission more accurately than START alone (AUROC 0.88 versus 0.77), and reached 0.93 for predicting ICU admission or death. The authors note this single-centre study only derived and internally validated the model, which is designed to support senior ED decision-making rather than replace it. Prototype development and prospective evaluation are still needed before clinical use.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• how an NHS trust used agentic AI to reduce routine IT help desk demand,

• research showing UK patients turning to AI chatbots before contacting their GP or NHS services,

• and how practices can set clear rules for controlled AI use that staff can actually follow.

Subscribe now

2. Case Study – Your Fortnightly Practical

Image Source: essert.io

Creating a One-Page AI Policy For Your Practice

Case Presentation: Dr Harry has been using Heidi for his consult notes for a few months. Other clinicians in the practice have started using it too.

The practice manager has noticed something else. One clinician is using ChatGPT to tidy up patient instructions. A nurse has asked another AI tool to rewrite an educational handout. Reception has used Google AI to draft a patient message about a missed appointment.

The issue is that everyone is making their own judgement call and it is unclear which tools are approved, what patient information can be entered, what patients need to be told, or who is responsible for checking the final output.

The practice needs a simple, shared rulebook that staff can actually follow.

Approach: Create a one-page AI governance policy that covers the basics: approved tools, patient information, consent, review, and ownership.

1. List the approved tools
   Start with a short table. Name the AI tools the practice allows, what each one can be used for, and who is allowed to use it. For example, a tool like Heidi may be approved for clinical documentation, while general AI tools may only be approved for non-identifiable admin writing.

2. Set the patient-information rule
   Be explicit. Staff should not paste patient names, NHI numbers, contact details, clinical histories, referrals, results, or correspondence into public AI tools. Identifiable patient information should only be used in approved clinical tools with the right privacy and security arrangements.

3. Include a consent script
   Give clinicians a simple sentence they can use every time. For example: “I am intending to use a secure clinical documentation tool to help draft my notes from the consultation. This allows me to focus on you during the consultation. Your audio is not stored, and I review and approve the final note myself. Are you comfortable with that?” The wording can be adjusted, but it should be consistent across the practice.

4. Make human review non-negotiable
   AI can draft a note, letter, patient message, or instruction sheet. It should not send, file, or finalise anything on its own. A clinician must check the output before it goes into the PMS or reaches a patient.

5. Give the policy an owner

   The practice manager should keep the AI tool register up to date. Agree on who will be the clinical lead responsible for clinical safety. Any new AI tool should be checked before staff start using it, not after it has already become a habit.

Outcomes: The practice moves from informal AI use to controlled AI use. Staff know which tools they can use, what information must stay out of public AI systems, how to explain AI use to patients, and who checks the final output.

For practice managers, this is a great starting point. Before adding another AI tool into the clinic, make sure the rules are clear enough that a busy GP, nurse, or receptionist would make the same decision on a different day.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

Rotherham NHS Trust Cuts IT Help Desk Calls by 28% with Agentic AI

The Rotherham NHS Foundation Trust has reduced IT help desk call volumes by 28% after introducing an AI-powered autonomous agent for routine IT support queries.

Image Source: netcall.com

Launched on 28 January 2026 with Netcall’s Liberty Converse platform, the agent acts as a first-line support route before staff reach the IT service desk. It interprets requests, retrieves relevant information from internal systems and troubleshooting guidance, resolves common issues in real time, and directs staff to the right support channel when needed.

For queries that still require IT input, the system can automatically raise a pre-populated service ticket with the correct categorisation and context. The trust says 41% of IT queries are now handled through self-service and AI agents, helping reduce pressure on phone lines and freeing IT teams to focus on more complex issues.

Key Features:

• Reduced call volume: IT help desk calls fell by 28% after the AI agent was introduced, easing pressure on the service desk.

• Routine query resolution: The agent handles common IT issues in real time using internal systems and troubleshooting guidance.

• Automated ticket creation: When escalation is needed, the system raises a pre-populated service ticket with categorisation and context already included.

• Continuous improvement: The trust analyses live staff interactions to refine accuracy, expand coverage, and improve the user experience.

• Next phase: Rotherham is now looking at out-of-hours support, where the agent could resolve or redirect non-urgent requests so only priority incidents reach on-call teams.

Implications for the Health System and Clinicians: For practice managers and health service leaders, Rotherham shows that healthcare AI does not need to sit directly in the consultation room to deliver operational value. Some of the fastest returns may come from applying AI to internal bottlenecks that quietly drain staff capacity every day.

Routine IT support is a high-volume, low-complexity demand. The caveat is that call reduction is not the whole story. The next questions are whether staff satisfaction, resolution quality, escalation safety, and total cost hold up as the agent expands into out-of-hours support. Still, this is a clean example of agentic AI being used for real operational work rather than another chatbot.

UK Study Finds Patients Are Turning to AI Chatbots Before Their GP

A new study from King’s College London has found that AI chatbots are already changing how people access healthcare.

The survey of 2,093 UK adults found that 15% have used AI chatbots for health advice instead of contacting a GP or other NHS service. 1 in 10 have used AI for mental health therapy or wellbeing support instead of seeing a trained professional.

The main reasons were convenience (46%), curiosity (45%), uncertainty about whether the concern was serious enough for a GP (39%), and NHS waiting times (25%).

Among those using AI for health advice, 20% said the AI had not encouraged them to seek professional advice, while 21% said they had decided against seeking healthcare because of something a chatbot said.

Image Source: kcl.ac.uk

Key Findings:

• AI is already substituting for some first-contact care: 15% used AI chatbots for health advice instead of contacting a GP or other NHS service.

• Mental health use is emerging: 10% used AI for mental health therapy or wellbeing support instead of seeing a trained professional.

• Access pressure is a driver: Convenience, curiosity, symptom uncertainty, and NHS waiting times were the most common reasons for using AI.

• Some patients are being steered away from care: 21% decided against seeking professional advice because of something a chatbot said.

• Public trust remains divided: 37% support AI being used in NHS clinical decision-making, while 38% oppose it, with women and younger adults more cautious.

Implications for Healthcare Systems:
Patients are already adding AI to the care pathway themselves, often before they speak to a clinician. In general practice, the risk is delayed care, AI-generated advice entering consultations, or patients assuming a chatbot can safely triage symptoms that need proper clinical assessment. For practice managers and clinical leads, this points to the need for clear patient messaging, safer signposting, and staff awareness that AI may already be shaping the consultation before the patient walks in.

Read the full study here.

4. Vitals – Quick Bytes

• Health NZ moves AI breast screening from scoping to procurement

  Previously, we covered Health New Zealand’s early scoping of AI support for BreastScreen Aotearoa. That work has now moved a step closer to implementation, with a new national tender for AI-supported mammogram reading and breast density reporting. Rather than replacing radiologists, the proposed system would sit inside the existing double-reading process, taking on one of the two mammogram reads while clinicians remain responsible for diagnosis and follow-ups. Validation is expected to begin from July 2026, including testing across Māori, Pacific and Asian mammogram datasets, before comparison with radiologist reading in a prospective trial. With screening demand rising, the real test will be safe validation, workflow integration, and equitable performance.

• Australian survey finds patients may accept AI summaries, if clinicians stay in the loop

  A mixed-methods Australian survey of 275 consumers found cautious support for AI-generated consultation summaries. Participants saw potential for AI to save time, improve access, and make summaries easier to read, but remained concerned about accuracy and data use. In a scenario task, they preferred the AI-generated summary over a clinician-written version for capturing key issues, empathy, and which summary they would prefer to receive. The trust condition was clear: 88% said they would be more comfortable if a doctor reviewed AI-generated summaries before sharing them, and nearly 91% said clinicians should remain responsible for accuracy. For practices, the lesson is simple: AI-assisted communication may be acceptable to patients when it is useful, clear, and visibly supervised.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• Health NZ's culturally grounded “BroPilot” rollout in Hauora Māori Services,

• how an AI sedation system performed against anaesthetists in a multicentre RCT,

• and how to set up your new Heidi Remote quickly.

Subscribe now

2. Case Study – Your Fortnightly Practical

Image Source: Heidi Health

Setting Up Heidi Remote for the First Time

Case Presentation: Last time, Dr Harry used Heidi Tasks to automate his follow-up to-do list.

This week, his practice asked him to take over the Friday rest home visits, covering 8 to 10 elderly residents. So far, he has been balancing his laptop on a bedside table or propping his phone against a fruit bowl. He also noticed some of his patients glance at his phone with skepticism even after giving consent.

He has just received Heidi Remote, Heidi’s new wearable microphone, and wants to set it up before next Friday.

Approach: Pair Heidi Remote to the Heidi app, wear it during the round, and let audio sync automatically.

1. Confirm compatibility and charge
Heidi Remote pairs with the Heidi iOS, Android, and Desktop apps (not yet the web browser version). Use the supplied charging cable rather than a fast charger. A full charge takes about 1.5 hours and gives 14 hours of use.

2. Power on, pair, set a default template.
Hold the Power button for 2 seconds until the LED turns solid white. Keep the device within arm's reach of your phone, then in the Heidi app go to Settings, then Add Device, and follow the prompts. Set a default note template (for example, a rest home progress note) that all synced sessions will use.

3. Wear it and record
Clip Heidi Remote high on your chest or top pocket, facing outward. At each bedside, gain consent, then press the Record button once to start (LED red), again to pause (orange), and hold for 2 seconds to stop and save (white).

Important: Heidi advises keeping the device away from anyone with implanted pacemakers or defibrillators (magnetic components).

4. Let it sync
Out of Wi-Fi range, Heidi Remote keeps recording offline and stores audio on-device (32GB, over 100 hours of capacity). Back in range, sessions sync automatically into the Heidi app. Link each session to the resident and review the notes.

Note: Heidi Remote is in Early Access, so specifications may change. Heidi classifies it as an audio peripheral, with the clinician remaining responsible for clinical accuracy, consent, and data compliance.

Outcomes: Dr Harry's Friday visits run without a laptop or a propped-up phone. He clips Heidi Remote to his collar, presses one button at each bedside, and stays present with the resident.

By the time he reconnects at the practice, audio is syncing. He links each session, reviews, and copies into his PMS. He drives back to the practice with the documentation already done.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

Hauora Māori Services Customises Microsoft Copilot into "BroPilot", Grounded in Tikanga Māori

Health New Zealand has adapted Microsoft 365 Copilot into a culturally grounded AI tool called BroPilot, built within Hauora Māori Services and shaped around tikanga Māori. The programme used Dame Naida Glavish's Tikanga Best Practice Guidance as its foundational document. More than 200 staff with knowledge of Māori tikanga contributed to developing 16 measurable standard operating procedures grounded in Māori culture and values.

