🔍🤖 AI in health: a numerical overview of real uses, on-the-ground barriers, and levers for action for healthcare teams and management. #AIhealth #DigitalGovernance
💡📊 AI and care: this national report combines the voices of 1,097 respondents, maps priority use cases, and provides decision-making tools for establishments. #DigitalTransformation #PublicHealth
🔍➕ The report also allows for comparison of the maturity of different profiles (doctors, paramedics, management, suppliers) and types of structures, to adapt the level of ambition for AI projects. It highlights the needs for change support, dedicated time, and training arrangements, which helps to build a realistic skills development plan. Finally, it documents the success conditions observed in use cases (co-design, evidence of effectiveness, regulatory compliance, IT integration), useful for drafting specifications, roadmaps, and concrete funding applications.
Source: 📒 Results of the national survey on the adoption and use cases of artificial intelligence in health
📜🔗LIEN
1. Analytical summary
Context, survey, and issues
The report presents the results of a national online survey (10 February – 10 May 2025) involving 1,097 complete respondents, primarily from the health sector, including users and providers of AI solutions. It aims to measure the level of AI adoption in healthcare, identify barriers and accelerators, and document concrete use cases, without claiming exhaustive statistical representativeness. The profiles surveyed include doctors, paramedics, support functions, management, IT service companies, startups, and industry players, with coverage across metropolitan France and the overseas departments and territories. The medico-social sector is very poorly represented, which the report acknowledges as a major limitation. The document places AI within the structural transformations of the healthcare system (digital, data, organisation, European regulation such as the AI Act and EHDS), with a focus on trust, security, and value in use.
Operational contributions for stakeholders
The report provides quantitative data on areas of use (administrative, decision-making, clinical, training, prevention), perceived benefits (productivity, accuracy, personalised care) and obstacles (data security, reliability, regulation, skills, costs, IT integration). It clearly distinguishes the perspective of suppliers (certification difficulties, ROI, market access, data, human resources) and that of users (AI literacy level, training needs, trust criteria, perceived governance). Detailed use cases (Theremia, OptimPharma, Askara…) allow for the visualisation of concrete scenarios, observed gains, and success conditions (co-construction, clinical validation, regulatory anchoring). The document also offers an extensive glossary and references to regulatory devices and frameworks (HAS, CNIL, EHDS, AI Act, HDS, FHIR…) useful for anchoring AI projects within a secure and auditable framework.
2. Key points of the document
The national survey describes 1,097 complete responses, with strong interest (6,814 views, 3,297 starts) but moderate engagement, and a predominance of health professionals (98.5%) compared to the medico-social sector (1.5%). (p. 23–25)
The main barriers on the supplier side are difficulties in integrating into existing IT systems (53%), the challenge of demonstrating ROI (51%), lack of evidence of effectiveness (35%), and regulatory uncertainties (34%), which strongly shape product and market strategy. (p. 35–37, 41)
On the user side, AI is primarily used for administrative assistance and document management (51%), followed by diagnostic support (25%), therapeutic decision support (23%), and training (20%), illustrating a priority focus on task offloading and targeted clinical support. (p. 44, 48–49)
Most respondents report a low awareness of public support mechanisms (≈ 91% at a low or no level) and significant barriers to training (lack of time 54%, lack of quality criteria 38%, costs or inadequacy of training 36%), which constitutes a key lock for skills development. (p. 43, 50–51)
The detailed use cases (Theremia in neurology/psychiatry, OptimPharma on medication returns, Askara dental voice assistant) show measured gains (up to 87% of data entry time saved, 97% of reintegration possible) and highlight the importance of co-innovation between hospitals and startups and GDPR-compliant HDS hosting. (p. 79–81)
3. Action points for local stakeholders
Integrate AI into a formalised digital transformation strategy, with explicit involvement from governance (management, CME, CS, DSI) and clear communication to teams, in order to reduce the massive 'I don't know' regarding the role of the bodies (45.8%). (p. 52)
Organise modular training pathways in AI for professionals (foundations of AI literacy, workshops by profession, local use cases), articulating dedicated time, quality criteria for training, and recognition of AI referents identified in the survey. (p. 43, 50–51)
