Artificial intelligence (AI) for health promotion and disease reduction: Knowledge synthesis

📜🔗LINK to the source

1. ANALYTICAL SUMMARY

A still embryonic global state of affairs. In the face of chronic diseases (41 million deaths/year according to WHO, 45.1% of Canadians affected in 2021), the report explores the use of AI to promote healthy lifestyles and prevent disease in OECD countries (p. 3, 7). It examines 22 studies conducted between 2019 and 2023, predominantly American (10/22), focusing on dietary behaviour, smoking cessation, physical activity, mental health, and preventive care (p. 6, 17). The authors highlight a major deficit: very few rigorous studies exist, which undermines any decision to support AI initiatives on solid foundations (p. 3-4).

An inventory of effects, challenges, and success conditions. The document provides a clear typology of systems (mobile applications 15/22, chatbots 9/22, websites, software, platforms) and a mapping of impacts, positive but often uncertain (10 studies) or even negative (3 studies) (p. 17, 21). Its true value lies in the systematic ranking of challenges (8 subcategories), opportunities, and suggestions (7 subcategories), and in two Canadian guidelines for AI governance (p. 28-35). It thus provides decision-makers and practitioners with a framework to anticipate the conditions — technical, ethical, human — for a successful deployment.

2. KEY POINTS OF THE DOCUMENT

1. A concentrated and recent corpus, but geographically unbalanced. 22 studies (2019-2023), including 7 for the year 2023 alone, mostly in the United States (10), then Australia (3), Japan (3), United Kingdom (2), Spain (2), Canada (1), Italy (1). The number of participants varies from 21 to 139,164. No French study (p. 17, 36).
2. The mobile application and the chatbot dominate the landscape. Mobile applications (15/22) and conversational agents (9/22) are the main vectors, ahead of websites (6/22), software (5/22), devices and platforms (1/22 each). Six studies combine two types of systems (p. 17, 37).
3. Real effects on behaviours, but fragile over time. All studies report positive impacts, but 10 report uncertain results and 3 report negative. Smoking cessation is the most solidly demonstrated effect (4 statistically significant studies); regarding health status, 6 studies show an effect on weight loss, 2 on mental health, 1 on blood sugar, 1 on blood pressure — some effects do not persist over time (p. 21).
4. The need for human interaction is a common thread. Frustration with the chatbot, declining engagement over time, and the need for human contact recur as recurring limitations; several studies conclude that AI alone is not sufficient to establish lasting change (p. 21, 28-29, 32).
5. A governance still imported from care, not from prevention. The analysis of the environment reveals only two guidelines, both Canadian (Canada Health Infoway 2023; Healthcare Excellence Canada 2021), focused on general healthcare and not on health promotion — hence the invitation to cautiously transpose their recommendations (p. 34-35).

3. ACTION TRACKS FOR LOCAL ACTORS

1. Use the ranking of the 8 challenges (p. 28-29) as a pre-assessment framework before adopting any AI tool: short-term effectiveness, human interaction, specialised needs, methodological issues, interrupted participation, provider concerns, technical limitations, unreliable data. This avoids purchasing an unfulfilled promise.
2. Systematise the role of the human in the system. As AI alone becomes exhausted, plan from the design stage for a human relay (professional, peer supporter) to sustain long-term engagement (p. 21, 32).
3. Apply the suggestion grid "consideration of the end user" (p. 31-32) : integrate the audiences (elderly, young people, those in precarious situations), plan for suitable readability and accessibility features, and involve users in the design rather than delivering them a ready-made product.
4. Check the actual availability and language of the cited solutions. Many tools are English-speaking or limited to one country (e.g. CALO Mama Plus, Japanese edition not distributable elsewhere, p. 29). Do not assume that a documented solution is deployable in a French context.
5. Draw inspiration from the implementation phases of the Canadian guides (annex 3, p. 49-50) — establishment of leadership, design/approval, steering, sustainability — as a framework for project management, while keeping in mind that they aim for care, not health promotion.
6. Unmet need to document locally : the total absence of French data and cost-effectiveness evaluation (p. 39) requires, before any deployment, a local evaluation phase (acceptability, equity of access, digital literacy of the target audience).

4. ADDITIONAL REFERENCES

1. HAS — First keys to the use of generative AI in health (October 2025). Educational guide for professionals in the health, social and medico-social sectors, structured around the guidelines A.V.E.C. (Learn – Verify – Estimate – Communicate), with a summary infographic. Direct operational French supplement to this report. 👉 https://www.has-sante.fr/jcms/p_3703115/fr/premieres-clefs-d-usage-de-l-ia-generative-en-sante
2. HAS — Artificial intelligence in health: using it well and protecting oneself (user guide + FAQ, June 2026). Produced with France Assos Santé and the CNIL; focus on literacy and personal data protection, aimed at the general public. Extends the dimension of “consideration of the end user” of the report. 👉 https://www.has-sante.fr/jcms/p_4092354/fr/intelligence-artificielle-en-sante-la-has-publie-des-reperes-pour-les-usagers
3. ANS / Delegation for digital health — Guide to implementing ethical AI in health (public consultation, May-June 2025). Operational ethical criteria linked to the phases of development of an AI system (design, learning, validation, commissioning, monitoring) and to the European Regulation on AI (2024/1689). Directly addresses the need for ethical governance highlighted by the report. 👉 https://esante.gouv.fr/actualites/intelligence-artificielle-en-sante-une-concertation-publique-pour-un-deploiement-ethique