Image Source: news.microsoft.com

The rollout sits under HealthX, Health NZ's programme for scaling digital and AI solutions, which began in August 2025. Since January 2026, more than 2,050 Copilot licences have been issued, with deployment sequenced to leadership and digital teams first.

Adoption is supported through twice-weekly drop-in sessions where staff bring real work. Within BroPilot, kaimahi have built their own role-specific AI personas, such as Smartmāori for research, Nanny for kaupapa advice, Coco for day-to-day tasks, and Doc for executive-level documents.

Key Features:

• Cultural foundation: 16 standard operating procedures developed with input from over 200 staff with Māori tikanga knowledge.

• Practical adoption model: Drop-in sessions every Monday and Wednesday let staff bring real work into a secure enterprise environment.

• Personalised AI personas: Staff build their own agents within BroPilot for specific functions, from research and kaupapa advice through to admin and document drafting.

• Workflow applications: Reported uses include summarising long documents, extracting actions, clarifying responsibilities, improving report structure, and turning dense material into clearer outputs.

• Scale of rollout: Over 2,050 Copilot licences issued since January 2026 under HealthX, with deployment continuing across an organisation of more than 90,000 staff.

Implications for the Health System and Clinicians: For clinicians and practice managers watching AI rollouts in Aotearoa, BroPilot is a useful local case study in how adoption tends to succeed or fail on the human side, rather than the technology. The combination of cultural grounding, enterprise governance, and learning through real work shows what a thoughtful rollout can look like inside a complex public health system. We look forward to seeing outcomes data on productivity, retention, and clinical or administrative workflow.

AI-Driven Anaesthesia System for GI Endoscopy: a 418-Patient Multicentre RCT from China

A multicentre randomised controlled trial in npj Digital Medicine tested an AI system that automates sedation during gastrointestinal endoscopy. Known as AAS-GE, this automated anaesthesia system uses reinforcement learning to adjust the sedative drug, ciprofol, in real time based on patient vital signs.

The trial ran across five hospitals in China between January and August 2025, enrolling 418 adults with low anaesthetic risk, aged 18 to 65 undergoing routine GI endoscopy. Half received AI-controlled sedation, the other half clinician-managed anaesthesia. The primary safety outcome was hypoxaemia, defined as oxygen saturation < 92% at any point during the procedure.

Image Source: npj Digital Medicine

Key Findings:

• Non-inferior safety: Hypoxaemia rates were almost identical between groups (14.42% vs 14.29%), with no significant differences in severity or the lowest oxygen levels recorded.

• Faster induction without higher drug use: The AI group reached anaesthesia depth significantly faster (1.55 vs 1.90 minutes), with no significant differences in total ciprofol doses or recovery duration.

• More intraoperative body movement: Body movement was significantly more frequent in the AI group (17.79% vs 6.67%). Affected patients received lower maintenance doses, suggesting the AI kept patients in a slightly lighter state of sedation than clinicians would.

• No difference in other adverse events: Rates of swallowing, jaw lifting, coughing, vasopressor use, and supplemental oxygen were similar between groups.

Implications for Healthcare Systems:
This trial moves AI a step beyond decision support and into autonomous clinical delivery. The AI matched anaesthetists on safety while being faster, which is meaningful given global anaesthetist shortages and rising procedural volumes. However, it is important to note that the model was trained on data from a single hospital, and the more frequent body movement suggests it dosed more cautiously than clinicians would. Human-in-the-loop oversight was mandatory throughout, with anaesthetists retaining full override authority.

Read the full trial here.

4. Vitals – Quick Bytes

• Te Aho o Te Kahu Pilots AI-Assisted Pathology Reporting in Three NZ Labs

  An AI-assisted Structured Anatomical Pathology (ASAP) application went live with Pathologists at Waitematā, Christchurch, and Dunedin laboratories as part of New Zealand's national Cancer Control Agency’s CanShare programme. ASAP uses AI to draft pathology reports and tag key clinical details with SNOMED CT, a standard coding system that makes cancer information easier to share and reuse across the health system. The pilot is testing how well ASAP fits into pathologists' day-to-day work, the quality of the reports, and how quickly coded results can be shared with treating teams. Delivered with Health NZ Te Whatu Ora, AWS, and Deloitte, the pilot runs for two months and will inform whether AI-assisted structured reporting has a wider role in Aotearoa.

• Robotic Phlebotomy Achieves 94.5% First-Stick Success in Dutch Multicenter Trial

  A multicenter trial evaluated an autonomous robotic phlebotomy device (Aletta, by Vitestro) across three Dutch hospitals, enrolling 1,633 patients for routine outpatient blood draws. The device uses near-infrared imaging, ultrasound, and Doppler to locate a vein, then performs the venipuncture itself. First-stick success was 94.5% overall, holding up in patients with difficult venous access (92.7%), obesity (97.4%), and those aged 65 or older (93.4%). Adverse events were mild and uncommon (0.6%), and 82% of surveyed patients said they would prefer the device or had no preference over manual phlebotomy.
  Caveats: the trial was manufacturer-funded, several authors are company employees, the routine-use cohort had no manual control group, and 6% of patients were screened out because no suitable vein could be identified.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• Whakarongorau Aotearoa's new AI mental health Welcome service,

• what a 1000-people RCT reveals about conversational AI compared with face-to-face group therapy,

• and how you can automate your daily to-do list with Heidi Tasks.

Subscribe now

2. Case Study – Your Fortnightly Practical

Video Source: Hendrix Health

Automate a To-do List Using Heidi Tasks

Case Presentation: The admin pile has been getting smaller for Dr Harry. Using Heidi, his volume of leftover work at the end of each day has noticeably dropped.

Today, he wrapped up a busy morning clinic. Across 15 patients, there were referrals to make, repeat bloods to order, results to chase, and a couple of patient callbacks he had promised to do himself. He had a mental list, a few sticky notes, and a vague sense that he had probably forgotten something.

Dr Harry wonders whether Heidi can help him track these action / to-do items consistently, without relying on memory or paper.

Approach: Capture follow-up actions automatically from each consultation with Heidi Tasks, then review and manage them from a single “Master” list across patients.

1. Open the Tasks panel during or after a session
Click the Tasks button at the bottom right of your Note to open the Tasks panel on the side. Heidi will have already identified to-do items from the transcript and listed them automatically. You can adjust how the panel behaves and turn auto-generation on or off in your settings.

2. Review what Heidi has captured, organised by category
Each task is auto-categorised into one of five categories. Order covers things like blood tests or imaging requests. Coordinate covers referrals to specialists or allied health. Communicate covers patient callbacks, results delivery, and reminders. Document covers letters, summaries, or reports that still need to be drafted. Action covers general tasks that don't fall neatly into the above categories. You can edit a task title, change its category, or add a manual task by clicking "new task" at the bottom of the list.

3. View the evidence or generate a document directly from a task
Click the three dots button on the right of any task. From there, you can view the task evidence (which links you to the exact point in the transcript where Heidi picked up the task), or generate a document based on the task in one click. Heidi will automatically mark the task as Done once the document is ready. This can be undone if needed. As always, as the clinician you remain responsible for reviewing the evidence and confirming clinical appropriateness before actioning any task or sending any document.

4. Manage outstanding to-dos through the “Master” Task List
Navigate to your “Master” Task List by clicking the Task button on the very left. This list shows every outstanding to-do across all patients and sessions in one place, so you can work through your tasks at the end of the day without scrolling through individual notes. Tick items off as you complete them, or delete tasks that no longer apply.

Note: Heidi Tasks is currently in a phased beta rollout. Enable it by turning on Heidi Labs in Preferences.

Outcomes: With Heidi Tasks, every referral, test order, callback, and outstanding letter is already captured and categorised by the end of his morning clinic. Dr Harry spends the last 20 minutes of his day working through the Master Task List, rather than re-reading every note to remind himself what he meant to do. The risk of a forgotten test result or an unsent referral is significantly reduced.

This is very helpful for a busy GP managing dozens of small to-dos across a full day of patients.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

Whakarongorau Aotearoa Launches AI Welcome Service for Mental Health and Family Harm Helplines

Whakarongorau Aotearoa (New Zealand Telehealth Services) will launch a Microsoft Azure AI-powered Welcome service in May 2026 to support people contacting its helplines, while they wait to connect with a trained counsellor. The service will first deploy across SMS to the national 1737 Need to talk? helpline, and webchat for the Whakarongorau-run Women's Refuge service.

Image Source: geneticliteracyproject.org

Whakarongorau handles more than 7,000 interactions daily across health, mental health, and social services, reaching over 735,000 people in Aotearoa annually. Mental health demand is rising sharply: at-risk contacts have grown 153 percent since 2019, and the average session now takes around 50 percent longer.

When someone makes contact, the Welcome service identifies itself as AI and lets the user choose to engage or wait in silence. It gathers contact details and the reason for contact, checks in on how the person is feeling, and offers non-clinical support such as breathing techniques. By the time a kaimahi (staff member) joins, they have context to focus on care from the first message.

Key Features:

• Clear AI disclosure: Users are told upfront they are interacting with AI and can choose to engage or wait in silence.

• Non-clinical scope: The service does not provide clinical guidance, diagnosis, or medical advice, using empathetic language to hold the person until a counsellor is available.

• Context handover: Counsellors join the conversation with contact details, reason for contact, and emotional state already captured, reducing time spent on basics.

• Co-designed and tested: Developed with clinical oversight and frontline staff input, then tested with mental health service partners and people with lived experience.

Implications for the Health System and Clinicians: For New Zealand’s mental health and crisis support sector, this is a practical example of AI absorbing pre-clinical work that traditionally consumes counsellor time at the start of each interaction. The narrow, non-clinical scope to use AI to keep people engaged while they wait, is the key design choice.

One caution is that the service has not yet launched and no outcomes data is available. The real test will be whether it holds up safely across diverse users in distress, and whether wait-time experience and counsellor workload measurably improve once live.

Conversational AI in Mental Health: What a 995-Student RCT Reveals

A randomised clinical trial published in JAMA Network Open compared a conversational AI mental health platform with face-to-face group therapy and a waiting list control among 995 university students reporting psychological distress in Israel.

Participants were assigned to a 12-week AI platform (Kai, KAI.AI Inc), 12 weekly 90-minute group therapy sessions led by licensed psychologists, or a waiting list. The AI drew on multiple therapy and mindfulness techniques, with users encouraged to engage at least 3 times weekly. Those with active suicidality, current psychotherapy, or severe mental disorders were excluded.

Image Source: Journal of the American Medical Association (JAMA)

Key Findings:

• Greater anxiety reduction with AI: The AI group showed significantly greater anxiety reduction than both group therapy and the waiting list. Group therapy did not differ significantly from control group.

• Clinical transition rates: Among participants meeting baseline clinical anxiety thresholds, 57.9% in the AI group moved into the non-clinical range, compared with 14.4% in group therapy and 9.8% in the control group.

