REALMENT is a four-year project, funded under the EU’s Horizon 2020 programme. Its aim is to improve existing treatments through the better use of data – mainly by developing methods which allow the vast amount of data stored in registries and biobanks to be read alongside the more tightly defined data produced by clinical trials (RCTs). The project is now reaching the end of its first phase. To mark this, two REALMENT researchers, Professor Ole Andreassen and Dr Elise Mirjam Koch, both from the University of Oslo, talked to ECNP press officer Tom Parkhill about the project.
Tom Parkill: Professor Andreassen, tell me a little bit about REALMENT; where did the idea come from?
Ole Andreassen (OA): Perhaps we can begin by thinking about personalised treatment in psychiatry, which is difficult to achieve with current tools and knowledge. We want to put forward better science and data integration, to move psychiatry into precision medicine. We have several drugs today that are effective, but they have side effects, and they don’t work for everyone. So that's the low-hanging fruit REALMENT aims for in the short term, let’s say: improving current treatment. Afterwards, we hope to be able to develop new drugs but that will take 20 years.
The question is, how do you move forward, and how do you discover who will benefit from treatment and who will not? And of course, who will get side effects and who won’t? At the moment there’s a lot of trial and error in psychiatry, and it’s expensive; if you want to start new clinical trials, you quickly run out of money. There’s not enough funding in the world to fund all the clinical trials needed to unpick this information. So I said, OK, let's try using real world data, which we have in abundance, which we have in biobanks and registries.
With real world registries you have huge amounts of data. You can develop AI models and train the models to interrogate big data. We have records of genotypes, we have millions of variants, we can understand different risk factors and outcomes, and of course we also have information about which drugs people have been treated with. We have data from nearly two million people across the Nordic countries and the UK, and that is the basis of REALMENT. We developed new AI and modelling tools to discover the trajectories of response and non-response. And then we validated these via smaller, genotyped RCT studies, and then finally we put this into an electronic management platform, otherwise doctors wouldn’t understand it. If you are using big data, then there’s just too much information, so you need a management platform to guide the doctors to select the best treatment. This is our aim and our story.
Elise Mirjam Koch (EMK): The name REALMENT refers to real-word data and mental disorders. By real-word data, we refer to data sources that are not clinical trials, such as health registries, electronic health records, hospital records, electronic health records, laboratory databases, and self-reports.
It sounds very complex to reconcile your two data sources, registries and clinical trials, and of course clinical trials would normally exclude anyone who didn't meet very stringent parameters, whereas registries and so on tend to be much more inclusive. How did you manage to reconcile that?
OA: You are right – registry data includes everyone in the population, and by comparing with clinical trials, we aim to better integrate these two types of findings. The beauty of real-world data is that you include the placebo effect; the placebo effect actually helps you. You’re not in the situation where the patient is thinking “You're gonna give me this drug, but I'm not sure if this is the real thing or not”. It’s much more like we are saying, “Here's the drug for you. This is a good prescription”. So that’s the real-world situation. In RCTs you have a much more stringent, much more controlled study, but they're not so representative. So, by using AI tools to look at the whole population we get a more general, predictive tool set. RCTs are just there to show significant difference between the intervention and control group, but real-world efficiency is what we are concerned with in the clinic.
EMK: I think of course there are both advantages and disadvantages to both data sources. It's true that the real-world data is affected by heterogeneity and confounders, but also I think it's a huge benefit to have this. Real-world scenarios include factors like polypharmacy and comorbidities, these are considerations which are not so well captured in randomised controlled trials. Real-world data is such a rich source of information. For example with registries, we can not only estimate treatment outcomes or treatment responses. We can also look at long-term outcomes such as hospitalisations, or functional outcomes like employment and so on.
REALMENT focuses on the optimisation of existing treatments for schizophrenia, bipolar disorders, and major depressive disorders. What are your successes? Is there anything that you can translate directly to the clinic, which doctors can start using?
OA: We're still finishing our studies, but we have preliminary data now. With antipsychotics, for example, we have ongoing work which has obtained a better prediction of the treatment response – which we hope will improve even more. In addition, with a new tool that we are deploying to clinicians, we can better predict cardiovascular and metabolic side effects. Thus, we can now say, OK, there is a likelihood of you developing increased BMI or cholesterol or lipid disturbances if you are treated with this drug. That is very close to being a finished product. What was needed was a tool which would work in the clinic. I think that's been a good success story for REALMENT. We have an SME [small-medium enterprise] which took our discoveries, for example, on cardiovascular disease effects and turned them into something clinically useful. This was a low-hanging fruit, given that there’s a lot of good information already out there for cardiovascular events. So, for example, if a patient starts antipsychotic treatment, you would like to know how much weight will this patient gain, or will they have cholesterol changes. So that I think is the most clinically relevant outcome so far.
Obviously, the clinical trials tend to belong to drug companies and the databases tend to be much more open and available. Are there any practical difficulties you will work with? Did you actually have to work with specific pharmaceutical companies on this?
