Personalised medicine : PreMeDICaL, a new joint Inria-Inserm project team in Montpellier

PreMeDICaL is a joint team between Inria and Inserm, and more specifically with IDESP (Institut Desbrest d’Épidémiologie et de Santé Publique), a joint research unit between Inserm and the University of Montpellier.

It is an interdisciplinary team, composed of researchers in machine learning and statistics but also clinicians.

PreMeDICaL stands for "Precision Medicine by Data Integration and Causal Learning".

The name refers, first of all, to the field of application. It is also a nod to the studies carried out by students before becoming a doctor.
It indicates that we are upstream of care, developing new strategies to improve patient care.

Finally, it allows to precisely define the team's themes with the integration of heterogeneous data and causal learning, creating a perfect bridge with IDESP, whose objectives are to optimize and model prevention and care of chronic diseases with massive health and environmental data.

We develop causal inference methods that allow, among other things, to establish optimal treatment policies, i.e. to know which patient to treat and when, by combining different sources of data: data from randomized clinical trials and so-called observational data such as data routinely collected by hospitals. The integration of heterogeneous data is accompanied by numerous methodological challenges such as the management of missing data.

The team's objective is to bridge the gap between theory and practice and to implement the tools and methods developed in clinical practice. Interdisciplinary work is therefore at the heart of our activities.

We study in the first place respiratory diseases (the World Health Organization (WHO) predicts that in 2050 one person out of two will suffer from allergies) and in particular asthma, one of the specialties of the team members and of Montpellier.
Our links with the IDESP allow us to access databases and doctors who contextualize the results.

At present, treatment strategies are far from being personalized.

When the results of a clinical trial indicate that a treatment is effective on average, this treatment will generally be given to all patients suffering from the disease in question. However, there is very often heterogeneity of response to the treatment: that is, some patients may benefit more or less from the treatment.

The methods we are developing, which use complementary data sources, make it possible to move towards greater personalization and even to predict the effect of a treatment in a population that has not been studied in a randomized trial. This makes it possible to accelerate the availability of therapeutics on the market, which is crucial, particularly in times of health crisis.

The impact is important for the health of patients but also from an economic point of view, as it can have a direct impact on drug reimbursement policies. Of course, beyond the statistical and machine learning methods which, combined with the data and the business expertise, allow to have new evidences on the effectiveness of treatment. It is imperative to work with all stakeholders such as health authorities to transfer research results into the public domain.

The team members are on several sites but we do regular updates. It is still a little early to evaluate the implication of having several institutes, in particular for contracts, recruitments, etc.

For the moment, this diversity is a source of richness because it facilitates exchanges with different scientists, for example with members of the IDESP research axis led by Isabella Annesi-Maesano, on the understanding of environmental determinants of chronic diseases (Exposome).

We have many links with industry, in particular with the pharmaceutical industry (Sanofi, ALK), medical equipment developers (General Electric) or consulting firms (Capgemini Invent, Quinten).

This is mainly done through CIFRE theses or research contracts. Pascal Demoly is also the co-founder of a start-up AdviceMedica (a collaborative self-help solution for doctors).

We want to continue to strengthen our collaborations with industry.

The field is growing rapidly and researchers from different communities are interested in the problems of causality and data fusion.

The field is therefore bubbling and arouses a certain craze, which can be explained by the societal context and by the associated challenges, which are both theoretical and applied.

Integrating the notions of causality in machine learning offers promises to alleviate the problems of these methods in the face of distributional changes, for example, which would make them more robust, and machine learning can also bring new perspectives to the classical methodologies used in biostatistics to evaluate treatment effects.

However, even if a lot of work is motivated by applied considerations, there is little concrete impact on clinical practice. We hope to achieve this goal through software development, training of the various actors, and the integration of the organizational and structural constraints of the field from the project design stage.

I co-supervise with Nicolas Molinari, another member of the team, a student who has a medical thesis to do a PhD in statistics. This type of profile is crucial because it is imperative to have intermediaries between clinicians and more mathematical profiles, who are able to understand the issues and constraints of each, and demystify certain methods, in order to carry out projects successfully.

On another project, a PhD student in statistics (quite theoretical) works hand in hand with a clinician to define dynamic treatment policies and we meet regularly all together. This work will hopefully result in a nice project with theoretically based results that could be adopted by the medical community (we use information that is available in practice).

The research work is always motivated by a practical question in the application area, we then develop generic methods that can be applied to other areas and then return to the original question to propose practical solutions. Iterations between clinicians and methodologists are constant.

I am involved in the R statistical and data analysis software community. I participate for example in the organization of international conferences on the subject, develop libraries. I am part of Rforward, which aims to increase the participation of women and other underrepresented groups.

I am also coordinating a project on the management of polytrauma patients, in collaboration with the Traumabase, EHESS, and École Polytechnique, started before I joined Inria.

Finally, like all researchers, I participate in numerous committees.
But there are too many solicitations as a woman, because of the imposed quotas, with an effect that does not seem to be significant on the presence of women in the field (I would be happy to analyze the data on this subject).

I plan to replace some of these committees with more communication to the general public and to young people in particular, as I am convinced of the need for more awareness of science and scientific careers.