• Depression, well-being, and life satisfaction: All three outcomes favoured AI over the waiting list at 12 weeks and 3 months, with smaller and less consistent differences between AI and group therapy.

• No effect on PTSD: No significant differences emerged between any group at either time point, suggesting the platform did not address trauma-related symptoms.

• Therapy-seeking intention fell with AI: Intention to seek therapy declined from 29.8% to 23.5% in the AI group, remained stable in group therapy, and rose to 37.0% in the control group.

Implications for Healthcare Systems:
The results suggest conversational AI may have a role as an early-intervention tool for mild-to-moderate distress, particularly where access to face-to-face care is challenged. Here are some limitations: outcomes were self-reported, participants were university students in one country, attrition reached 35% at 3 months, and PTSD was unaffected. The drop in therapy-seeking intention warrants caution, as it may reflect genuine recovery or unmet need that goes unescalated.

Read the full trial in JAMA here.

4. Vitals – Quick Bytes

• Victoria Expands Virtual Hospital Pilot to 400 Patients

  The Victorian Government is expanding its first Virtual Hospital pilot, jointly led by The Royal Melbourne Hospital and Austin Health. For heart failure and post-cardiac surgery patients, the programme uses video, phone, and remote monitoring to deliver hospital-level specialist care at home and in regional hospitals through virtual wards and ward rounds; and a virtual foetal medicine service with the Royal Women's Hospital. There is also a planned neurology programme in partnership with Northeast Health Wangaratta. Since December 2025, more than 260 patients have been treated and over 1,000 hospital bed days saved, with the pilot now scaling to 400 patients by the end of June. The figures come from early government announcements rather than independent evaluation, with a formal review scheduled for June 2026. (premier.vic.gov.au, thermh.org.au)

• Frontier LLMs Still Stumble on Differential Diagnosis, Large Benchmarking Study Finds

  A cross-sectional study in JAMA Network Open tested 21 off-the-shelf large language models on 29 standardised clinical vignettes from the MSD Manual, scoring 16,254 responses in total. This included GPT-5, Claude 4.5 Opus, Gemini 3.0, and Grok 4. Using a new multidimensional benchmark called PrIME-LLM, the authors found that reasoning-optimised models scored higher than non-reasoning ones, and all models did best on final diagnosis and management. However, failure rates for differential diagnosis were over 80% across every model, while final diagnosis failure rates stayed below 40%. The authors conclude that current LLMs jump to single answers too quickly, and cannot yet be trusted for unsupervised, patient-facing decisions. The caveat is that the study tested base models only. Tools that retrieve from trusted sources (such as Heidi Evidence) were not tested, which may lift real-world performance.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• how air medical helicopter teams uses AI to speed up patient transfers,

• what we know about AI diabetes coaching programme from an RCT,

• and how you can add patient context in a scribe session to improve the output.

Subscribe now

2. Case Study – Your Fortnightly Practical

Video Source: Hendrix Health

Add Context to Support Continuity of Care

Case Presentation: Dr Harry has come a long way. He has built a dedicated ADHD assessment template, used Ask Heidi to generate ad-hoc referral letters and support documents, reviewed trusted sources and guidelines before complex consultations, and managed multilingual consults with patients and their whānau, all within Heidi.

Today, he is seeing a patient for a f/u appointment. The previous visit generated a detailed note covering initial presenting complaint, history, developmental background, and a management plan. At the end of f/u consultation, Dr Harry found himself manually entering details from the prior visit to complete today's note.

He wonders whether there is a way to bring that past clinical context into the current session so that Heidi’s output reflects the full clinical picture, not just what was discussed today.

Approach: Use Heidi's context window to load relevant information from previous visits, so that the generated note is more targeted and clinically complete from the outset.

1. Open the context tab before starting the session
At the top of a new Heidi session, the default tab is the note tab. To the very left of that is the context tab. Open this before transcribing the consultation.

2. Load the relevant history into Heidi and label it clearly
Copy and paste the relevant past notes from your PMS or EHR into Heidi’s context window. Add a clear label directly above the pasted text so that Heidi understands this information is from a previous encounter and not the current one.

For example: "History from last visit: [paste prior notes here]".

Another option is uploading individual files directly from your computer by clicking the paper clip icon at the bottom left corner.

Note: You can also include other relevant background information, such as letters returned from a specialist and radiology or pathology reports. This could also be ambulance notes in an ED context.

3. Transcribe the consultation as usual
Gain consent and transcribe the visit normally. The note that Heidi generates will take into account everything in the context window alongside what was captured in the transcript. The result is a note that reflects continuity of care rather than treating the visit as a single, isolated session. You will still need to review the final note for accuracy before copying it into your PMS or EHR.

4. Keep consent in mind before loading any context
For patients known to you who have previously consented, load the prior context before the session as outlined above. However, if a patient known to you withdraws consent at the start of a f/u appointment, delete the session entirely. This will also remove any context already uploaded, respecting their decision.

For new patients, gain consent before using Heidi at all. If consent is confirmed, you can add the background clinical information into the context window after the consultation instead.

Important: Do not use the context window if the patient declines the use of an AI scribe.

Outcomes: By using the context window, Dr Harry no longer needs to manually enter a patient's history or cross-reference multiple documents. His notes already reflect the patient's relevant history, are more consistent across visits, and require fewer edits before being copied into his PMS.

For patients like his ADHD follow-ups, where continuity across multiple long appointments matters, this makes a meaningful difference to both documentation quality and admin time after the consultation.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

AI Stroke Detection and Care Coordination Now Live Across 2,000 US Hospitals

Viz.ai, an AI-enabled stroke detection and coordination tool, is now in use across 2,000 hospitals in the United States. The system analyses CT scans within seconds, and when a suspected large vessel occlusion is identified, sends an alert to the treating specialist’s smartphone. This removes the manual chain of phone calls, image sharing, and delayed handovers that traditionally adds critical minutes between a scan being taken and a decision being made.

Image Source: Viz.ai

At regional hospitals that transfer stroke patients to larger centres for treatment, the system automates care-team activation and streamlines transfer coordination. The treating specialist can view the imaging and begin coordinating care within minutes of the scan being completed.

The platform has also been extended into air-medical transport. A Life Flight module, deployed as a pilot at the University of California, San Diego, allows helicopter dispatch and flight crews to join the same secure communication thread as stroke and surgical teams. This gives flight crews access to patient details, medications, and imaging findings before landing at the referring hospital, reducing the need for repeated handovers on arrival.

Key Features:

• Faster transfers: At Adventist Health & Rideout, a regional stroke centre in California, the time from patient arrival to transfer out was reduced by 44%, bringing performance in line with American Heart Association benchmarks

• Earlier detection: Time from scan completion to identification of a vessel occlusion fell by 84%, with specialist notification dropping from 45 minutes to 7 minutes

• Low adoption burden: Life Flight teams were trained on the new module through a single 15 to 20 minute session covering workflow steps, device settings, and test activations, with no additional training required for the existing stroke team.

• Clinician satisfaction: Overall satisfaction with the communication process improved by approximately 60% across all roles, with 100% of respondents preferring the new workflow

Implications for the Health System and Clinicians: For health systems that transfer stroke patients between regional hospitals and specialist centres, this is a practical example of AI reducing delays by improving how teams communicate, rather than by making clinical decisions. The value is in removing bottlenecks that slow treatment when minutes matter most.

The caution is that the strongest results come from early, single-site evaluations. Whether these gains hold consistently across different hospital networks and patient populations will need longer-term evaluation at scale.

AI-Powered Diabetes Prevention: What We Learned

On 3 April 2026, a Johns Hopkins-led phase III randomised controlled trial was presented at the NIH Pragmatic Trials Collaboratory's Grand Rounds. Published in JAMA in late 2025, the trial evaluated whether referral to a fully automated, AI-powered Diabetes Prevention Program (DPP) was non-inferior to referral to a human coach-led DPP among 368 adults with pre-diabetes and overweight or obesity across two US clinical sites, followed for 12 months.

The AI-led DPP used a reinforcement learning algorithm delivered via a mobile app and Bluetooth-enabled scale. The human-led DPP comprised four CDC-recognised programmes with the highest designation for effectiveness, delivered remotely.

Image Source: Journal of the American Medical Association (JAMA)

Key Findings:

• Non-inferiority confirmed: 31.7% in the AI-led group and 31.9% in the human-led group achieved the primary composite outcome, which included weight loss, physical activity, and HbA1c targets at 12 months. A risk difference of -0.2%, well within the pre-specified non-inferiority margin of -15%.

• Higher initiation and completion with AI: 93.4% of those referred to the AI-led DPP actually began the programme compared with 82.7% in the human-led group, and 63.9% completed it compared with 50.3%.

• Different engagement patterns: Among those who achieved the outcome, 74% were programme completers in the AI group compared with 56% in the human-led group, suggesting the AI programme depends more on sustained app engagement to deliver results.

• No study-related adverse events, with 85.1% of participants retained at 12 months.

Implications for Healthcare Systems:
AI-based programmes can run on demand and are not constrained by the scheduling, staffing, and capacity limitations that restrict traditional DPPs. The important caveats are that participants were motivated volunteers with high educational attainment from only two sites, and the trial measured surrogate outcomes rather than diabetes incidence. The value of the AI-led DPP appears to be accessibility and reach rather than superiority over human coaching.

Read the full trial in JAMA here.

Watch the Grand Rounds session on AI for Diabetes Prevention here.

4. Vitals – Quick Bytes

• Western Australia Funds AI Pilot to Improve Hospital Workflow and Efficiency

  The Western Australian Government is investing A$700,000 in an AI pilot at Royal Perth Hospital to predict length of stay and streamline hospital workflows ahead of the 2026 winter season. The system will schedule tasks including medical imaging, laboratory tests, pharmacy medication packs, and discharge summaries, with the aim of improving bed availability and reducing delays. A new system-wide data platform will feed live dashboards for the State Health Operations Centre to track demand, capacity, and patient movement across the public health system. If successful, the pilot may be scaled across multiple hospitals and integrated with other digital health initiatives.

• US Survey Finds Public Openness to AI in Healthcare is Declining

  A national survey of 1,007 US adults commissioned by The Ohio State University Wexner Medical Center found that only 42% are open to AI being used in their healthcare, down from 52% in 2024. The belief that AI can improve healthcare efficiency also fell, from 64% to 55%. At the same time, 51% of respondents reported using AI to make a health decision without consulting a medical professional, and 62% use AI to understand symptoms before deciding whether to seek care. Researchers noted the decline is consistent with the natural technology hype cycle and expect trust to stabilise as the true use cases for AI in healthcare become clearer. (medicalxpress.com)

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• expansion of AI-assisted hospital room technology across the US,

• what a large RCT tells us about AI in stroke care decision support,

• and how you can make the most out of Heidi’s functions when you’re speaking multiple languages during a consultation.

