Scaling cardiac ambulatory services with a hybrid approach of human expertise supported by machine learning

Mark Hashemi is the founder and managing director of Technomed Limited, a UK-based company delivering cardiology data analysis services to healthcare providers. Technomed has developed the ECG On-Demand cardiac reading service, and specializes in 12-lead ECG interpretation and ambulatory ECG Holter analysis using a hybrid approach of machine learning (with Cardiologs’ AI) and human expertise. Their current customers include general practices (GP), hospitals (NHS and independent sector), and consultant cardiologist private practices.

In 2018, Technomed became one of Cardiologs’ first customers. We met with Mark to discuss Technomed’s ECG On-Demand service, the reasons why they chose Cardiologs, and the benefits that AI-based ECG analysis has brought them.

Technomed & ECG On-Demand at a glance

Mark, what is ECG On-Demand?

ECG On-Demand® is our cardiac ECG reading service. We’ve been providing this service at scale for around four years, and the last couple of years, things have really taken off. At the moment, we’ve got 15 cardiac physiologists, doing a combination of both ambulatory Holter recording analysis and 12-lead ECG interpretation. We’re probably now the largest Holter analysis provider in the United Kingdom. We’re doing close to 1,000 Holters a week and we’re still growing.

Who are your customers in the UK?

We serve both the private and the public National Health Service (NHS) sectors. We’re very strong in particular in the London private practice sector, but since 2020 our NHS service has grown significantly.

We are working with high caliber NHS hospitals, such as the Royal Brompton and Harefield hospitals, Barts Health NHS trust, and Liverpool Heart and Chest Hospital. They have high profile operators there, who are very satisfied with our services, and that’s probably due to the quality of our reports, both from the standpoint of nice-looking electrograms and excellent report summaries. They also seem to appreciate being able to talk to the person who wrote the report. Most of our cardiac physiologists have had senior roles (band 7+) within the NHS Cardiology departments, so they really understand what they’re talking about, and they can communicate their findings more effectively to the end customer.

How does your service work?

We have two main user scenarios. Our primary care customers are looking for somebody to overread their 12-lead ECGs to make sure they’re not over-referring into secondary care or the hospitals. So in this situation, we’ll put an ECG machine inside a GP practice, and this ECG machine is networked to a server, so that anything that it records is automatically received and checked by us.

The majority of GP practices have their own ECG machines, however, unless the GP is really confident, what they tend to do is rely on the onboard algorithm of the ECG machine. So when the ECG machine says there’s a serious abnormality, the GP will generally automatically book the patient in for a hospital appointment. What we found is that around 3/4 of those abnormalities are actually false positives, because artifacts are often identified as arrhythmia. By using a cardiac physiologist to overread the GP practices ECGs, we tend to reduce the secondary care referral rate by about 75%. This huge drop gives you an idea of the over sensitivity that these traditional ECG algorithms suffer from. Without our overread, they probably would have been referring almost half of all their patients into the hospital.

We also propose to equip those general practices with Holter monitors, so that they can fit the recorder locally and bypass hospital waiting lists. Pre-COVID-19, it was not unusual to wait 6-months for a test and now the waiting time is longer. Using our service, Holter monitoring is often performed the same-day or next day. Customers upload the ECG recordings into our servers, and wait for our report back to them. Our average turnaround time at the moment is between two and three hours, which is pretty spectacular versus the traditional pathway.

What about your hospital customers?

Secondary care or hospital customers don’t tend to need 12-lead ECG overread, because they already have staff that can read 12-lead ECGs. Their main problem usually is that they don’t have enough cardiac physiologists to do the Holter analysis. So they will have long waiting lists. The other problem, especially over the past year, is that they don’t want to bring patients into the hospital and patients don’t want to go to the hospital. That’s why we operate two different models in this setting.

The standard model is where the hospital fits the recorder on the patient in the clinic. This represented the majority of the work before the pandemic, but since March 2020, there’s been a large explosion in requests for us to ship monitors directly out to patients. These monitors are delivered in a kit, with comprehensive instructions for the patients to fit the monitor themselves. The patient then drops the monitor at the post office, and it comes back to us via a courier service. In this case we do the data upload for the hospital, and deliver the final report to them in the same way as the first model, so that the hospital actually never sees the patient, only their test result. Pleasingly, we found that 19 out of 20 patients can manage to fit the recorder themselves without any problem. What’s more, from an electrogram quality, they’re virtually indistinguishable. Patients are more than capable of fitting the recorders to as good if not better standard sometimes, probably because they spend more time fitting and are invested in their results.