OA: We use biobanks and registry data from the Nordic countries and UK. They don't openly share any data, but you get access to it as a researcher and then you can work according to a “federated learning” system. Let’s say you develop a model in Sweden, add data from Denmark and validate in Norway and Finland, this concept is federated learning. So we can analyse the data across all of these different sites and then bring them together, a bit like a meta-analysis. And then the effect sizes for the different markers or genes or family history, diet, lifestyle or whatever can be translated into the clinical trials, which of course are much smaller in size. But you're right, we're still working on finalising the validation in the clinical samples. Changing drug priorities was more of a practical challenge which meant that we had to readjust the project, but we have other resources.
So you come up with the idea, I guess four or five years ago and then you took it to the EU for funding.
OA: This is a Horizon 2020 project, and we gathered a team of strong collaborators from academia and industry.
You’re reaching the end of the end of the process now, what happens? Are you looking for another round of funding, are you looking to take it forward in any way?
OA: Definitely. In the Nordic countries and with the Brits we have now developed this infrastructure which leverages the health registries and biobanks. That pipeline is quite complex, you can't have one PhD student building and maintaining that kind of infrastructure; this is a big structure, a big project was needed to build up the infrastructure, so now we will follow up and use it for more projects and expand it to other areas. We have a series of European and Nordic and national projects building on this infrastructure.
EMK: There are several projects we're planning, and indeed already working on, where we apply what we have learned from REALMENT. There’s also an NIH project.
So you want to expand your geographical input. You mention existing relations with the Nordic countries plus the UK, but you also mentioned the NIH.
OA: We are part of a funded project where we contribute with the infrastructure that was developed through the EU projects. This is seen as highly valuable for research, and until recently, the US NIH grant also funded this type of work. However, the future is more uncertain. Our approach started as a northern European project, but we now also have Italy as a partner. We want to make sure that they are aligned, and that what we're developing in the healthcare systems of Scandinavia is also validated in southern Europe.
What was the ECNP's involvement in this?
OA: ECNP has been responsible for dissemination, which has been very good, as ECNP can promote REALMENT via meetings and other types of professional dissemination. My impression is that ECNP is very good with communication, and that has worked well for us. We are now getting to the final phase where we have real results to report. We’ll be working with ECNP on dissemination, to make sure the findings reach the right people. In November ECNP will host a free a webinar series on the ECNP Knowledge Hub with the REALMENT highlights. Keep an eye on ECNPs social media account or the website for updates.
EMK: ECNP has also been involved in the task force to harmonise treatment outcome definitions from real-world data, which has been going for a year. We’ve been working with ECNP and with other EU project teams, to make recommendations on how to define treatment outcomes from real world data sources.
What do you think were the difficulties in putting this project together and running the project? EMK: One thing we noted were the great opportunities from real-world healthcare data in Nordic countries. But you need to be aware of some specific technical differences in types of data between the countries. Data harmonisation is important. In Scandinavia, we have a lot of data in the registries, but this amount and quality level of registry data is not seen in other European countries.
So this would be one of your take-home messages, that we need to set parameters for data harmonisation?
EMK: Yes, I would say we do have to standardise real-world data to enable replication and meta-analyses.
OA: In this project we had a real advantage: we have known one another for many years, and we know how our partners work. Stable teams together with young scientists are also important. As regards challenges, COVID hit us quite badly, it delayed us a lot, and we are still trying to overcome some of the difficulties from that time. And it took some before the partners settled to their different ways of working, but once we got going it was good. I guess most projects are like that.
There is a lot of promise in using real-world data to develop new AI tools for precision medicine. One thing we learned is that AI tools should be targeting a clinical use case, as undertaking precision psychiatry is context dependent. Each of us have different kinds of questions and different data uses. So you also need a certain amount of “precision” in the way you apply these precision tools. This was a new thing for me: I hadn't realised how targeted the clinic is towards specific use cases.
One of the most successful parts of what we did was getting companies to sit together with scientists and stakeholders and ask, “What do you want to predict? What is your clinical setting? Are you working in an acute ward or an outpatient clinic?” If you are an emergency ward you don’t think so much about chronic conditions, you need to treat people here and now. If you're in an outpatient clinic, your patients have higher levels of functioning and less severe symptoms. It's obvious when you articulate it, but for the planning of the project, we did not build in that kind of granularity on the practical clinical use.
EMK: I’d add just one thing, and that is that we can combine data from different sources – different data from different registries. We can also use self-reports or even clinical notes from text extraction and then combine the data to get better prediction. I think this is a really great data source.
So I guess REALMENT is actually the first major project to actually look at combining these data sources.
OA: Yes, I absolutely agree. Precision medicine is not only building on one modality: it’s not just “look at brain imaging, that's the only way”, or “genetics is all you need to do”. We need to combine our resources and apply multidisciplinary approaches. This will eventually make us able to implement a precision psychiatry framework.
Thanks very much to you both, great to talk to you.
*****
You can find more information on the REALMENT project here. 