Subscribe now

2. Case Study – Your Fortnightly Practical

Video Source: Hendrix Health

Managing Multilingual Consults with Heidi

Case Presentation: Last time, Dr Harry used Heidi Evidence to review trusted sources and guide his approach to a clinical question. It is quick, reliable, and completely ad-free to get the answers he needs.

Today, he is seeing a patient whose mother speaks only Cantonese, while the daughter speaks English. While being fluent in both English and Cantonese, Dr Harry wants to communicate clearly with both patient and whānau, but without doubling his workload by manually translating explanations or writing separate documents.

He considers whether Heidi can support multilingual transcription and communication within the same session.

Approach: Changing Heidi’s input language settings to capture multilingual consults and using Ask Heidi’s translation capability generate patient-friendly explanations in multiple languages.

1. Set input languages before starting the consult
Adjust Heidi’s input settings to match the languages spoken in the room, for example English and Cantonese. This allows Heidi to process mixed-language conversations without breaking workflow or requiring separate documentation. Heidi can listen for up to 3 languages when transcribing. For example, English, Te Reo Māori, and Cantonese.

2. Conduct the consultation naturally
Allow the conversation to flow between you, the patient, and family members in their preferred languages. There is no need to force everything into English, as it reduces accuracy and rapport.

Heidi captures the consult and generates a structured note in English (or any output language you select) for clinical documentation.

3. Generate a patient-friendly explanation
After completing the consult, click the Ask Heidi bar at the bottom of your screen:

“Create a simple patient explanation of the diagnosis and management plan in English.”

This ensures clarity, especially for complex conditions like ADHD or mental health presentations.

4. Translate for the patient and whānau
Follow-up in Ask Heidi with: ”Translate this explanation into [language of your choice].”

Then, review the output before providing both versions of the patient explainer to the patient and their family.

Important: Do not rely on this workflow if you cannot verify the translation yourself. You remain responsible for accuracy and clinical safety in any case.

Outcomes: Dr Harry is now able to manage multilingual consults without slowing down or compromising communication quality. Instead of fragmented conversations and duplicated admin, he delivers clear, structured explanations in both languages within minutes. This improves patient understanding, supports whānau involvement, and reduces the risk of miscommunication.

Dr Harry maintains a single, efficient workflow while adapting to the diverse patient populations in Aotearoa New Zealand.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

AI-Assisted Hospital Rooms Expand Across 25 Hospitals (WVU Health System)

WVU Health System has begun rolling out AI-assisted “intelligent hospital room” technology across all 25 hospitals in its network, covering more than 3,000 beds across West Virginia, Ohio, Pennsylvania, and Maryland. The rollout is delivered through a partnership with hellocare.ai, and is one of four new partnerships the AI company has announced with US health systems in 2026.

Image Source: Hellocare.ai

When a patient is admitted, bedside devices, video systems, and monitoring tools connect into a single virtual care platform that is integrated with the Epic EHR. The platform combines AI, secure video collaboration, real-time communication tools, and clinical workflow integration to support patient observation, fall prevention, remote consults, and communication between care teams, patients, and families.

Virtual care teams, including remote nurses, can then support bedside staff through virtual nursing workflows, while clinicians and specialists can also join care remotely through virtual rounding and telehealth.

Key Features:

• Virtual nursing support: Remote nurses support bedside teams through virtual nursing workflows and communication.

• AI-assisted patient monitoring: Continuous observation may help detect deterioration earlier and supports fall prevention.

• Integrated communication: Video, messaging, and EHR integration allow easier communication between care teams, patients, and families.

• Virtual rounding and telehealth: Specialists can review and contribute to care without being physically on-site.

• Large-scale rollout: Deployment across 25 hospitals and 3,000+ beds signals standardised infrastructure rather than pilot-stage experimentation

In some hospitals, additional features such as digital whiteboards, door signage, and patient infotainment systems are also being added.

Implications for the Health System and Clinicians:
This reflects a broader shift in how inpatient care is organised, separating physical presence from clinical contribution. Tasks that do not require hands-on care, such as monitoring, communication, and routine support activities, can now be delivered remotely.

For clinicians, the potential value is better communication, more support with monitoring, and less pressure on bedside teams. For health systems, the appeal is clear: improving patient safety and workflow efficiency without relying only on traditional staffing models that are vulnerable to shortages.

The rollout is large and the platform is clearly integrated into hospital workflows, but there is no published data on staff and patient outcomes. The real question is whether these benefits hold up in practice.

AI Stroke Decision Support: What a 21,000-Patient Cluster RCT Reveals

A large cluster-randomised clinical trial in The BMJ evaluated an AI-based clinical decision support system (CDSS) for acute ischaemic stroke across 77 hospitals in China, involving 21,603 patients admitted within seven days of symptom onset.

The system combined AI-assisted imaging analysis, classification of stroke cause, reminders for pending assessments, and evidence-based treatment recommendations during admission and at discharge.

Image Source: the BMJ

Key Findings:

• Fewer new vascular events at three months: The primary outcome, a composite of ischaemic stroke, haemorrhagic stroke, myocardial infarction, or vascular death, occurred in 2.9% of patients in CDSS hospitals versus 3.9% in usual care hospitals, a relative risk reduction of about 26%.

• Improved stroke care quality: The composite adherence score for evidence-based stroke care was higher in the intervention group, 91.4% versus 89.8%. The biggest gains were seen in secondary prevention and inpatient care measures, including dual antiplatelet therapy, anticoagulation for atrial fibrillation, dysphagia screening, and DVT prophylaxis, where baseline adherence was lower.

• Benefits persisted over 12 months: New vascular events remained significantly lower at six months, 3.4% versus 4.8%, and at 12 months, 4.0% versus 5.5%, driven primarily by reductions in recurrent ischaemic stroke.

• No clear difference in disability, mortality, or bleeding: Despite fewer recurrent vascular events, the trial did not find significant differences in disability, all-cause mortality, or moderate to severe bleeding, which suggests the system improved secondary prevention without increased safety risk.

Implications for Healthcare Systems:
AI-based decision support can improve stroke care quality and reduce recurrent vascular events when embedded directly into routine workflows, particularly in time-critical, protocol-driven pathways like stroke. A CDSS can help clinicians follow guidelines more consistently and reduce variation in care at scale. The caution is important: this study was conducted at the hospital level rather than individual patients, in a population where most strokes were relatively mild, and it did not assess cost-effectiveness or include thrombectomy decision-making. As such, further validation across different health systems and patient populations will be needed before widespread adoption.

Read the full study here.

4. Vitals – Quick Bytes

• New Zealand’s National Diabetes Roadmap Signals AI-Enabled Screening

  Health New Zealand has launched a National Diabetes Roadmap that includes exploring AI in diabetic retinopathy screening and piloting retinal photo screening in primary care settings. The strategy also aligns HbA1c thresholds with international standards to support earlier detection and intervention. Alongside workforce expansion and community-based care, this signals a shift toward more technology-enabled diabetes management at a national level. However, this remains policy direction rather than proven deployment. The takeaway is simple: AI is now part of the national plan, but impact will depend on whether these pilots translate into real-world reductions in complications.

• Studies on AI-Enabled Stethoscopes Suggest Better Detection, But Implementation Limits Impact

  Two recent studies examined whether AI-enabled stethoscopes can improve detection of cardiovascular disease in primary care. In a prospective study of 357 at-risk patients, AI-augmented digital auscultation substantially improved sensitivity for moderate-to-severe valvular heart disease compared with standard auscultation, detecting 12 previously undiagnosed cases versus 6 with routine care, though with lower specificity.

  Meanwhile, the larger TRICORDER cluster-randomised implementation trial across 205 UK primary care practices found no significant increase in population-level detection of heart failure, atrial fibrillation, or valvular heart disease after 12 months, largely because real-world uptake was inconsistent and use declined over time. Taken together, it is clear that the technology itself performs well. But to deliver real clinical impact also relies upon workflow integration, sustained clinician use, and access to follow-up testing and specialist care.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• how Patchwork Health’s AI rostering tool helps NHS hospitals reduce shift gaps and agency spending,

• AI-prediction of breast cancer risk after an all-clear mammogram,

• and how you can use Heidi Evidence to quickly review research and guidelines.

Subscribe now

2. Case Study – Your Fortnightly Practical

Video Source: Hendrix Health

Using Heidi to Find Evidence

Case Presentation: Dr Harry have been using Ask Heidi to turn his completed ADHD assessment notes into ad-hoc documents when he doesn’t have a template, and it’s been working well.

Today, he is about to see a new patient who describes a long history of inattention, disorganisation, and poor follow through dating back to school, but the presentation is not straightfoward. They are now struggling at work, sleeping poorly, feeling anxious, and drinking alcohol regularly.

Dr Harry is not sure whether this is uncomplicated adult ADHD or another condition presenting in a similar way. He does not want to rely on memory alone or waste time seraching across multiple sources while the patient is in the waiting room.

He wants a fast way to review quality evidence, sense-check the differential, and identify what else he should ask before making any decisions.

Approach: Use Heidi Evidence when the clinical picture is unclear and you want a faster, more efficient way to review trusted sources, including local guidelines and research studies.

1. Enter the clinical question in plain language
Start by typing the scenario into Heidi Evidence as you would naturally think about it in practice. For example:

“I’m about to see a 25M, long history of inattention and disorganisation now affecting work, some anxiety, poor sleep, regular EtOH use, ?ADHD ?something else, what should i think about patient is in waiting room”

You do not need to write a perfect academic search query. A clinical shorthand is usually enough for Heidi Evidence to return a useful starting point.

2. Review cite sources and key points from Heidi Evidence
Heidi Evidence can help explore alternative explanations and common overlaps, in this case anxiety disorder, depression, sleep disorders, substance use, autism spectrum, and medical contributors. This is especially useful in cases where the symptoms are non-specific and the differentials are not yet clear.

If you use Heidi Evidence to review evidence, for example by asking, “What is the latest research on psychedelics in mental health treatment?”, Heidi Evidence would often generate a summarised answer with key points at the top, followed by a more detailed response. Each point is linked to its source, so you can quickly see where the information is coming from and open the original article or guideline if you wish to.

3. Ask focused follow-up questions to guide the consult
Once the initial answer appears, you can narrow the query further by asking follow-up questions. You can also use it like Ask Heidi, or link the question to a consultation or patient.

You can also upload your own custom sources, so Heidi’s answers are tailored to your hospital or practice environment, rather than generic evidence and guidelines.

4. Use the evidence to support, not replace, clinical judgement
Heidi Evidence does not make diagnostic recommendations. Instead, it gives you a fast and efficient way to review evidence and guidelines related to the clinical context.

You still need to review the output critically and apply clinical judgement before making decisions on further management.

