Last week an acquaintance asked me for advice on creating an app to support patients who have a health problem which impacts hundreds of thousands of people in UK, and tens of millions world-wide. I tried (as always) to give helpful advice, and I also told him that what he wanted to do was very important, and indeed part of the future of healthcare. Every health system in the world wants to move care out of hospitals and focus more on prevention (instead of treatment), and such apps can support this move. Its also a great way of using AI to improve health outcomes!

Of course, huge numbers of people are already using AI in health contexts. According to OpenAI, 230 million people ask ChatGPT about health and wellness every week. Chatterji et al (2025) and Ayre et al (2025) also report large numbers of people using LLM chatbots for health. However, the current focus is primarily on existing “use cases”. For example, the recently announced ChatGPT Health seems (as far as I can tell) to focus on fitness and diet (there are tens of thousands of existing apps that support this), mental health (likewise), and health information (people have used Google for this for decades). Perhaps ChatGPT will be more effective than existing apps in these use cases, this remains to be seen; I can see ways in which it will probably be better than MyFitnessPal and Strava, and also ways in which it will probably be worse.

Anyways, the point is that since ChatGPT Health is largely replicating what existing health apps do, I suspect it may not have a major impact on health outcomes. In order for AI to really make a difference, we need to explore novel use cases (of course this is true of most new technologies, its the new use cases that have the most impact).

Helping patients manage illness

I think one very promising area is using AI apps to help people manage health conditions, which is what my friend wants to do. Of course this is a huge space, and includes (amongst other things):

• Monitoring disease status: Is a disease getting worse, and (if so) should action be taken, such as contacting a doctor?
• Supporting compliance/adherence to treatment regimes: Helping people take medication as prescribed, use medical devices correctly, follow dietary and other restrictions, etc
• Personalised information about status and expectations: Illness of course can be very stressful, and personalised information about status and expectations can reduce stress.

There are many existing apps in this space, such as Glucose Buddy for diabetics, but I think this is an area where AI can make a real difference in health outcomes and improving quality of life.

ASICA: Example of monitoring disease

A concrete example of using AI to monitor disease is our ASICA project, which supports people who have melanoma (a type of skin cancer). Such people are supposed to regularly take pictures of relevant parts of their body, and send them to clinicians who check the pictures. Unfortunately many people do not take pictures, or take poor pictures which are useless.

The ASICA app helps and encourages people to take good pictures. From an AI perspective, its uses computer vision to check image quality, based on what clinicians have told us they need; it deliberately does not do diagnosis, the goal is to help patients take good pictures. It also uses an LLM chatbot to help people provide important contextual information, and more generally provide advice and help on monitoring melanoma. We plan to start an evaluation with patients in March.

Of course ASICA is targeting a specific niche, its not a general tool for monitoring any disease! But the general concept is acquiring data about an individual at home (via sensors, cameras, questionnaires), addressing data quality issues (which are common in patient-acquired data), and then perhaps analysing the data for insights about the disease; and I think this concept could be applied in many more contexts.

Compliance and adherence

People with health problems are often asked to take medication, use medical devices, restrict their diet, etc. Unfortunately, as with melanoma skin checks, a lot of people do not do this, or do it poorly, and this can have a major impact on their health. This is called compliance or adherence in the medical literature.

The same concept of acquiring, cleaning, and analysing data can be used for compliance. For example, several years ago we started working with a medical-device manufacturer to create apps which used AI to help patients use a device correctly. The vision was to combine sensor data from the device, camera data about the patient using the device, and patient questionnaires (supported by LLMs) to detect problems in device usage, and help the patient do better. Unfortunately the project got cancelled because of commercial issues, but again I think the concept is great and could lead to real improvements in health and quality of life.

Information about status and expectations

Long-term illness is scary and stressful. Good information about the patient’s status and expectations for the future can reduce stress, which will increase quality of life. Of course there are loads of information sources about disease, including ChatGPT Health, but this information is more useful and effective if it is personalised to the patient’s circumstances. I have not tried ChatGPT Health (its not available in the UK), but other apps and LLMs I have tried are not great at personalising complex medical information, and indeed make mistakes when doing so (paper).

One project we did in this space many years ago was Babytalk-Family (paper). This system ingested the full electronic patient record of a baby in a neonatal intensive care unit, and summarised it for the baby’s parents. The goal was reducing stress, and parents were very appreciative of the system, in fact the hospital we worked with kept running Babytalk-Family even after the research project ended. We are also looking at explaining AI health prediction models, so that patients have a better understanding of the model (blog).

Better information in this sense may not help people live longer, but if it improves quality of life by reducing stress, this is an important contribution.

Final thoughts

There are major challenges in creating real-world apps that support disease-management. Challenges are varied and include technical (eg, ensuring advice is safe), organisational (eg, connecting apps to relevant clinicians), HCI (eg, getting people to use an app for years or decades), social (eg, convincing people in deprived communities to use the app), etc. So it will not be easy, but I think AI could make such apps much more powerful, and enable them to have a major impact on health outcomes and quality of life for people who are managing long-term health conditions. Which is why I told my friend that I was happy and excited to learn that he wanted to develop such an app!