The Toulouse University Hospital and Collective Thinking, a French company specialising in artificial intelligence, are combining their respective expertise to understand the impact of the types of care on patient health. Their artificial intelligence will analyze the textual content of medical records in order to improve patient care and prognosis.

Today, research and the improvement of care pathways is limited because most medical data is made up of free texts (hospital reports, medical observations, etc.), making it almost impossible to analyse them on a large scale. To overcome this, the University Hospital of Toulouse and Collective Thinking are collaborating using a novel approach based on three artificial intelligences capable of analysing all of these millions of free texts, which are virtually unexploited to date.

“This is a further step towards personalized medicine, taking better account of the specificities of each patient,” emphasizes Vincent Susplugas, Managing Director of Collective Thinking. The first application framework chosen by the Toulouse University Hospital and Collective Thinking is that of patients suffering from head trauma, the primary cause of acquired disability in young people. This project is one of the 10 winners of the Health Data Hub’s call for projects “AI for an improved experience of the healthcare system”.

Beyond the benefits for the prevention of complications of this pathology, this innovative approach can be replicated in the analysis of care pathways for other chronic pathologies, with important public health issues at stake.

The APSoReN project “Improving the Care Pathway of the Brain Injury Patient by Developing a Neuron Network Artificial Intelligence Model Applied to Massive Data Sets”, has just been selected among the 10 winners of the call for projects launched by the Health Data Hub and the Grand Défi of the Innovation Council on the theme: “Artificial Intelligence for an Improved Health System Experience”.

Following a rigorous selection process involving 138 projects between 20 December 2019 and 1 June 2020, the APSoReN project was selected in the category: “Development of population models for prevention or therapy based on innovative data analysis techniques”. It aims to model the care pathways of head injury patients in order to identify the risks of poor care in order to offer them an adapted and personalized follow-up.

For this purpose, an innovative process of automatic structuring by artificial intelligence of the clinical information contained in text form in the medical files of the Toulouse University Hospital will be coupled with data from the National Health Data System (SNDS), in compliance with the regulations in force relating to the protection of personal data, patient information and respect for “data processing and freedom” rights. The APSoReN project will also contribute to enriching the Health Data Hub’s data catalogue currently being set up.

Head injury (CT) is a serious pathology with a high public health stake, with 150,000 new cases/year in France, 20% of which will have after-effects. For the past 2 years, Toulouse University Hospital has been working on improving the care of these patients with the Federation of Cognitive, Psychiatric and Sensory Disabilities (FHU HoPES).

Professor Xavier de Boissezon, PU-PH in the Department of Physical Medicine and Rehabilitation of the University Hospital of Toulouse and coordinator of the APSoReN project, insists :

“CT is going to become the 1st cause of acquired disability in young subjects. The care pathways of these patients are very often complex and erratic, due to medico-social problems but also because some of them present what is called anosognosia: the after-effects of CT on the cognitive level mean that these patients are not always aware of their disorders. Escaping the care pathway, they may then be lost from sight for several years, often with disastrous socio-professional consequences. As the WHO points out, this pathology and its consequences can really be described as a silent epidemic”.

Identifying at-risk patients by identifying the factors with a poor prognosis, particularly at the social level, is therefore essential, with medical, economic and public health issues at stake. By avoiding late hospitalizations through early and adapted care, it is possible to promote an early return to work and better professional reintegration, thus reducing the costs of care for the community and contributing to a better efficiency of the French health system.

With the help of Collective Thinking, a company specializing in Deep Learning and Natural Language Processing (NLP), the Toulouse University Hospital will analyze the entire contents of the medical records of TC patients who have been seen there and then reconstruct their care pathways by linking these data with those of the SNDS, which are already integrated into the Health Data Hub. Once reconstituted, these care pathways will be subject to new treatments by the AI in order to determine the factors of poor prognosis, which will then make it possible to establish recommendations for prevention and personalized care.

Vincent Susplugas, founder and director of Collective Thinking, explains how this approach is innovative:

“This is the first time that the textual content of patient records will be exhaustively exploited for the analysis of care pathways. Usually, this type of study is based on already structured data, such as the patient’s age, treatments, or socio-professional category, for example. In the case of this project, it is the free text of medical observation notes, letters, hospitalisation or operating reports that will be used. These documents contain a wealth of relevant information, but are often difficult to question and exploit because of their lack of structure. This is the interest of the AI technologies we have developed at Collective Thinking. »

The treatments carried out on the basis of patient records from the Toulouse University Hospital will make it possible to generate a new set of clinically relevant data, in structured (and therefore easily queryable and usable for cohort or care pathway modelling, for example) and anonymised form, enabling them to be shared with the Health Data Hub and the scientific community.

The results from the RN algorithms that will be applied to care pathways to identify the determining factors, whether medical or social, could create a precedent and a new method of analysis that can be replicated to other subjects, as confirmed by Dr. Edouard Dufetelle, Medical Director of Collective Thinking:

“We are very excited to work on this project. Beyond the issue of the CT care pathway and its medico-economic stakes, we are convinced that this approach can then be reapplied to other types of care pathways, particularly in the context of chronic pathologies such as diabetes or chronic kidney disease. In terms of public health, the expected benefits are potentially very significant.

We are fortunate in France to have a colossal database in the form of the SNDS, which is the envy of many countries. However, this database is incomplete and our aim is precisely to enrich it with more clinical content. The information is already available within hospitals, and our ambition is to use our expertise in AI to better develop and exploit it. »

The launch of the project is expected in the autumn of 2020 and the first results within a year.

Translated from Le CHU de Toulouse et Collective Thinking lauréats de l’appel à projets du Health Data Hub et du Grand Défi de la BPI