Outcomes: Using Heidi Evidence, Dr Harry can approach a more complex consultation in an evidence-informed way. He can quickly review research studies and local guidelines by asking questions. This helps him practise more safely, and make clinical decisions with greater confidence when the presentationis not straightforward.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

AI Rostering Tool Rolls Out Across NHS Trusts to Cut Shift Gaps and Agency Spend

(a February update worth revisiting)

Patchwork Health, a doctor-led workforce management platform, has launched an AI-powered Preference-Based Rostering tool for NHS hospitals, designed to build clinician rotas that better match staff preferences while still meeting compliance and operational requirements. The tool is already being used at four NHS trusts across London, Yorkshire, and the South West of England.

Image Source: Patchwork Health

The system collects clinician shift preferences, alongside workforce rules, minimum staffing requirements, and service demand. Its AI algorithm then processes thousands of possible roster combinations to generate a flexible rota in minutes, distributing night shifts, weekend work, and longer shifts more fairly than traditional manual scheduling.

Once the rota is produced, managers spend less time making repeated manual changes after publication, while clinicians gain more control over their working patterns. This is intended to reduce rota-related frustration, cut shift gaps, and lower reliance on temporary agency staff, while keeping service delivery stable.

Key Features:

• Operational impact: In an early 10-week ward trial comparing the tool with a traditional roster-building process, Patchwork reported a 97% reduction in unfilled shifts.

• Cost reduction: Temporary staffing costs fell by 98%, from £18,000 to £400 over the same 10-week period.

• Preference-based design: The system is built to account for both positive and negative clinician shift preferences, with Patchwork’s modelling suggesting rosters can meet 98% of negative preferences at the point of creation.

• Established NHS footprint: Patchwork says it already works with over 200 healthcare organisations, is partnered with over half of NHS trusts and health boards across the UK, and has supported more than 120,000 healthcare professionals through its wider workforce platform.

Implications for the Health System and Clinicians:
This is one of the more practical healthcare AI rollout stories from recent weeks because it targets a real operational bottleneck. Rostering is a high-friction problem that affects clinician retention, burnout, staffing gaps, and agency spend.

The broader lesson is that some of the most useful AI in healthcare may come from workforce tools that make services easier to operate, not just diagnostic and clinical decision support. The obvious caution is that the strongest result so far come from early company-reported trials, so the real test will be whether these gains hold across full NHS rollouts at scale.

AI in Breast Screening: What the BRAIx Risk Score Means for Personalised Screening

A population cohort study in The Lancet Digital Health evaluated the BRAIx risk score, an AI-derived mammography-based tool that was repurposed from detection, to the prediction of both current and 4-year breast cancer risk.

The study used a training dataset of nearly 400,000 women, then tested the model in an independent cohort of over 96,000 women from BreastScreen Victoria, Australia with external validation in a Swedish screening population.

Image Source: The Lancet Digital Health

Key Findings:

• Strong intermediate-term risk prediction: The BRAIx risk score was a strong predictor of breast cancer diagnosed within 4 years after an all-clear screen, with consistent performance across both Australian and Swedish screening datasets.

• High-risk subgroup identified despite an initial all-clear: Among the top 2% of women with the highest BRAIx risk scores who were initially cleared at screening, 9.7% were diagnosed with breast cancer within 4 years, a risk level higher than that reported for BRCA1 and BRCA2 mutation carriers over the same timeframe.

• Outperformed traditional risk factors: The BRAIx risk score was a stronger predictor of 4-year breast cancer risk than age, family history, mammographic density, recall status, screening round, or country of birth. After accounting for BRAIx, mammographic density was no longer a significant predictor in the Australian test data.

• The score alone was as good as full risk models: Full multivariable models performed only marginally better than the BRAIx risk score alone, suggesting that traditional risk factors contribute little additional predictive value once the AI score is already accounted for.

Implications for Healthcare Systems:
For screening programmes, the practical implication is a shift toward more personalised, risk-based pathways, such as closer follow-up or supplemental imaging for women at highest intermediate-term risk, rather than a one-size-fits-all screening model. The caution, however, is important: this was still a risk-prediction study, not a prospective implementation trial, so any policy change will still need real-world validation, cost-effectiveness modelling, and careful attention to equity.

Read the full study here.

4. Vitals – Quick Bytes

• New Zealand explores AI Integration for Breast Screening

  Health New Zealand is exploring how AI can safely support BreastScreen Aotearoa, with current scoping extending beyond mammogram reading to breast density reporting, image quality support at acquisition, and breast cancer risk stratification. The move comes as screening demand rises, radiologist workforce shortages persist, the screening age extends to 74, and transition from largely paper-based, opt-in process to a digital-first, opt-out register through Te Puna. International evidence, including Sweden’s MASAI trial, suggests AI-supported screening may improve early cancer detection and reduce interval cancers. There is a clear opportunity for New Zealand, but the real challenge is implementation, ensuring any adoption is safe, equitable, and workable within the existing infrastructure. (healthcareitnews.com)

• UK Pilots AI and Robotic Bronchoscopy to Speed Lung Cancer Diagnosis

  The NHS has launched a pilot at Guy’s and St Thomas’ that combines Optellum’s AI risk stratification for low-dose CT lung scans with Intuitive’s Ion robotic bronchoscopy to identify and biopsy small, high-risk lung nodules, including lesions as small as 6 mm. The aim is to shorten diagnosis by replacing weeks of repeat scans and invasive procedures with a single 30-minute biopsy appointment as targeted lung screening expands across England. More than 300 robotic biopsy procedures have already been carried out, with 215 patients going on to receive cancer treatment. The main caution is that this remains a pilot, so impacts on waiting times, outcomes, cost-effectiveness, bias, and equity still need full evaluation before wider rollout.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

Hi Pulse Readers - this week, we’re diving into:

• India’s AI-powered E-ICU model revolutionising critical care delivery,

• what 1.5 million Australian encounters reveal about AI virtual triage and ED demand shift,

• and how you can “ask Heidi to do anything” for you.

Subscribe now

2. Case Study – Your Fortnightly Practical

Image Source: Heidi Health

“Ask Heidi to Do Anything…”

Case Presentation: Last time, Dr Harry managed to build a dedicated ADHD assessment template to streamline his documentation. His consult notes became faster, more structured, and consistent across follow-ups.

Today, he met a young child whose presentation raised concerns, and he decided the child would benefit from a referral to a paediatrician for specialist input. However, he does not often write paediatric ADHD referral letters, and he has no pre-built template for this.

Dr Harry has already finished documenting the assessment with his ADHD template. But now he needs to draft a structured referral letter from scratch. He stays back an extra 30 minutes at the end of the day to do this.

He wonders whether Heidi can help with these kind of ad-hoc tasks – where no template exists and the structure is different from his standard notes.

Approach: Instead of writing from scratch or building a new template for a task you perform only occasionally, use Ask Heidi to repurpose the consult you have documented.

1. Finalise and review the consult note first
Complete and review your note as usual. Confirm accuracy of history, examination findings, medications, and impression.
Ask Heidi should work from an accurate, reviewed source document.

2. Ask Heidi to generate a specialist referral letter with a prompt
Simply type what you want to create into the Ask Heidi to do anything bar at the bottom of the screen, or tap the Ask Heidi button on mobile to open the text box and start typing.

Prompt example:

“Using this consult note, draft a referral letter to a paediatrician for ADHD assessment. Include relevant history, developmental details, family background, school concerns, previous interventions, current medications, risk assessment, and a clear clinical question. Write in formal New Zealand referral style.”

Note: A prompt is the instruction you give an artificial intelligence to tell it what to do. The clearer and more specific the instruction, the better the output.

3. Refine the output with follow up prompts in the same Ask Heidi chat box
Heidi analyses the existing consult note and context, and generates a structured referral letter in a new chat window. You may then tell Heidi to:

“Expand the developmental history and include psychosocial context.”

“Condense this to one page while retaining all clinically relevant information.”

This iterative prompting takes seconds, not another 20 minutes of rewriting. You will still need to review the final version for accuracy before copying and pasting it into your PMS or EHR referral system.

4. Apply the same workflow to other ad-hoc tasks
From the same consult note, you can also use Ask Heidi to:

• Draft a school support letter

• Generate a patient-friendly explanation of why referral is needed

• Summarise the case for a MDT discussion / conference

• Pull in medication information or add billing codes

Multiple potential outputs within seconds and no templates required.

Outcomes: By using Ask Heidi, Dr Harry no longer feels slowed down by work he does not have a template for or is less familiar with. Whether it is a paediatric referral, a school letter, or a one-off report, he can generate a refined output in minutes.

Instead of being disrupted by ad-hoc administrative tasks, he now has a flexible workflow that adapts to whatever the consult requires, without adding unnecessary time to the end of his day.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

India Launches AI-Enabled E-ICU to Extend Private Specialist Critical Care to Public District Hospitals

An AI-enabled E-ICU Command Centre has been launched in Ghaziabad, India, linking a tertiary “base” command centre at a private hospital with the ICU at a public district hospital. The model is designed to strengthen critical care capacity by combining real-time, remote specialist oversight with AI-supported monitoring.

Image Source: IndiaMedToday

When a critically ill patient is admitted to the district hospital ICU, the system connects bedside monitors and the hospital information system to a centralised dashboard. AI-driven analytics then continuously assess patient data, support risk stratification, and generate early alerts for clinical deterioration. A central team of critical care specialists will be available 24/7 at the command centre to assist the on-ground medical teams with timely interventions and management.

This allows high-quality intensive care to be more accessible and affordable for economically weaker sections of society.

Key Features:

• Continuous remote oversight: Round-the-clock specialist supervision from a central command centre supports on-site ICU teams.

• Integrated clinical data: Bedside monitors and hospital systems feed into a unified dashboard for consolidated patient visibility.

• AI-supported early warning: Algorithms assist with deterioration detection and risk stratification, aiming to enable earlier intervention.

• District-level strengthening: The model focuses on expanding access to structured critical care support in non-tertiary settings.

Implications for the Health System and Clinicians:
For clinicians, the value proposition is the integration, earlier detection and access to immediate expert input when risk escalates. This results in more equitable healthcare delivery. For practice managers and health system leaders, this is infrastructure-level AI adoption. The technology is embedded within workflow, data, monitoring, escalation pathways, and governance, rather than deployed as an isolated tool.

The unanswered question remains outcomes. There is currently no published data on mortality, ICU length of stay, transfer rates, or adverse event reduction. The implementation architecture is clear, while the clinical impact still needs to be demonstrated.

AI Virtual Triage in Australia: What 1.5m Patient Encounters Show

A cross-sectional study in Mayo Clinic Proceedings: Digital Health evaluated Australia’s AI-based Virtual Triage and Care Referral (VTCR) system, deployed by Healthdirect Australia as a 24/7 digital front door to care.