Launch targeted pilots on use cases with a high perceived impact (documentary assistance, reports, flow management, medication returns), relying on the described feedback (Theremia, OptimPharma, Askara) to negotiate support, evidence of effectiveness, and HDS hosting. (p. 39–40, 79–81)
Structuring partnerships with suppliers around co-development programmes, real-world demonstrations, performance guarantees, and compliance (AI Act, DM/DMDIV, GDPR, EHDS), reflecting the expectations expressed by 58% to 60% of supply-side stakeholders. (p. 36–37, 41; )
Mobilising national and regional mechanisms (ANAP, DNS, ARS, France 2030 programmes, regional hubs) to finance access to data, clinical evaluation phases, change management support, and inter-institutional projects, particularly targeting less mature territories and structures. (p. 10–11, 13–18; )
4. Additional References
Articles referenced in Health Practices on the theme of AI -https://pratiquesensante.odoo.com/2-6-intelligence-artificielle-numerique
G_NIUS – "Analysis report on AI adoption in health" (summary sheet of the same report, useful as a short operational version). 2026.https://gnius.esante.gouv.fr/fr/a-la-une/actualites/rapport-danalyse-enquete-adoption-ia-en-sante
Hub France AI – "Putting AI at the service of health" (presentation page of the working group, including other resources and summaries on AI in health and its regulation). 2025–2026.https://form.typeform.com/to/pyDZjMyE?utm_source=xxxxx&typeform-source=gnius.esante.gouv.fr
Regulation (EU) 2025/327 on the European Health Data Space (EHDS) – Legal analysis article detailing objectives, implementation timeline (general from 26/03/2027) and impacts on health data systems. 2025.https://dsih.fr/articles/5784/enquete-sur-les-cas-dusage-de-lia-en-sante-du-hub-france-ia
5. Cross-sectional analysis – Values of Health Practices
Literacy: The document offers a very developed glossary and educational explanations on AI, digital health, and cybersecurity, but it remains largely oriented towards an audience already familiar with these issues. (p. 13–21)
Empowerment: Beneficiaries/patients are primarily mentioned as recipients of optimised care, but are rarely directly associated with the design of solutions or the evaluation of AI projects. (p. 32, 39–40, 79–81)
Participation: The participation described mainly concerns professionals (providers, users, establishments) through surveys and workshops, with few formalised mechanisms for co-construction with users. (p. 3–4, 10–11, 79–81)
Community health: The collective dimension is addressed at the level of territories (role of ARS, GHT, CPTS, regional hubs) and digital ecosystem dynamics, but little from the perspective of community approaches such as health promotion. (p. 10–11, 26–27)
Ethics: The report emphasises responsible AI, safety, transparency, algorithmic biases, and regulation (AI Act, GDPR), but does not engage in a systematic analysis of cultural or social biases in case-by-case uses. (p. 19–21, 32–33, 41)
Human rights: It refers to data protection, care safety, and equity of access through regulatory frameworks and the EHDS, but without an explicit reference to human rights or social health inequalities. (p. 13–18; )
Intersectorality: The text primarily mobilises the health field, with openings towards the medico-social and digital ecosystem (companies, hubs, institutions), but addresses little the links with the environment or the social in a broad sense (One Health partially excluded). (p. 4–5, 10–11, 23–27)
Partnership: Models of collaboration between hospitals, startups, industry, and institutions are described (co-development, pilots, clusters, hubs, working groups), without detailed methodological formalism but with concrete examples. (p. 10–11, 34–37, 79–81)
Fighting against discrimination: The risks of algorithmic bias are identified, but discrimination (gender, age, socio-economic status, origin) is not analysed in detail and the principles of non-judgment and diversity are more implicit than structured. (p. 19–21, 32–33)
6. Assessment of the reliability of the resource
Scientific relevance: the methodology is clearly described as operational and exploratory, non-academic, with explicit recognition of the limits of representativeness (healthcare/social care imbalance, self-selected sample). The concepts and references used (TAM, UTAUT, AI Act, EHDS, HAS frameworks, CNIL, DM/DMDIV) are aligned with the current state of regulations and applied literature.
Operational relevance: very high for decision-makers and practitioners who wish to structure or question their AI strategy; the report provides perception figures, usage typologies, concrete cases, and an actionable glossary, while remaining cautious about statistical generalisation.
