IESO Digital Health: digital therapy can provide the same level of impact as a human therapist can
“We are riding a wave of behavioural change and thus have a chance to reach people who would otherwise not benefit from therapy.”
In the past 4 years, IESO has been undergoing a transformation from a technology enabled cognitive behavioural therapy (CBT) provider to a much more ambitious AI-based solution. Does this journey
surpass your wildest dreams?
Andy: Actually, it has been a long-term vision of ours to build a massively scalable solution to address mental health problems around the world. The first challenge was to provide mental healthcare to people wherever they are and whenever they need it. Then we used the data produced in this process to find out what works and how we can improve the quality of our service. In our core market, the UK, we were already able to achieve this. But if you look at the problem on a global scale, these mental conditions are literally everywhere. There is probably no family on this planet unaffected by mental health problems such as depression or anxiety. In some areas of the world, even less than one mental health pro fessional is available per 100,000 people in the population. This is why we really have to make a difference. I believe that using artificial intelligence, data and new technologies is the way to go. Right now, we are experiencing a convergence of enabling factors such as large volumes of data, various new consumer technologies, computational power and deep neural network based approaches. In addition, the majority of people in the world have a computer sitting in their pocket. This gives us an unprecedented opportunity to reach them. I know it is a very audacious goal we are pursuing, but now is the time to do it.
So it is a combination of internal and external achievements that made this goal come into reach?
Andy: Correct. We already had this vision when Ananda Impact Ventures made the first investment in IESO. It was fantastic how they supported us. It was always clear for us why we had to pursue this goal and the type of approach we would be using. But the specifics of how we could do it were still rather vague in 2015/16. This is why Ananda’s belief in us has been absolutely central to our success. IESO probably wouldn’t exist otherwise and we’re enormously grateful for it.
Which major challenges are you facing today and where do you see your biggest advantage in the market right now?
Andy: Valentin always told me that automating systems can be effective when we observe how human beings behave and then replicate best practice with technology. What is unique about IESO is that we combine our huge high-quality data set with the knowledge from tracking what happened with each patient who was ever treated by us. That in itself is a great starting point, but there is an enormous engineering and clinical validation challenge ahead. We will have to provide evidence on a very rigorous level and prove that our approach is not only clinically effective, but also cost-effective and scalable. What is important is that we have a very precise idea of what success looks like. For example, we know which level of clinical improvement we consider to be meaningful and how people should engage with our products. Particularly in the United States, there is an abundance of psychotherapies, both human-delivered and provided by apps. In terms of quality, it is sometimes hard to tell these apps apart, because there is a lot of fancy marketing going on. But we feel very strongly about these tools. If it were my son or my daughter in need and he or she downloaded something that is not evidence-based and clinically proven, I would be very concerned.
This is probably a concern shared by many potential patients and relatives when it comes to using a technology-based therapy solution. How large does your data set have to be to make sure that a “digital therapist” is clinically effective in helping patients?
Valentin: The size of the data set plays indeed a vital role. Ours is the largest set of conversation data between therapist and patient available in the world. My point is that if you look at the advances made in the past decades such as MRI scans or genomics, technology has supported people tremendously in getting better physically, but unfortunately not mentally. One of the main reasons is that relevant information for the mental treatment of a patient comes mostly in the form of language. You can easily do a chemical analysis of a drop of blood, but language is much more difficult to analyse. Fortunately, research work spurred by the development of new platforms such as Alexa or Google Assistant has led to the development of deep learning methodologies for language processing. Once you know how to analyse language, the challenge is to come to meaningful conclusions. This is exactly where our large data set is a major asset. Last year, we shared our first conclusions in JAMA Psychiatry, one of the leading journals in this space. Now that we can extract the right signals from the language, it is time to raise our game.
In psychotherapy, it is not only about the content of the language, but also about the emotional style or empathy with which we communicate. How far advanced is today’s technology in this respect?
Valentin: Given the right amount of data, the deep learning methods we are using today are definitely able to make a distinction between the content and the style of communication. These methods can even capture language elements that we, as human beings, are not aware of.
So it all comes back to large data sets to distil information and provide reliable therapy, right?