Analysing 1,552,592 patient encounters between April 2023 and March 2025, the study examined whether the AI-driven symptom assessment improved alignment between patient care intent and recommended pathways. This includes self-care, GP, urgent care, virtual emergency, or in-person emergency services.

Image Source: Medical Republic

Key Findings:

• Care alignment: The propotion of patients appropriately selecting lower-acuity, non-urgent care mroe than doubled, from 21% to 53%.

• Uncertainty eliminated: Patients unsure what care to seek dropped from 43.2% to 0.2% – a 99.6% reduction.

• Emergency department diversion: Intent to attend in-person ED fell from 37% to 25% as virtual pathways became available.

• Virtual care uptake: Among those who initially planed to attend in-person ED, virtual emergency care rose from < 1% to 11%, with additional growth in telehealth and urgent care use.

• System-level impact: If results seen in Victoria were replicated nationally, modelling suggests around 2,400 unnecessary non-urgent visits and 19,000 unnecessary ED visits could be avoided annually.

Implications for Healthcare Systems:
AI virtual triage demonstrates potential for improving care alignment, reducing patient uncertainty, and redirecting low-acuity demand away from emergency departments toward telehealth and urgent care pathways. The early signals are strong, but validation of actual service utilisation, cost savings, and hard clinical outcomes will be needed before large-scale adoption.

Read the full review here.

4. Vitals – Quick Bytes

• NZ Government funds $4.6m for 10 AI-in-health research projects

  The Health Research Council has awarded $4.6 million across ten artificial intelligence research projects aimed at transforming healthcare in Aotearoa. Funded initiatives span across AI-assisted stroke and radiology imaging, heart failure management tools, digital pathology for gastrointestinal cancers, postoperative monitoring, primary care diagnostic pilots, and ethical frameworks for responsible AI use. All projects are scheduled to run through to completion between late-2026 and late-2027. This investment reflects continued national focus on evaluating both technical accuracy and implementation, safety, and equity of AI within New Zealand’s health system. (hrc.govt.nz)

• PRIMARY-AI: a propsed framework for judging whether AI helps or harms primary care

  A Nature Medicine correspondence argues that AI is already being deployed at scale in primary care, but most evaluation still focuses on technical performance metrics rather than outcomes that actually matters in general practice, such as continuity, coordination, accessibility, and people-centered care. The authors propose PRIMARY-AI, an outcomes-based standards framework with a three-tier approach: safety and fairness, real-world effectiveness and reliability, and primary care-specific attributes including people-centeredness, coordination, and continuity. They plan to develop validated outcome sets and minimum evidence thresholds via a multi-stakeholder Delphi method beginning in 2026, with final standards targeted for October 2026.

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in the next edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

1. Triage – Your Fortnightly Rundown

Hey Pulse Readers, we’re back! And we will be in your inbox every second Wednesday from now on. Let’s dive right into it:

• promising results in AI physiotherapy clinic with expansion across the UK,

• what 11 randomised trials found about the impact of AI in cardiology clinical outcomes and operations,

• and how you can get Heidi to write notes that are more “like you”.

Subscribe now

2. Case Study – Your Fortnightly Practical

Image Source: Heidi Health

Designing Documentation That Matches Complex Consults

Case Presentation: Harry is a busy GP in Aotearoa who recently started offering adult ADHD assessments after the 1 February 2026 rule change by the New Zealand Government.

These appointments are long, often over 60 minutes and across multiple visits. He found that his standard GP note templates (SOAP style, short problem lists) do not fit the structure of an ADHD assessment or the f/u plan.

He is stuck rewriting the same sections every time and has to stay behind after hours to finish his notes.

Approach: Build a fit-for-purpose ADHD assessment template by uploading a de-identified example document to Heidi, then iterating until the output matches how you want to document.

1. Pick a “gold standard” example note
Choose one ADHD assessment note you are happy with (3^rd party or self-drafted), focusing on headings and structure (not patient specifics). Aim for the exact output style you want to reuse as your baseline structure.

2. De-identify properly, do not be sloppy
Replace names, NHI, DOB, address, and other unique identifiers with placeholders (for example, [PATIENT], [DOB], [ADDRESS], [DATE]).

3. Upload and create a template in Heidi
Use Heidi’s template creation workflow to upload your de-identified note and have Heidi generate a “Adult ADHD assessment” template according to your preferred structure and wording. Heidi supports building templates through conversation and adapting from uploaded existing templates. Alternatively, reach out to the Hendrix team, and they can do this for you.

4. Test and refine
Run a dummy case through the template, check the output, then adjust the template instructions until it reliably produces the level of detail you need and you have to make as little edits as possible before copying and pasting into your PMS or EHR. You will still need to review the notes for accuracy in any case.

Save and favourite it for fast access.

Outcomes: With a dedicated ADHD template built from his own documentation style, Harry stops re-writing the assessment from scratch each time. His consult notes become consistent across clinicians, easier to review at follow-up, and super quick to complete, which matters when assessments routinely take > one hour.

Instead of fighting his usual GP templates, he now has a repeatable workflow that fits the new reality of ADHD care in primary care.

Disclaimer: Hendrix Health is the official New Zealand partner for Heidi Health.

3. The Pulse - Your Fortnightly Update

AI Physiotherapy Clinic Now Expanding Across 11 NHS Regions After Pilot Programmes

Flok Health, an AI powered physiotherapy clinic, has expanded across 11 NHS regions to provide patients with same-day video appointments for back pain via a smartphone app. This follows large-scale pilots in Scotland and England in early 2025, and the rollout now covers more than 2.4 million patients across England.

Image Source: Flok Health

The service begins with a mobile phone video assessment, with the AI system evaluating patient symptoms. If approved, they receive weekly video appointments with the AI physiotherapist, available at any time. Patients can also send voice messages to the clinical team and receive responses during working hours.

Those with more complex symptoms have a further telehealth assessment with a clinician before digital treatment; or otherwise, referral to a more appropriate NHS service. This keeps clinician time focused on higher complexity care.

Key Features:

• Waiting list impact: In one of the pilots, waiting lists for back pain were cut by more than half, and waits for all MSK conditions fell by 44 percent in under 10 weeks.

• End-to-end pathway: The AI physiotherapist manages patients from initial assessment through treatment and discharge, recording interactions and outcomes into the patient’s EHR.

• Regulatory readiness: Class IIa medical device approval under EU regulations supports autonomous delivery of diagnostic decisions and treatment pathways without requiring clinician input.

• Patient reported improvement: Average MSK-HQ score improvement of 6.16 points, with anything above 5 considered to be clinically significant.

Implications for the Health System and Clinicians:
For the UK, this is a clear signal that commissioners will back regulated, integrated digital pathways where the operational case is strong. For ANZ services watching from a distance, the useful lesson is the combination of regulatory clearance, and measurable waiting list impact.

In the long term, the real test will be sustaining outcomes across diverse populations, maintaining clinical oversight, and ensuring equitable access for patients who might not engage well with smartphone-delivered care.

AI in Cardiovascular Care: What 11 Randomised Trials Tell Us

A new systematic review in Journal of the American College of Cardiology examined randomised controlled trials of the application of AI in cardiovascular care. It aims to investigate AI’s impact on cardiology patient outcomes and health systems.

Image Source: Journal of the American College of Cardiology (JACC)

Key Findings:

• ECG: > 50% of the trials focused on ECG-based AI, mainly to screen and flag higher-risk patients earlier, prompting early detection or escalation. One trial showed reduced 90-day mortality, but another trial showed worse outcomes. Overall, modest clinical benefits and often seen in high-risk subgroups.

• Computed Tomography: CT-based AI tools were used to support coronary risk assessment and procedural planning. It helped reduce unnecessary downstream invasive testing, and promoted consistent preventive care (e.g. statin use) when actionable findings are identified.

• Echocardiography: Results were mixed. For routine echo workflows (e.g. EF assessment), AI performed as well as sonographers and helped save time, but performance was not strong enough in specialised context such as stress echo for coronary disease detection.

• Evidence and reporting gaps: Only 11 out of 930 RCTs (2021 to 2024), met inclusion criteria. None of which reported patient-reported outcomes (e.g. quality of life, symptoms) and only 5 RCTs declared adherence to AI-specific reporting standards (CONSORT AI or SPIRIT AI). Many studies had moderate to high risk of bias linked to industry funding or financial interests.

Implications for Healthcare Systems:
AI in cardiology shows promise for screening, early detection, standardised interpretation, as well as operational / workflow efficiency. More high-quality evidence is needed to validate AI’s effectiveness in hard clinical outcomes (e.g. treatment response rates, patient-reported outcomes) before widespread clinical reliance and scaling.

Read the full review here.

4. Vitals – Quick Bytes

• AI in ED sepsis, early alerts and smarter diagnostics, but limited evidence

  A narrative review outlines how AI could support sepsis care in the emergency department, from early warning alerts (e.g. TREWS) to improving diagnostic work ups such as targeting blood cultures and combining biomarkers with routine data (for example Sepsis ImmunoScore). The main caution is that most sepsis AI has only been validated in observational studies, with limited randomised evidence. The incomplete ED data capture and generalisability also remain challenges. The takeaway is simple: promising tools, but they still need rigorous trials and careful workflow integration before wide rollout.

• US-based Optain Health acquires EyePACS to scale AI retinal screening in primary care

  Optain Health is combining their AI-powered robotic retinal imaging with EyePACS’ teleophthalmology network, expanding retinal screening in primary care and community clinics. A portable, robotic camera captures retinal images at the point of care to detect diabetic retinopathy, glaucoma, and age-related macular degeneration. Optain is also positioning this platform for “oculomics”, using signal from the eye to inform broader cardiometabolic risk assessment. The combined platform’s specialist multi-condition grading, EHR integration and analytics helps support population health programmes. (fiercehealth.com)

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in next week's edition of The Pulse.

Have a great day & see you in two weeks!

Your Hendrix Health Team

1. Triage – Your Weekly Rundown

Hi Pulse Readers - this week, we’re focusing on:

• how AI is transforming physiotherapy care in Scotland with NHS Lothian’s innovative pilot,

• what 86 clinical trials reveal about the impact of AI in healthcare delivery,

• and how you leverage AI tools like ChatGPT to achieve your New Year’s resolutions.

Subscribe now

2. Case Study – Your Weekly Practical

Image Source: generated with Recraft.ai

Achieve your Practice’s New Year’s Resolutions with AI

Case Presentation: New Year’s resolutions can drive personal and professional growth, but busy schedules in healthcare make it difficult to stay on track. Whether you want to improve clinic efficiency, foster team well-being, or enhance work-life balance, LLM tools like ChatGPT or Claude can act as your planning assistant to help you take actionable steps.

Approach: Prompt an AI tool to break your resolutions into manageable tasks, optimise time management, and ensure steady progress throughout the year.

