Introduction

Hello everyone, my name is Amir. I've been here before a couple months ago. Before I start, anyone here from my previous talks by chance? Okay, a few hands in the audience, good.

So this will be even more interesting for you guys and I hope I can enthuse the rest of you for the next 15 -20 minutes or so as well.

I'm a secondary care doctor by background and over the past two or three years I've pivoted my career into AI and health tech and I'm joined by my colleague Lydia. I'll let her introduce herself.

Hi, my name is Lydia. I'm a clinical pharmacist by background. and worked across primary and secondary care before I jumped into digital health and been in this field for five years now.

Now working as a medical operation with Amir at Tandem. Amazing, thank you.

Why AI Medical Assistants Matter in Clinical Practice

So today we're going to talk to you about a practical application of AI in healthcare. You can see the heading up there.

How are we actually using medical assistants in real life? life.

My previous talk was all about how anyone can build AI tools these days using lovable cloud, et cetera, but what actually makes it to the front line is very different.

So we're going to talk a bit about what it takes to get to the front line and what you require to get there.

As an example, we'll be showing you our company, Tandem. It is an AI medical assistant, funnily enough, suitable to the theme that we'll be discussing.

So I promise

The Documentation Burden—and the Opportunity

there's not really many slides oh before i show that slide i've got a question for everyone so don't answer if you saw it how much of a clinician's time do you think is spent on documentation at the moment 80 we've got anyone else okay so over 50 is a consistent theme i'm glad to hear that you're on the right tracks but it's actually not as high as that it is very high

still 40 which accounts for three percent of global gdp which is spent on documentation and healthcare. Now, for a moment, imagine what happens if we give that time back to clinicians.

They spend more time on the patient, they're more empathetic, they're more attentive, and there's less cognitive burden on the clinician as well, so they have more time to think as opposed to worrying about remembering what to type and what to document.

What Tandem Is and Where It’s Deployed

So, what is Tandem?

As I said, we're one of the leading AI medical assistants in the world right now. Now we're operating across 5 ,000 plus care organizations across Europe in 11 or so more countries now at this stage. These slides are a bit out of date even.

And we've deployed across NHS trusts across the country as well as private health care institutes across Europe. We're a Series A funded organization and we're really making an impact through what we're doing. And we'll show you exactly what that is in a moment.

Live Demo: An Ambient AI Scribe in a Consultation

So, what I'll do is hand over to Lydia, who's going to play doctor, and we're going to put on some amateur theatrics for you. I'll play patient, and we'll walk you through what it actually looks like today when you go to see your GP or a hospital doctor.

So, Lydia, over to you. Okay, can you see this okay? No. Change.

Okay. You should be able to see that. Brilliant. Okay.

So, Lydia, over to you. Perfect.

Setting Up the Workflow and Adding Context

So now you're in Tandem homepage. You can see we try to make the interface as simple and intuitive as possible.

So now we're just going to do a quick role play for you guys. So you can see Tandem in action. I'm going to start the consultation.

Before I start, so I'm about to see Amir, who's been referred by his GP following a severe cough. He has baseline conditions as well. I received the GP letter here, as you can see.

Oh, I think we flipped over to another window. Oh, gosh. Okay. That is okay.

We can imagine that there is a GP referral letter on the screen there. And I am about to add that onto Tandem.

So you can see we can add context onto Tandem. You can add in text format, PDF, audio, and take pictures. So right now I am going to just add as text and click on save.

And now Tandem has a bit of background history of Mimir, and then we are going to start the consultation now. There we go.

Role Play: Clinician–Patient Conversation

So, hi, I'm Dr. Lydia, so one of the emergency doctors here. I understand that your GP has sent you in. How can I help you?

Well, before we do that, let me just make sure that the microphone is picking up the right source. So you can see here, I'm just going to set that over to my MacBook. Brilliant. Okay, so it should hear us now.

Hello, doctor. My name is Amir. So I've been referred to you for my GP. I feel dreadful.