So in the end, less than 5% of the patient population still needs to go to the clinic. I think direct fitment is here to stay. Whether or not it will outgrow the traditional route, I don’t know, but the only thing I would say is that we’ve now decreased the cost to fit a monitor at home down below what it costs to fit a monitor in hospital and it’s more convenient.

The switch to Cardiologs

What challenges were you facing prior to using Cardiologs?

We’ve been through quite a number of devices over the last 10 years. Cardiac monitoring devices are changing all the time, and the devices are associated with a certain software package. So we have used plenty of solutions, such as Mortara, CardioSpy, Scottcare, Medical Algorithmics, etc. and of course have considered SpaceLabs. In the last two or three years, there’s also been a wealth of wearable technology. Unfortunately, one doesn’t really work without the other, both the device and the software components are critical, so you can have the nicest device in the world but if the data analysis package is poor, it doesn’t work. Trying to find that combination of good devices and software is difficult. So we’ve been very lucky within the last four years to find an excellent device supplier that provides a really well engineered device and an excellent data analysis supplier, Cardiologs. The Cardiologs system is particularly attractive as it interoperates with various devices.

What was the main reason for transitioning your ECG analyses to Cardiologs?

We started using Cardiologs about four years ago. Right from the start, we loved the user interface design, which was really well thought out. The user interface had been designed with a really fresh approach and no preconceptions. It was nice and simple, because the company was focused on the core requirements of a Holter analysis system, rather than the peripheral bells and whistles, which are used only 1% of the time. In a lot of the older analysis platforms that we’d used before, we probably only ever used less than 50% of the features. Those solutions were very inefficient, because the core components were lost in the middle of all these unused features.

So what we were looking for was an easy to use interface that had an equivalent performance. A really good cardiac physiologist should be able to produce a good report, irrespective of the Holter analysis platform that they use, but what will vary greatly between different analysis platforms is the amount of time it takes to produce that report.

How did Cardiologs change your business?

Initially, we wanted to make sure that the software worked well both in terms of performance and efficiency. So we tested Cardiologs, by taking a fixed data set of 100 Holter recordings that we fed through an old fashioned Holter analysis system and compared it with review using the Cardiologs system. Both systems were used by the same physiologists. We found that both systems gave the same results, which was what we expected. But we discovered that there was quite a substantial difference in analysis time between both solutions. On average, there is a 40% reduction in analysis time with Cardiologs, with the time benefit even higher with longer recordings. That was the key thing that really convinced us to switch to Cardiologs.

The medical community recognises now that the longer you monitor, the more likely you are to pick up the rhythm disturbances. So there is a move towards extended monitoring periods, and therefore we needed to have more efficient data processing systems in order to keep our costs down and remain competitive, while delivering longer monitoring sessions.

When we have hospitals that come and visit our facility, they’re quite staggered at the amount of work that the physiologists can now do in a day. Most of our physiologists are doing around 25 Holter reports a day plus a load of 12-lead ECGs, all with a wonderful report quality.

How does the Cardiologs algorithm perform?

Cardiologs’ AI helps a lot with the rhythm classification and the isolation of the interesting strips. You still have to confirm its selection, but you don’t want to be spending hours relabeling ventricular tachycardia to artifact, you want a system that doesn’t need much correction, and that’s what Cardiologs is about. What we find with machine learning algorithms is that the artifact rejection is pretty spectacular, and it makes a real difference.

What about the Cardiologs 12-lead solution?

The main reason we are using Cardiologs to support our 12L ECG overreads is for quality assurance. We don’t rely on the algorithm, but it certainly makes it more difficult for our physiologist to make a mistake. Whether one likes it or not, humans are unfortunately subject to human error. So what we’ve really tried to do is to reduce the probability of error, and we certainly go to great lengths to get it down to an absolute minimum with Cardiologs. The algorithm makes an initial finding, and our physiologist needs to confirm or reject the finding. What we find is that the amount of editing required using Cardiologs’ machine learning algorithm is substantially less than conventional ECG machine algorithms.