1. Set Professional and Personal Goals
Prompt Example:
"I am a [GP, practice manager, nurse, etc] at a busy medical centre in [country]. Help me set New Year’s resolutions for streamlining clinic workflows and improving staff morale. Also, suggest a goal to maintain a better work-life balance. I would like to especially improve my [fitness] [sleep] [stress management] [etc] this year."

Type “Regenerate” and refine as needed to create additional options.

2. Develop Actionable Plans for Practice Improvements
Prompt Example:
"I want to improve patient wait times and reduce admin workload. Create a step-by-step plan that excludes [technology upgrades, etc], but includes [staff training, etc]."

3. Build a Well-Being Strategy for the Team
Prompt Example:
"Suggest a weekly wellness program for my team, including short mindfulness sessions, healthy snack options, and team-building activities."

4. Track Key Metrics and Celebrate Achievements
Prompt Example:
"Create a system for tracking staff engagement levels. Suggest ways to recognise and celebrate milestones with the team."

5. Manage Time Effectively
Prompt Example:
"Provide tips for managing my time as a [GP, practice manager, nurse practitioner], including strategies for delegation and maintaining focus on high-priority tasks."

Outcomes: By integrating AI into your New Year’s resolution planning, you can balance personal well-being with professional goals, enhance clinic operations, and cultivate a supportive team environment. Make 2025 a year of growth and achievement for you and your practice!

3. The Pulse - Your Weekly Update

AI-Powered Physiotherapy Clinic Now Available in Scotland

Image Source: Flok Health

NHS Lothian has launched a large-scale pilot of an AI-powered physiotherapy clinic designed to provide personalised care to nearly one million patients across the region.

Based on video footage of human physiotherapists, the system replicates a structured video-call experience, offering rapid access to musculoskeletal (MSK) care.

Key Features:

• CQC Approval: The solution is the first digital MSK clinic to be approved by the Care Quality Commission in the UK.

• Patient Feedback: During earlier trials, 100% of patients rated their experience as equal to or better than with a human physiotherapist, with 86% reporting symptom improvement.

• Streamlined Access: Patients can self-refer through NHS Lothian’s existing portals or be referred by participating GPs for same-day digital appointments.

• Data Integration: The service links seamlessly with NHS systems, ensuring coordinated care and robust clinical and cost outcomes analysis.

Implications for the Health System and Clinicians:
This pilot highlights the potential of AI to address treatment delays and improve care accessibility. It also demonstrates a shift towards integrating digital solutions into routine clinical workflows. While promising, challenges such as ensuring long-term patient engagement, clinical oversight, and equitable access remain critical to address.

With plans to expand the service to other MSK conditions, solutions like Flok Health could redefine how physiotherapy is delivered at scale.

Evaluating AI’s Impact: What 86 Clinical Trials Reveal

Image Source: The Lancet Digital Health

A comprehensive review in The Lancet Digital Health analysed 86 randomised controlled trials (RCTs) assessing AI’s role in clinical practice, providing insights into its potential and limitations.

Key Findings:

• Care Management Improvements: AI systems for insulin dosing and hypotension monitoring enhanced patient outcomes by increasing the time patients stayed within target glucose and blood pressure ranges. In oncology, AI-assisted radiation therapy and prostate brachytherapy showed reductions in acute care rates and tumour volumes.

• Behavioural and Symptom Impact: Trials revealed AI-driven diabetic retinopathy risk predictions boosted patient referral adherence, while nociception monitoring systems decreased postoperative pain scores.

• Clinical Decision Support: AI predictions on cancer mortality prompted more serious illness discussions between oncologists and patients. However, other trials, such as AI tools for atrial fibrillation stroke risk, showed limited impact on prescribing behaviour.

Implications for Healthcare Systems:
While AI shows promise in improving care management, enhancing patient adherence, and assisting in clinical decision-making, the variability in outcomes highlights the need for more robust, multicenter trials. To ensure effective integration, future research should prioritise patient-centred outcomes, equitable access, and transparent reporting.

Read the full review here.

4. Vitals – Quick Bytes

• AI Advances Mammography Screening in Germany

  Germany’s PRAIM study reveals that AI integrated into mammography screenings has improved cancer detection rates by 17.6%. Using AI for triage and safety net alerts streamlined workflows, reduced radiologist workload by 56.7%, and enhanced positive predictive values for recalls and biopsies. This large-scale implementation, over 460,000 screenings, underscores AI's potential to improve early cancer detection while maintaining accuracy and efficiency.

  Explore the study here for more on its groundbreaking findings.

• CHAI Releases Open-Source “Nutrition Label” for Healthcare AI

  The Coalition for Health AI (CHAI) has launched an open-access model card on GitHub, offering a comprehensive "nutrition label" for healthcare AI algorithms. Designed to improve transparency, the card details key information like training data, bias mitigation strategies, and ongoing maintenance, enabling healthcare providers to assess AI models effectively. (fiercehealth.com)

We’d love to hear your thoughts, so join the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in next week's edition of The Pulse.

Have a great day & see you next week!

Your Hendrix Health Team

1. Triage – Your Weekly Rundown

Ho ho ho Pulse Readers - this week, we’re highlighting:

• China’s virtual hospital with self-evolving AI doctors,

• how OpenAI’s Sora is opening new doors for visual patient education,

• and how you can use AI to take the stress out of planning your next medical centre team event.

Subscribe now

2. Case Study – Your Weekly Practical

Image Source: generated with Recraft.ai

Planning Your Next Event with ChatGPT

Case Presentation: Organising a festive Christmas party for your medical centre can be challenging, especially with busy schedules. Tools such as ChatGPT can help simplify the process, offering creative ideas for food, activities, and decor while considering budget and time constraints.

Approach: Use ChatGPT as your planning assistant by providing detailed prompts to generate ideas tailored to your team’s preferences and resources.

1. Define Your Theme and Vision
Prompt Example:
"I’m planning a Christmas party for our medical centre staff. We want a fun and festive theme suitable for doctors, nurses, and admin staff. Please suggest 3 creative themes with decor, outfit ideas, and activities."

2. Create a Festive Menu
Prompt Example:
"Plan a Christmas party menu for 30 people, including finger foods, a main dish, desserts, and non-alcoholic drinks. We have a mix of vegetarians and meat eaters. Include easy-to-prepare and budget-friendly options."

3. Plan Activities and Entertainment
Prompt Example:
"Suggest engaging Christmas party activities for a medical centre team. We want team-building games, a Secret Santa gift exchange, and some fun prizes. Include instructions for at least one activity."

4. Set a Budget and Shopping List
Prompt Example:
"Help me create a budget for a Christmas party for 30 people. We need decor, food, drinks, and small gifts. Break it down into categories, and provide a detailed shopping list."

We hope these tips help to get the creative juices flowing… try it for your next event!

3. The Pulse - Your weekly Update

China’s Self-Evolving AI Virtual Hospital

Image Source: Tsinghua University’s Institute for AI Industry Research

Tsinghua University’s Agent Hospital, a self-evolving virtual healthcare platform, is slated for public release in 2025. Utilizing AI-generated doctors, nurses, and patients, the system simulates real-world clinical workflows, offering advancements in training of AI agents for clinical applications.

Key Features:

• MedAgent-Zero Framework: Allows AI doctors to learn continuously from interactions, literature, and outcomes.

• Simulated environment covers diagnostics, treatments, and follow-ups.

• AI doctors achieved up to 93.06% accuracy in tasks like diagnosis of respiratory conditions

Implications:
By simulating tens of thousands of cases in days without the need for human-labeled data, the system accelerates the proficiency of AI agents to carry out clinical tasks such as triaging, diagnosing and treating patients with increasing accuracy.

While those agents could help to alleviate clinician shortages and free clinician’s time for complex tasks, it raises ethical concerns about accountability in patient care, and broader questions about the respective roles of doctors and AI agents in the future.

Still a long way to go, but definitely one to watch.

For more insights, check the full study here.

OpenAI Unveils Sora: What are the implications for healthcare?

Image Source: generated with Sora, via Sora.com

OpenAI has launched Sora, an AI model that generates videos from text prompts. This innovation shows potential to significantly enhance healthcare communication and education by enabling personalised, visual content creation.

Key Features:

• Dynamic Video Creation: Sora can generate videos up to 20 seconds long based on user-provided text prompts.

• Versatile Applications: Tools like "Remix," "Loop," and "Blend" allow users to customize and enhance video content.

• Accessible: Available to ChatGPT Plus and Pro subscribers, Sora democratizes video production with minimal technical requirements.

Implications for Patient Care:
Sora could redefine how healthcare providers communicate with patients:

• Personalised Patient Instructions: Clinicians could create tailored videos explaining post-operative care, medication usage, or chronic disease management.

• Physiotherapy Demonstrations: Visual guides for physical therapy exercises could improve patient adherence and outcomes.

• Health Education: Public health campaigns and condition-specific education could be delivered in engaging, accessible video formats.

Challenges and Considerations:
While promising, Sora's applications in healthcare are not without challenges:

• Quality and Accuracy: Ensuring videos are clinically accurate and free from visual inconsistencies is still currently challenging, especially with the early maturity state of the technology. Also, videos cannot be generated with accompanying audio as of today.

• Ethical Use: Safeguards will be critical to prevent misuse, such as generating misleading or harmful content.

• Integration with Workflows: Providers will need efficient processes to incorporate text-to-video tools into clinical practices.

As Sora evolves, its potential to enhance patient education and support healthcare professionals makes it a tool worth watching in the AI healthcare landscape.

We also expect to see other solutions develop in the text-to-video space that are more suited to narrower healthcare use cases.

4. Vitals – Quick Bytes

• Scientists Propose Global Collaboration on AI Virtual Cell
  Researchers from Stanford, Genentech, and the Chan-Zuckerberg Initiative propose a global effort to create an AI-powered virtual human cell. This model could simulate cellular behavior, accelerate drug discovery, and advance personalised medicine. Achieving it will require massive datasets and unprecedented collaboration across disciplines. (stanford.edu)

• Neuralink Launches Study to Connect Brain Implants with Robotic Arms
  Neuralink has announced the CONVOY study to test its N1 brain-computer interface implant for controlling assistive robotic arms. The implant interprets neural activity, enabling individuals with quadriplegia to operate devices or robotic limbs with their thoughts. This advancement marks a step toward integrating robotics with neural technology for restoring mobility and independence. (mobihealthnews.com)

• AI Surpasses Experts in Neuroscience Predictions

  A new study shows large language models (LLMs) outperform human neuroscientists in predicting experimental outcomes, achieving 81.4% accuracy compared to 63.4% by experts. By analysing patterns in vast scientific datasets, models like BrainGPT can uncover hidden insights to guide future studies and streamline discovery processes.(nature.com)

We’d love to hear your thoughts, so joint the conversation by leaving a comment below:

Leave a comment

We are taking a break to recharge the batteries over the holiday period and will be back with your next issue of The Pulse the week of January 6th.

