Statistical analysis at the service of patients
Will doctors soon be using a robust mathematical decision-making tool to ease their workload and facilitate more efficient patient care? The Modal project team from the Inria centre at the University of Lille is about to take up this major scientific challenge.
The stakes are high. On the one hand, population growth, chronic illnesses, multimorbidity and the ageing of the population, compounded by shortages of staff and financial resources, are prompting healthcare facilities to acquire innovative resources. On the other hand, computerisation by healthcare providers (hospitals, practices, medical biology laboratories) and insurance companies (including l'Assurance Maladie, the French national sickness insurance scheme) has led a massive accumulation of patient care data.
The solution is to reuse this data and integrate it into artificial intelligence-based procedures to create care pathways adapted to the needs and resources. In recent years, a number of data science teams have taken up this challenge, but their progress has been slowed by the limitations and inefficiency of the available methods and statistical analysis tools.
Synergies between mathematicians and doctors
So how have French scientists managed to overcome some of the major methodological and scientific hurdles? By focusing on interdisciplinarity.
For two years, researchers in applied mathematics from the Modal project team in Lille worked in close collaboration with doctors and biostatisticians from the multi-disciplinary Metrics research team at Lille University Hospital who contributed their expertise and brought a different perspective to bear on the relevance of the modelling choices and the results obtained. This work formed part of an exploratory project (see box) on "PAtient paTHways in Hospital contexts" (PATH), created in 2021. That’s the secret!
Together, they first sought to define a patient pathway mathematically, in the broadest sense. Realising fairly quickly during the course of discussions that each pathway is unique and dependent on the pathology and context, the scientific team focused on an initial use case for which the university hospital had precise usable data and expertise: the DAMAGE cohort including very frail elderly people who are exposed to the risk of repeated hospitalisation and for whom doctors have to decide whether to hospitalise them again or send them home or to nursing homes for the elderly.
"Collaboration with Metrics' clinical researchers was therefore initially essential, in order to identify specific use cases and gain access to the data", explains Sophie Dabo, head of the exploratory action for the Modal team. The data was processed securely and anonymously after the signature of an agreement between Inria, Lille University Hospital and the University of Lille. There are now plans to allow the Modal team to process data from the INCLUDE health data warehouse (EDS) as soon as authorisation has been granted by Lille University Hospital Scientific and Ethics Committee and the French Data Protection Agency (CNIL).
"But the project didn’t end there. Our collaboration has also been very useful in subsequently sorting through these resources and learning how to extract data of interest from them", explains Sophie Dabo. It was then a matter of getting these elements to tell their story.
Child's play for a mathematician? It all depends on the context.
The data collected on the cohort of elderly people consisted mainly of data that can be easily pooled and therefore processed”, says the researcher. “For the EDS INCLUDE data, it will be quite a different story. The data exists in a wide variety of formats, which we won't be able to analyse without the help of health database specialists.
Validating this initial model
The next step is to validate the model and use statistical learning techniques to train the machine to extract data of interest automatically and determine the presence or absence of clear clinical signs of a given pathology, for example.
In the same way that driver assistance software needs to be capable of distinguishing between dangerous and non-dangerous situations in order to warn of risks and prevent accidents, assistance software needs to capable – in the context of patient pathways – of identifying the data of interest obtained from observations of new patients, and of analysing it using a suitable model which, once trained, can provide decision-support advice: concerning the existence of risks for a given pathway (death, multiple hospitalisations, adverse events, etc.).
"The aim is clearly not to replace practitioners, but to offer them a powerful support tool capable of automating certain data extraction and analysis steps in the construction of patient pathways", explains Sophie Dabo.
And the adventure is set to continue: to extend these activities and meet societal challenges concerning patients and their care pathways, the creation of a future Inria project team composed of health and statistics researchers is currently being discussed.
What is an exploratory action ?
This original scheme implemented by Inria enables scientists to explore a line of research and assess its potential before the formal creation of a project team, for example. Inria issues calls for projects every year. Interested applicants must describe their project, situate it in its context, and then present the scientific challenges, lines of research, expected results and the material and human requirements for its implementation. "These exploratory actions provide an opportunity to establish new collaborations and innovative research projects, such as those we have developed with Lille University Hospital", enthuses Sophie Dabo.
Find out more
- Santé numérique : quand la recherche monte en puissance (in French), INSERM, 30/6/2022
- La place du numérique dans le parcours “ville-hôpital” des patients atteints de cancer (in French), Hospitalia, 15/3/2021.
- AP-HP and Inria launch URGE, a research project aimed at analysing and optimising patient visits to A&E, Inria, 21/2/2023.