7. Multiple Choice Questions – 5 questions
Part 1 — Questions (without answers)
Question 1 (p. 23–25):
What is the main imbalance of the sample highlighted in the survey?
a) An overrepresentation of healthcare/social care professionals
b) A total absence of AI solution providers
c) A near absence of responses from the healthcare/social care sector
d) An exclusive overrepresentation of hospital doctors
Question 2 (p. 35–37, 41):
What major barrier do providers identify for the adoption of their AI solutions by health clients?
a) The low demand for administrative support tools
b) The difficulties of integration into existing systems
c) The lack of internal training offers
d) The total lack of interest in AI in healthcare
Question 3 (p. 44, 48–49):
Which area of AI use is most reported by users in the survey?
a) Medical research and the development of new therapies
b) Telehealth and connected medicine
c) Administrative support and document management
d) Prevention and public health
Question 4 (p. 50–51):
Among the following proposals, which barrier to AI training is cited first by professionals?
a) The lack of digital teaching materials
b) The lack of time to train
c) The obligation to train only in person
d) The prohibition of AI training by management
Question 5 (p. 79–81):
What quantified impact is reported in the OptimPharma use case regarding the returns of unadministered medications?
a) 50% reduction in the costs of purchased medications
b) 87% of data entry time saved in experimentation
c) 10% decrease in the number of prescriptions
d) 20% reduction in avoidable hospitalisations
Part 2 — Commented correction
Question 1:
✅ Correct answer: c) A near absence of responses from the medico-social sector.
📝 Explanation: The report highlights that only about 1.5–2% of respondents come from the medico-social sector, which constitutes a significant limitation and primarily directs the results towards the reality of the health sector. (p. 23–25)
Question 2:
✅ Correct answer: b) The difficulties of integration into existing systems.
📝 Explanation: 53% of providers cite the difficulties of integration into existing information systems as a major barrier, related to technical debt, interoperability, and hospital architecture constraints. (p. 35–37)
Question 3:
✅ Correct answer: c) Administrative assistance and document management.
📝 Explanation: 51% of users report using AI primarily for administrative assistance and document management, far ahead of other areas such as diagnosis, prevention, or telehealth. (p. 44)
Question 4:
✅ Correct answer: b) The lack of time for training.
📝 Explanation: The lack of time is mentioned by 54% of respondents as the primary barrier to training in AI, ahead of cost, content relevance, or scepticism. (p. 51)
Question 5:
✅ Correct answer: b) 87% of data entry time saved in the experiment.
📝 Explanation: In the OptimPharma use case, the clinical experiment shows "up to 87% of data entry time saved" in the circuit for returns of unadministered medications. (p. 80)
8. Frequently Asked Questions (FAQ)
Who led the national survey on the adoption of AI in health?
The survey was led by the working group "Putting AI at the service of health" of the Hub France IA, with the support of the DGE and the participation of the HAS. (p. 3–4, 8, 12)
What types of professionals responded to the questionnaire?
Respondents include doctors, paramedics, support function professionals, management, solution providers (publishers, manufacturers, integrators, IT service companies) and actors from public and private establishments. (p. 10–11, 23–25, 34–35)
What are the main current uses of AI according to users?
The dominant uses are administrative assistance and document management, followed by diagnostic assistance, therapeutic decision support, medical training, and then prevention, public health, and telehealth. (p. 44, 48–49)
What are the main expectations of professionals for adopting an AI solution?
They primarily request practical demonstrations of performance, concrete examples of effectiveness, technical support and guidance, as well as a guarantee of regulatory compliance. (p. 45)
How do providers perceive regulatory obstacles?
They highlight the length of certification processes (CE, FDA), the high cost of compliance, limited access to data, barriers to market access, and the complexity of requirements (AI Act, GDPR, MD/IVD). (p. 37, 41; )
What is the level of awareness of public support mechanisms for AI?
Approximately 91% of respondents report being poorly or not at all informed about public support and training mechanisms for AI, revealing a major visibility deficit. (p. 50)
What do the use cases presented at the end of the report show?
They illustrate how AI can optimise precision medicine (Theremia), reduce waste and data entry tasks (OptimPharma), or lighten the administrative burden on practitioners (Askara), provided there is co-design, compliant hosting, and validation by field teams. (p. 79–81)
9. Rewriting in Easy-to-Read Language
Title
AI in Health: Survey in France
9.1 Easy-to-Read Summary – Context and Issues
This document discusses artificial intelligence in health.
A large online survey took place in 2025.
1,097 people fully responded to the questionnaire.
Most work in hospitals or medical practices.
Few people come from the medico-social sector.
The aim is to understand how AI is used and what the issues are.
9.2 Easy-to-Read Summary – Contributions for the Field
The document shows where AI is already helping professionals.
It explains the benefits and difficulties encountered.
It provides concrete examples of AI tools and projects.
It helps managers choose action priorities.
It highlights the importance of training in AI.
It reminds us of the rules to follow to protect patients.
9.3 Key Points in Easy-to-Read Language
Many professionals use AI for administrative tasks.
AI also helps with diagnosis and medical decisions.
People lack time to train in AI.
Many are unaware of the public assistance available.
Data security is a major concern.