Wishing you all a relaxing festive season & see you in the New Year!

Your Hendrix Health Team

1. Triage – Your Weekly Rundown

Hi Pulse Readers - this week, we’re highlighting:

• how TrialGPT is streamlining patient recruitment for clinical trials,

• how a Microsoft open-source model is democratizing medical imaging analysis,

• and how you can quickly create impactful public health posters with AI.

Subscribe now

2. Case Study – Your Weekly Practical

Image Source: generated with Recraft.ai

Creating Public Health Posters with Recraft.ai

Case Presentation: Public health campaigns require engaging visuals to effectively communicate messages. Designing these posters can be time-intensive and resource-heavy.

Approach: Recraft.ai simplifies the process by using AI to create professional-grade public health posters tailored to your target audience. The platform, which recently shot to the top of the text-to-image leaderboard, provides a generous free tier and enables rapid poster creation with minimal effort.

How to Use Recraft.ai for Public Health Posters

1. Set Up Your Project:
   Log in to Recraft.ai and select “New project” and then use the “Frame” object to select the format for your poster (1:1, 4:3, etc).

2. Input Your Key Message:
   Use the text prompt feature to describe your poster’s goal, e.g., “Create a poster encouraging flu vaccination for young adults.” You can use the text tool to add specific text you would like to include in the frame, as well as any reference images.

3. Customize the Design:
   Once generated, edit colors, fonts, and layouts to align with your branding. Add images, icons, or QR codes for additional information or links to resources.

4. Generate Variations:
   Let Recraft.ai generate multiple design options. Review and refine the one that best fits your needs. You can also choose different styles such as hand-drawn or scientific styles.

5. Download and Share:
   Export the finished poster in high resolution for printing or digital distribution.

Outcomes: Recraft.ai empowers practice managers and clinicians to create visually appealing, impactful posters quickly and cost-effectively, ensuring critical messages reach their intended audience.

3. The Pulse - Your weekly Update

Matching Patients to Clinical Trials Using TrialGPT

Image Source: Nature Communications

Recent advancements in artificial intelligence have transformed patient recruitment for clinical trials, a traditionally labor-intensive process. Researchers have developed TrialGPT, a large language model (LLM)-driven framework that matches patients to clinical trials with remarkable efficiency and accuracy.

Key Features:

• Three-Step Matching: TrialGPT employs retrieval, matching, and ranking to identify trials suited to individual patients, recalling over 90% of relevant trials with less than 6% of the initial collection analyzed.

• Enhanced Accuracy: With an accuracy rate of 87.3% in predicting patient eligibility at a criterion level, TrialGPT delivers results comparable to human experts while offering detailed explanations and sentence-level evidence.

• Efficiency Gains: A pilot study demonstrated a 42.6% reduction in screening time, significantly accelerating patient-trial matching processes.

Impact:
TrialGPT has the potential to revolutionise clinical trial recruitment by reducing human errors, expediting trial matches, and enabling greater scalability.

For more insights, check the full study here.

Microsoft Unveils Open-Source Model for Medical Imaging Analysis

Image Source: Microsoft

Microsoft has unveiled BiomedParse, a GPT-4-powered AI foundation model designed to transform biomedical image analysis. This new tool unifies object recognition, detection, and segmentation, enabling clinicians to analyse medical imagery - from tumors to COVID-19 infections - using simple text prompts.

Key Features:

• Unified Framework: Unlike previous tools that separately perform object detection and segmentation, BiomedParse integrates these tasks for a holistic image analysis experience.

• Natural Language Input: Users can specify objects for analysis using simple text prompts, eliminating the need for manual bounding boxes.

• Data-Driven Precision: The model was trained on BiomedParseData, a dataset of over 6 million image-mask-description sets covering 64 object types and 82 subtypes.

Implications:
BiomedParse outperformed previous best methods like MedSAM and SAM in all evaluated metrics, particularly excelling in segmenting irregularly shaped objects, a common challenge in medical imaging. Its ability to identify and segment objects without additional user input reduces errors and increases efficiency.

Available under an Apache 2.0 license and deployable on Azure AI, BiomedParse is positioned to democratize access to advanced biomedical imaging tools.

4. Vitals – Quick Bytes

• Study Highlights Bias in AI Driven by Racial Differences in Lab Testing
  A recent study from the University of Michigan highlights how racial disparities in laboratory testing could introduce bias into AI models used in healthcare. The research found that African-american patients were significantly less likely than Caucasian patients to receive certain tests, such as complete blood counts or metabolic panels, during emergency department visits. This discrepancy risks skewing AI models trained on such data, potentially amplifying existing healthcare inequities. Read the study here.

• Hippocratic AI’s Patent Advances Safety in Healthcare LLMs
  Nvidia-backed start up Hippocratic AI has patented Polaris, a safety-focused framework that combines multiple AI models to verify outputs in healthcare applications. Its architecture, involving engines for tasks like medication reconciliation and chronic care management, aims to prevent errors while assisting with pre-op, discharge, and billing processes. With accuracy exceeding 99%, Polaris reflects a growing trend in safety-driven AI in healthcare (business wire.com)

• AI "Virtual Sitter" for Patient Monitoring
  Teladoc Health has launched Virtual Sitter, an AI-driven tool designed to detect and alert hospital staff to patient movements that may lead to falls. Using computer vision, the system monitors movements, identifies risky actions, and notifies remote staff for intervention. This system aims to improve patient safety and optimize staff resources in inpatient settings. (fiercehealthcare.com)

We’d love to hear your thoughts, so joint the conversation by leaving a comment below:

Leave a comment

PS: Thanks for your patience, we had a family emergency last week that prevented us from publishing our weekly newsletter, back to our usual weekly schedule now.

Have a great day & see you next week!

Your Hendrix Health Team

1. Triage – Your Weekly Rundown

Hi Pulse Readers - this week, we’re diving into:

• the heated debate over a study claiming AI outperforms doctors in diagnostic reasoning,

• a groundbreaking “ChatGPT for DNA” unlocking new possibilities in genomics,

• and we show you how AI-powered voice simulations can help clinicians hone their communication skills.

Subscribe now

2. Case Study – Your Weekly Practical

Image Source: generated with Recraft.ai

Practising Communication Skills with AI

Effective communication is a vital skill for healthcare professionals. Traditional training methods, such as role-playing and simulations, require significant time and resources.

Approach: AI tools like ChatGPT with voice mode offer an innovative solution for communication practice by simulating realistic patient interactions. Using simple prompts, healthcare professionals can engage in virtual role-play to refine their bedside manner and communication techniques.

Watch a video tutorial here, and/or follow the step by step below.

How to Use AI for Communication Practice:

1. Access Voice Mode: Open advanced voice mode in an AI tool like ChatGPT (premium feature - available as a mobile app or web application).

2. Set the Scenario: Begin with a clear prompt, such as: "Act as a patient. I am a physician practicing to deliver bad news regarding a serious diagnosis. Respond as a concerned patient."

3. Conduct the Simulation: Engage in the conversation, practising your communication skills. For example, deliver test results with empathy and offer supportive follow-up questions, such as: "How are you feeling about this news?"

4. Request Feedback: Ask the AI to provide feedback on your tone, clarity, or approach to ensure you improve over time.

Outcomes: Practising communication with AI allows clinicians to simulate a variety of scenarios, from delivering bad news to de-escalating potentially aggressive situations, in a low-pressure environment. This approach is flexible, scalable, and accessible, helping professionals build confidence and refine critical soft skills.

3. The Pulse - Your weekly Update

AI Outperforms Doctors? A Closer Look at a Controversial Study

Image Source: Dr Terence Tan LinkedIn via AI Health Uncut

A recent study published in JAMA claimed that a large language model (LLM) outperformed clinicians in diagnostic reasoning tasks, which was then later sensationalised in a NYT article and amplified widely, including by Elon Musk. Many experts have raised concerns about the study's validity due to its questionable methodology and small sample size, as explained in this excellent deep dive by Sergei Polevikov.

• The Study's Claim: The LLM achieved a 92% accuracy in diagnostic reasoning across six hand-picked medical cases, compared to 74% for physicians and 76% for physicians using the LLM as a tool.

• The Issues: Critics argue the study’s reliance on only six medical case histories from an original dataset of 105 undermines its credibility. Hand-picking curated cases and excluding 94% of the sample without transparency raises questions about bias and real-world applicability.

• Real-World Implications: While the study showcases the potential of AI in assisting clinicians, its limitations—lack of training for physicians on using the LLM, reliance on structured vignettes rather than real patient interactions, and minimal sample size—make its findings difficult to generalize.

Implications: This study highlights the need for rigorous standards in AI research, particularly in medicine. Oversimplified claims, amplified by media outlets, risk creating unrealistic expectations for AI's capabilities in healthcare, as well as frustrations in the medical workforce.

Cracking the Code: Evo’s Role in the Future of Genomics

Image Source: Science.Org

Scientists at the Arc Institute in Palo Alto and the Stanford Department of Bioengineering have unveiled Evo, a revolutionary AI model dubbed “ChatGPT for DNA.” Trained on 2.7 million microbial genomes, Evo excels at both interpreting and generating genetic sequences, opening new frontiers in genome research and engineering.

• What It Does: Evo simultaneously analyzes DNA, RNA, and protein sequences, enabling zero-shot function prediction and generating realistic genome-length sequences exceeding 1 million base pairs. It has already designed functional genetic editing tools and accurately predicted DNA changes' effects on bacterial fitness.

• Safety: Researchers intentionally excluded human-affecting viral genomes from Evo's training dataset to mitigate potential risks.

• Applications: Evo could drastically accelerate research, design proteins for drug development, predict disease-causing mutations, and provide unprecedented insights into biological systems.

Implications: Evo’s potential is groundbreaking, but its ability to “write genetic code” with ease raises significant ethical and safety questions. Striking a balance between innovation and responsible use will be crucial as this technology advances.

4. Vitals – Quick Bytes

• AI Journaling for Mental Health: AI-powered journaling tools like Mindsera and Rosebud are transforming the journaling experience by providing personalised prompts, emotional insights, and detailed feedback. These tools aim to enhance emotional processing and self-awareness, making journaling more accessible and effective for users coping with stress, grief, or chronic illness. While not a substitute for professional care, AI journaling offers a supplemental tool for maintaining mental well-being.
  (popsugar.com)

• Personalized Stroke Treatment with AI: The TRUSTroke project at Vall d’Hebron Hospital in Spain uses AI to analyse medical records and patient data, providing tailored treatment plans to optimise stroke recovery and prevent complications. (ft.com)

We’d love to hear your thoughts, so joint the conversation by leaving a comment below:

Leave a comment

Stay tuned for more insights in next week's edition of The Pulse.

Have a great day & see you next week!

Your Hendrix Health Team