Datavers relies on interdisciplinarity to improve patient pathways

Datavers is a new project team at the Inria centre at the University of Lille, joint with the University of Lille and, for the first time, the CHU Hospital in Lille. It was launched on 1^st August 2025, and is the result of a project that has been many years in the making: “In 2020, we started thinking about the theme we wanted to focus on, as the Modal team to which I belonged was coming to an end,” recalls Cristian Preda, head of Datavers. “As I'm interested in the statistical processing of data that evolves over time and I also used to work at the Lille Faculty of Medicine, I suggested that we focus on healthcare applications and the patient pathway in particular.” 

This idea brought people together, but Inria researchers soon felt the need to create an interdisciplinary team. “We didn't want it to be the preserve of mathematicians,” continues Cristian Preda. “We felt that the project made much more sense if it included hospital professionals.”  So in 2021, to test possible synergies prior to the creation of the team, Modal (run jointly by Inria and the Paul Painlevé laboratory at the CNRS) set up an exploratory project called “Path” (PArcours paTient en milieu Hospitalier – Patient Pathways in Hospital Settings), in which the METRICS team (University of Lille) also participated, bringing together clinicians and methodologists in ergonomics, biostatistics, medical informatics, data sciences and health economics. 

A Path leading towards Datavers

The project was notably based on the DAMAGE cohort, which includes almost 3,000 elderly people and is being studied by the geriatrics research team at Lille University Hospital, led by Professors Puisieux and Beuscart. The aim was to analyse how these patients fared in the year following their stay in the geriatric unit.

“But in the end, we managed to identify nine profiles that made sense from a clinical standpoint. We are currently preparing a publication which demonstrates the value of this data”, notes Cristian Preda.

Above all, this first collaboration has shown that interdisciplinarity opens up opportunities for new applications of statistical learning… and therefore highlights the relevance of a team based on this model. Of the dozen researchers involved in the Path exploratory project, six will join the future Datavers team: two from the Painlevé laboratory, one from Inria and three from the Metrics team. These include Emmanuel Chazard, a university professor and hospital practitioner (PU-PH) with a doctorate in Public Health: “We have large volumes of data derived directly from patient care,” he points out. They are much more complex than the data normally used by mathematicians to develop statistical tools, especially because it changes over time and is not purely quantitative. So we need methods that can cope with this complexity, and that's the whole point of working with mathematicians on real-life data.” 

From fundamental research to decision-making tools

The team's researchers will therefore be combining their expertise to focus on three areas. The first is to further fundamental research on the use of complex data by adapting statistical learning methods, both supervised and unsupervised. “Hospitals provide a highly heterogeneous source of data, in the form of signals, figures, reports, and so on,” explains Cristian Preda. “We will therefore be using both clustering methods, to identify groups of data with common characteristics, and regression models to explain clinical outcomes such as mortality, quality of life, risk of rehospitalisation, etc.”

The second focus consists in taking account of the temporal dimension of this heterogeneous data, which makes it even more difficult to process. These two areas clearly cannot be studied in isolation from each other and will therefore be closely intertwined. Finally, the team's third objective is to apply all this fundamental research to the patient pathway with one ambition: to propose models for decision support systems. 

Anticipating rehospitalisation and adverse effects... and even rethinking healthcare

For example, the researchers will be working with Professor François Pattou from Lille University Hospital on the follow-up of patients after bariatric surgery in order to use statistical learning to identify the mechanisms governing rehospitalisation, mortality and other events. 

As part of the NuMetaB CDP (cross-disciplinary project), led by Professors François Pattou and Guillemette Marot, they will also be trying to assess the impact of “omics” data (genomics, proteomics, etc.) on mortality and hospitalisation in the context of obesity and diabetes. “Generally speaking, hospitals are very interested in being able to predict how long it might take for a patient to be readmitted to hospital, in order to optimise the management of admissions and discharges,” explains Cristian Preda. 

Emmanuel Chazard also has other applications in mind: “Today, there are software programs capable of issuing alerts on adverse reactions when prescribing medication, but they are so inaccurate and their alerts so numerous that nobody pays attention to them anymore.”

In the longer term, thanks to the new statistical methods that the team is hoping to develop, it might even be possible to imagine a paradigm shift in healthcare: ongoing analyses of patient databases could predict the likelihood of a given pathology and enable carers to contact patients before problems arise.

Data confidentiality, a major issue

This scientific challenge is accompanied by another obstacle, this time of a legal nature: the confidentiality of patient data. Although this information can be used in the framework of specific calls for projects, clinicians cannot pass on their cohort information to researchers for their studies. “Obtaining the agreement of each patient is unthinkable, so we’ll be opting for a different solution: the development of our statistical learning tools will be based on synthetic data,” reveals Cristian Preda. 

Algorithms will therefore be used to create data sets resembling real data. However, it will still be necessary to make sure that they are realistic enough to enable the methods tested on them to be subsequently applied to real-life cases, which is another line of research for Datavers.

To make headway on these various issues, Cristian Preda is counting on calls for projects: “Our interdisciplinary approach will enable us to respond to and obtain the funding we need to recruit PhD students, who will be the driving force behind our team.” Datavers is now ready and able to move forward with improving patient pathway predictions.