I've had this cough for the past three, four days. I'm feeling feverish. I've got this lung disease, COPD. and yeah it's just making me feel absolutely rotten I can hardly breathe anymore sorry to

hear that so how much worse is your breathing now compared to normal I mean usually I manage okay around the house I do my own chores and activities now I'm struggling to go from room to room or I can hardly even attempt the stairs yeah um any chest pain no chest pain no coughing up any blood

no blood no okay and have you been using your rescue pack antibiotics and steroid yeah so my My GP gave me two medicines. I've been taking them for the past two days. It's done absolutely nothing for me. Okay.

And any travel recently or contact with anyone unwell? No, no travel. I live alone. Okay.

So I can see from your record that you have COPD, diabetes, high blood pressure, and the previous heart attack as well. Is that all correct? Yeah, unfortunately so.

I don't look it with my age, but yeah, I've got a past medical history to me. Brilliant.

And are you taking any regular medications? Yeah, God, I can't remember them, but some blood thinners or some, I don't know, you can find it on my record. No worries. Yeah, it has been documented on your GP referral. That's fine. And just to confirm, any allergies at all? No allergies. Perfect.

So your observation shows oxygen saturation of 92 on room air, respiratory rate 22, pulse 96, blood pressure 130 over 80 and a temperature of 38 degrees. degrees.

So, let me just listen to your chest quickly. Okay. There are some crackles and the reduced air entry at the right lower lung. So, the rest of the examination is shorting though.

So, my main concern is that you have a chest infection. Oh, dear. Likely a community acquired pneumonia causing an exacerbation of your COPD. Oh, no.

So, I would like to arrange some blood tests including infection markers and the chest x -ray and we'll start oxygen if needed and give you some inhalers. We'll also start you on IV antibiotics. Dog, that sounds like a lot.

Do I need to stay in hospital? Should I be worried? Yes, 100%. So I would recommend you to stay in the hospital today whilst we're monitoring your symptom.

Okay, that's not what I wanted to hear, but I appreciate your help. No worries, we'll get the investigation on the way. Amazing.

Round of applause, Lydia. Thank you so much. I feel cured

How the Scribe Captures and Structures the Note

You'll notice that Scribe has just been listening to everything we've said in the background. What I want to point out is that I hope at no point that felt like unnatural conversation.

Scribe is intelligent, obviously, it will listen to everything, so you can talk about the weather, you can talk about what bus route the patient took to come in, you don't need to worry about framing things like you would a typical dictation style. It will listen to everything and filter it intelligently through our LLM, so we'll go ahead and pause there.

The patient can leave the room, you can say things behind their back that you might not want to say in front of them, but particularly the case if you want to verbalize examination findings.

For example, I noticed the patient had tar -stained nails. I suspect they smoke a lot.

They were also visibly distressed, short of breath, sweaty. They looked terrible. I'm worried about them.

And once we're happy that we have all the information that we need, we go ahead and finish and create note.

That goes off to our correction layer, which has been sensitised essentially to medical terminology, pulling information from UK resources, essentially any language or voice you can imagine being online so we can translate 43 languages or so. And as you can see there, it's immediately pulled all of that information through the

template, which has predefined headings and prompts to essentially say put this information in this section.

There's loads of customisability that you can achieve through prompting your own templates, and these are for individuals who have kind of service -level access, clinical leads, service leads, et cetera, to make sure that there's a standardised level of documentation.

From Notes to Letters and Record Updates

And now we want to actually do more than just this.

We want to make a letter out of this. We want to send the patient home with a discharge letter, or we want to refer this patient on to a different specialty.

Again, we can do that by simply going ahead and creating a note, we'll select outpatient letter for the patient, and again, pulls that information through almost immediately, writes it as a nicely written long prose, no medical jargon, addressed to the patient letter.

If there's any clinicians in the room, I hope this looks somewhat familiar. And then we can even use this magic wand to give it stylistic and content change requests, requests, and it will go through the same LLM correction layer to filter those through.