Which benefits has Cardiologs brought to ECG On-Demand overall?

First, Cardiologs allowed us to scale. We wouldn’t be in our position if it wasn’t for Cardiologs. Last year, we grew fourfold, and this year, we’ll probably be disappointed if we don’t double. Our customer retention rate is pretty phenomenal. In the last five years, we’ve only lost one customer, and we pick up lots of customers from our competitors, with literally no salesforce, most of it by word of mouth. I think our clientele have worked out that our services are equally as good, if not better than the competition, and we deliver it locally, faster, quicker and at less cost.

There’s two routes that you can use to increase productivity and reduce costs. You either adopt new technology, or you outsource the analysis overseas, which is pretty common practice in the industry. The latter is not an option that I’m particularly keen on. We’re pretty passionate about keeping healthcare in Britain for our British customers, and the main reasons being quality assurance and regulatory oversight. The UK healthcare regulatory authority will not inspect a clinical service, if that clinical service is outside the UK. Cardiologs has helped us with keeping our clinical service in the UK and not needing to outsource overseas, by increasing our productivity.

Another benefit that we’ve seen in using Cardiologs is quality. From the report quality standpoint, the Cardiologs Holter analysis report formats are some of the nicest I’ve seen and it makes them very easy for the customer to understand. And then from a data quality point of view, there’s no doubt that the algorithms have helped. I would be quite surprised if we miss something horrific in our Holter analysis or 12L ECG work. The probability of a false negative is extremely low with Cardiologs.

With Cardiologs, we can provide a first class quality service for more people quicker, within the same budget. Because whether we like it or not, European healthcare systems tend to have fixed budgets. The whole purpose of using machine learning is to deal with the increase in demand from an ageing population, and hopefully keep costs at an affordable level.

How satisfied have you been with Cardiologs’ customer service and cloud services?

Customer service has been faultless, it’s an absolute pleasure to work with the Cardiologs team. One of the initial concerns for us was being reliant on a third party completely for operation. Basically, if the server went down, our service would go down. Having said that, in four years of working with Cardiologs, I can’t even remember when the application has been down, and it’s probably been for a minute or two. It’s a very stable system.

How does your team feel about Cardiologs?

All I would say is, I think they would be probably quite upset if I took the Cardiologs software away from them. It certainly does make their lives more pleasant.

For more information about Cardiologs, visit www.cardiologs.com

Legal & Regulatory mentions

Medical Device
The Cardiologs ECG Analysis Platform is a medical device intended for use by qualified healthcare professionals for the assessment of arrhythmias using ECG data in subjects over 18 years of age. Class IIa in Europe (CE2797) in compliance with the Medical Device Directive (MM 93/42/EEC amended by 2007/47/EC).
Class II in the USA according to the 510K clearance.

Warnings
This material with associated pictures is non-contractual. It is for distribution to Health Care Professionals only and should not be relied upon by any other persons. Carefully read the instructions for use prior to use. This material reflects the opinion of Health Care Professionals, not the opinion of Cardiologs.

Headquarters
Cardiologs Technologies SAS
136 rue Saint Denis, 75002 Paris
France

© 2021 Cardiologs Technologies — All rights reserved. MM-21–08 V1.00 (2021)

--

--

--

Our mission is to democratize access to expert cardiac care through medical-grade artificial intelligence and cloud technology.

Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

TechCon 2020 is here 🙌🏻

Why do You Need to Use AI in Digital marketing?

All About AlphaFold — Possibly AI’s Most Important Breakthrough So Far

How Fredrikson & Byron pivoted to Enview by Civic Eagle to deliver with speed to its clients

Machine Learning Vs. Artificial Intelligence

H+ Weekly — Issue #87

H+ Weekly - Issue #159

What Are Deepfakes and How Are They Created?

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Cardiologs

Cardiologs

Our mission is to democratize access to expert cardiac care through medical-grade artificial intelligence and cloud technology.

More from Medium

LIGHTING THE FUTURE WITH FLARE

Mind your REGEX or it can put your program into an infinite loop

REGEX that matches the chours of “Never going to Give you Up”

How to Build a Brand Using Jewelry Box Model

Don Troiano CEO Our Fashion Passion — Review 2022

Don Troiano Review