Andy: Exactly. There is another way to think about it. Imagine you are a therapist with ten or twenty thousand hours of sessions. Even with this level of experience, you will always face patients who are truly unique. An AI-based system can learn from hundreds of cases and from the sum of experiences of all psychotherapists within the network. Typically, a human therapist is also trained in a specific method, which gives the session with a patient an orientation from the onset. In a net work-based approach, the logic gets reversed: What is the right therapy method for a patient with a given set of mental problems? Such a system places the patient front and centre — a true paradigm shift. In addition, AI-based systems have the potential to continuously improve and capture elements that human-based therapy cannot. This level of scale and learning scope makes me truly excited. I know that data-driven psychotherapy may sound a bit cold and inhuman. But we have delivered a quarter of a million hours of therapy over the Internet and have witnessed that words transmitted this way can substantially improve the lives of our patients.
Where do you see the limits in terms of the type of mental health problems and patients to be treated?
Andy: Our solution is like a medical device or drug with a clear label: it will have precise conditions and severity levels that it is applicable for. For example, people with schizophrenia or dependency on opioids rather need complex, multidisciplinary teams to be able to improve their conditions. Our solution is clearly no panacea for everyone.
Valentin: We should also keep societal changes in mind. For example, we know that around 50% of patients who start face-to-face CBT do not complete or benefit from it, mostly due to a lack of engagement. A few years ago, hardly anyone was willing to use digital services for sensitive issues, for example mobile bank accounts, because people didn’t trust online services. In the meantime, this has changed dramatically. We are riding a wave of behavioural change and thus have a chance to reach people who would otherwise not benefit from therapy.
Are you in a position to also measure the longer-term impact that materialises way beyond psychotherapy?
Andy: We measure the outcome in the same way as a drug company would do to evaluate the efficacy of a new medicine. This means that we track the effect of every session since the beginning of our service in the same thorough way. As to how enduring the effects are, the healthcare system in the West is unfortunately very “transactional”: If you break a leg, you go to your doctor, and once you are recovered and released from hospital or therapy, you are good. Mental healthcare has been made to fit the same model, where patients are discharged following an episode of care, without any particular follow-ups being planned. Of course, digital therapy holds a promise that a longer-term measurement of effects becomes possible. But first we want to prove to the world that digital therapy can provide the same level of impact as a human therapist can — within the range of conditions and patients we discussed. So my message to you is: yes, we have big ambitions, but we are doing it in a very responsible way, step by step.
IESO is on a dynamic growth path. What impact do you anticipate on your organisational structure and team?
Andy: The number one priority will be to create a world-class team. This is my current obsession. We are very lucky that our mission attracts great talent and a lot of people want to be part of this. But it will radically alter the structure of IESO, since we need to move very fast while working in a highly regulated environment. Just like any business growing up we have to put many processes and systems in place. And since the world around us is definitely not standing still, we need to go like a rocket.
Are there major obstacles that you expect if you rocket upwards?
Valentin: If you are aiming to build something as innovative as we do, there are always challenges. One is to thoroughly understand the market, the other to master the regulatory requirements. All regulators of medical devices are aware of the fact that AI can be a great benefit to healthcare. They are solutionfocused and open to finding ways to make it happen as safely as possible. In addition, we intend to work directly and transparently with
regulators and subject ourselves to their scrutiny.
In this highly innovative field, where do you personally see the boundary between realistic vision and pure science fiction?
Andy: Futurist Carl Sagan said something in teresting
about ELIZA, the first language processing computer programme that
was developed by Joseph Weizenbaum at MIT back in 1975: “The human therapist, involved in the design and operation of the system, would not be replaced, but would become a much more efficient man since his efforts would no longer be limited to the one-to-one patient-therapist as now exists.” Today, this is not science fiction anymore. I have high hopes that we will get there soon.
Interview by Christina Möhrle
Dr. Andy Blackwell
Andy gained his doctorate in psychology from the University of St. Andrews before working as a medical research scientist in world-leading laboratories in the departments of psychiatry and experimental psychology at the University of Cambridge. In 2006 he became Chief Scientific Officer of Cambridge Cognition, a psychology and technology-focused business, and a year later joined the company board. Today at Ieso Digital Health Andy’s principle role is to develop and implement a global science and technology strategy to complement the commercial vision of the company.
Dr. Valentin Tablan
Valentin has spent nearly 20 years in the field of Natural Language Processing, Knowledge Representation, and Artificial Intelligence. Prior to joining Ieso Valentin completed his PhD from the University of Sheffield and was the lead scientist on the question-answering service that powers Amazon’s Alexa smart assistant. He is now using that expertise to help computers familiarise with the way patients express themselves during therapy. Under his leadership, Ieso has created the industry’s first (real-time) AI-enabled tools that augment therapists’ which in turn increases qualityand improves clinical outcomes.