Once we're happy, we get a prompt. Interesting. Let's take a note of that. So what does that actually say?

We went to copy the note so we can paste it onto the electronic record. Remember to review before transferring. We'll talk about clinical safety in a second, but this is essentially what I was highlighting in the beginning.

Clinical Safety, Verification, and Trust

What are the verification layers when you're using AI in a clinical workspace?

So we'll continue anyway. We're We're confident.

Not best practice. Don't do this.

Beyond Transcription: What Comes Next

Now this is all exciting if you were seeing it for the first time a year and a half ago. What's actually next? Ambient voice technology has made it smart on healthcare and it's here to stay.

But what is next? What more is achievable?

So chat. Obviously anyone can use chat GPT, Claude, whatever it may be, to ask medical queries. series, whatever, but how would you feel if your doctor was using ChatGPT and XE?

I would feel very uncomfortable, because there's no protection, there's nothing to say this is relevant to the UK guidelines, there's nothing to say that the information is pooling isn't just from some random resource on the internet.

This is certified as a medical device. So if you imagine heart valves, surgical tools, things that would have been traditionally validated as medical devices, we're treating the software the same way.

So this is a Class Class 2a under EU MDR, if anyone knows what that means, it's essentially as good as it gets for certifying medical device regulation.

So we're going to ask it, is my plan consistent with the guidelines? Do I trust Lydia's plan over here? Of course I do. But I just want to check what these guidelines have to say.

And it will pull information from UK guidelines as well as any guidelines that I have uploaded to it.

So I might be working in a private clinic or NHS hospital that has its own set of guidelines. It will pull information from that. that.

So broadly consistent with the guidelines, consider adding to the plan these other things. And it's referenced them for me as well, so I can immediately source check to make sure it's accurate.

There's way more fun you can have with this. How much more are we doing for time?

How much more can we play around with it for? Another five, ten minutes? Okay.

So we'll wrap up the demo part. I hope that's been enough to give you an idea of what is possible, okay? What is actually realistic now?

So we'll go back over to the slides. And again, we'll have to switch screens. Apologies for this. Change extended displays. Okay.

Getting to the Front Line: Procurement, Security, and Adoption

So what's realistic? This is realistic. Once again, like I said, this is being used now now in NHS organisations across the country, private organisations across the country.

We have the largest market share for GPs in the country. What does it take to get there?

So as I said, the AI is the cool part, the stuff we're showing you is all fun and exciting, but what does it take to get there?

Procurement and Contracting Realities

Procurement cycles are long in the NHS. It's a competitive bid.

You have to show up with evidence, you have to show up and compete against other suppliers to show why you should be selected. That can take months, if not years.

Contracting and data agreements built into that, so you need to make sure that you're secure. All of our data is on an EU database.

We don't store any audios deleted immediately after the interaction. Everything is compliant across EU standards.

Integration, Deployment, and Change Management

Deployment and integration. We do integrate with several EPRs, 100 -plus record systems there, actually, and that allows us to transfer the note directly into the system, not needing the clinician to copy and paste from one workflow to the other.

That takes a while to build. And then getting people to actually adopt the program is a whole change management process.

You need to influence behavior on a scale that is larger to imagine with clinicians who are really quite rooted in their old ways oftentimes, typing while they're sitting next to the patient. patient.

So, I won't bore you with this slide. This is kind of what we've touched on already. Practical considerations look like this.

All of the licensing that we have, you know, fought night and day to achieve on one slide that summarises it to make it look a lot easier than it is.

Conclusion and How to Get in Touch

And so, with that, a brilliant transition to entertain you. If you haven't already been entertained enough, and a QR code to the website if you want to find out more information.

These are our emails. Feel free to get in touch.

If there's anyone from healthcare in the room, come and say hello afterwards. We would love to get to know you.

And thank you again to Lydia, and thank you.