5. FREQUENTLY ASKED QUESTIONS (FAQ)

1. Is AI effective for health promotion? Partially. Positive effects are reported in all areas, but 10 studies out of 22 give uncertain results and 3 negative results (p. 21). The most established effectiveness concerns smoking cessation (p. 21).
2. What types of AI tools are involved? Mobile applications (majority, 15/22), chatbots (9/22), websites, software, platforms, and devices (p. 17).
3. On which behaviours does AI have the most impact? Dietary behaviour (10 studies), quitting smoking (6), physical activity (4), mental health (3), preventive care (3) (p. 6, 17).
4. What is the main limitation of these solutions? The difficulty in maintaining engagement over time and the need for human interaction; AI alone struggles to establish lasting change (p. 21, 28-29).
5. Are these results transferable to France? Caution: no French study, corpus dominated by the United States, tools often English-speaking or limited to one country (p. 17, 29). A local evaluation is necessary.
6. Are there guidelines for implementing AI responsibly? Yes: two Canadian governance guidelines are analysed (p. 34-35, 49-50), but they focus on care, not on health promotion.
7. Is this report a practical guide to be applied directly? No. It is a synthesis of preliminary knowledge; the authors also recommend an annual update given the rapid evolution of the field (p. 39).

6. REWRITING IN EASY LANGUAGE

What this report says

The report talks about artificial intelligence. It is also referred to as "AI".

AI consists of computer programmes. They can help to stay healthy.

For example: applications on the phone. Or robots that chat ("chatbots").

The researchers read 22 studies. These studies mainly come from the United States.

The studies talk about eating healthily, quitting smoking, moving more.

They also talk about mental health.

The results

AI helps a little. It works mainly for quitting smoking.

But often, we are not sure about the results.

Sometimes, people stop using the application. They get bored of it.

Many people want to talk to a real person. AI alone is not enough.

What to remember

We still lack solid evidence.

We must be cautious before using these tools.

We should always keep the help of a real person alongside AI.

7. CROSS-ANALYSIS — VALUES OF HEALTH PRACTICES

• Literacy : the document identifies the need for suitable readability and accessibility features (p. 31-32) but does not provide any simplification tools itself.
• Empowerment : participatory design is valued in some studies (involvement of users and therapists, co-creation with diabetic individuals), without being generalised (p. 23, 32).
• Participation : co-construction mechanisms are described occasionally (participatory design, stakeholder co-creation), but remain the exception rather than the rule (p. 32).
• Community health : the collective dimension is weak; interventions mainly target the individual (personal apps), even if the framework displays "populations or communities" (p. 11, 17).
• Ethics : data biases (data set partiality) and ethical governance are addressed through Canadian guidelines (p. 35, 49-50), but little at the level of the interventions themselves.
• Human rights : equity is established as a principle (Canadian guidelines "equity-focused", p. 49) and consideration of ethnicities and socio-economic levels is recommended (p. 32), without in-depth analysis.
• Intersectorality : collaboration between AI experts, health professionals, researchers, and patients is recommended for successful implementation (p. 32).
• Partnership : collaboration models mainly exist in the form of co-creation in design and institutional partnerships in the analysed guidelines (p. 32, 34).
• Combating discrimination : the document mentions consideration of ethnic origin, socio-economic status, visual impairments, and age (p. 32), and non-discrimination is included in governance (p. 50); however, the subject remains underdeveloped in substance.

8. EVALUATION OF THE RELIABILITY OF THE RESOURCE

Scientific relevance — strong on method, cautious on scope. The methodology is rigorous and transparent: rapid review in accordance with Cochrane and PRISMA-P guidelines, PCC framework from the Joanna Briggs Institute, double independent evaluation via Covidence, PRISMA flowcharts (p. 11-14). The institution (Obvia, U. Laval, ASPQ, FRQ funding) is credible and the authors declare no conflicts of interest (p. 2). In contrast, the evidential value is limited by the subject itself: the authors acknowledge the lack of rigorous studies, the strong geographical bias (United States), the heterogeneity of designs and sample sizes, and effects that are often short-term (p. 3-4, 21). The data stops at 2023 and the report itself recommends an annual update (p. 39).

Operational relevance — indirect for the French field. The document is a tool for monitoring and framing, not a toolbox. Its context is Quebec/OECD, with no French data; the solutions described are often English-speaking or not disseminable outside their country of origin (p. 29). Its operational strength lies in its grids (challenges, opportunities, suggestions, implementation phases), transferable as a framework for reflection.

⚠️ Internal inconsistencies noted (reported, not corrected) :

• Figure 1 (PRISMA, p. 13) : the total number of excluded publications is stated as n = 35, whereas the details (incorrect outcome measure n = 9, incorrect intervention n = 4, poor study design n = 17, incorrect population n = 4) total 34. Unexplained discrepancy of 1.
• Time window : the summary (p. 6) and the method refer to studies "published between 2019 and 2024", while the results specify that the included studies actually cover 2019-2023 (p. 17). The literature search was conducted in March 2024.
• No URL could be reported as dead: the two links from the environmental analysis (Infoway, Healthcare Excellence Canada) are referenced without any detectable anomaly in the provided text.

10. STRATEGIC HASHTAGS

#healthcarepractices #AIinHealth #HealthPromotion #HealthPrevention #DigitalHealth #HealthLiteracy #EthicsAI #EvidenceBased