Our Customers
Products & Pipeline
Science & DTx
For business
About us
Contact us
US Payers
next arrow
4 Min Read

From Real World Data to next Generation Mental Healthcare: my presentation at #AMLD22

30 Mar 2022
Valentin Tablan

Last month, I spoke at the #AMLD22 session on Applied NLP Technologies for Physical & Mental Health in Practice.

I would like first to thank the organisers for putting mental health on the same level as physical health. For too long we have been maintaining this dichotomy between physical and mental health which doesn’t make a lot of sense.

My talk, entitled “From Real World Data to Next Generation Mental Healthcare”, was about how AI can help drive progress in mental health. A canonical example of artificial intelligence, that is very well covered by our public discourse, is that of self-driving cars, so I wanted to use that as an anchor point. In this area there are two tiers of functionality: autonomous driving, where the car drives itself; and assistive features, where the car offers help to make driving simpler and safer. Underlying all those features sits a deep understanding of how cars work, the mechanics and physics of converting energy to motion, and controlling that movement.

The three directions of our AI work at ieso

Our AI work at ieso can be viewed along the same three directions: developing a fundamental understanding of how therapy works; building tools to support our therapists in the delivery of high-quality care; and building tools to automate elements of care, allowing us to increase the dose of therapy and relieve some of the pressure on the overstretched therapist workforce.

Uncovering the fundamentals of therapy

We do many types of work to uncover the fundamentals of therapy, but my talk focused on our work in natural language processing. I mentioned the work that Stephanie Brown and colleagues do on automatically detecting cognitive distortions in patient language. These are patterns of unhelpful thoughts that are known to associate with mental health conditions. In our work, we found eight different cognitive distortions that have statistically significant associations with different mental health diagnoses. This work is currently going through peer review, so watch out for a new paper on our papers page. I also covered our work on analysing therapists’ and patients’ language in therapy sessions that has already been published here, and here.

The tools we have built to support our therapists

Building on our work to understand the fundamentals of how therapy works, we have constructed a set of tools that support the delivery of good quality care by our skilled clinicians. These include the ‘Case progress monitor’, a statistical model that can monitor the changes in each patient’s symptoms and notify clinical supervisors when the patient does not progress as expected. The supervisors can then work with the treating therapist to make changes to the protocol and hopefully bring the case back on track.

Automation – why?

The last part of my talk covered the R&D work we do to automate elements of patient care. A problem faced by all mental health care systems in the world is the limited numbers of trained clinicians, which is evident even in wealthy parts of the world. Training more clinicians is clearly a part of the solution, but it is unlikely to be the only required solution given that the current access rate is growing towards 25% in the well managed and well-funded IAPT programme of NHS England. Some of the work we have been doing in recent years is to develop conversational agents that are capable of delivering elements of care directly to the patients.  

One of our clinicians told me that what happens in the therapy room is like a patient getting a prescription for a medicine from their doctor. The actual healing takes place in between therapy sessions, when the patient puts into practice the strategies they were given, when they do their activities and exercises, and internalise change. There is ample academic research that indicates that the more time patients dedicate to their between-sessions activities then the more likely they are to get better, and the faster their recovery will be.

Where next...

For the last few years, we have been running an internal R&D programme to develop conversational agents that are capable of automatically delivering care, supplementing the work that therapists do with the patients, and increasing the ‘dose’ of therapy that patients receive without requiring additional clinicians. These tools are currently being trialled with some of our patients, and the early results indicate a high level of patient engagement. On average, patients who have been given access to our therapy companion tools interact with them to a level similar to an additional therapy session. Our trial is still ongoing, so we don’t yet have data about clinical effectiveness, but these engagement levels, when considered alongside previous research on the impact of between-sessions work, are promising.

Valentin Tablan
From Real-World Data to next Generation Mental Healthcare: in this article, our Chief AI Officer Valentin Tablan summarises his talk at Applied Machine Learning Days 2022 for those who missed it