AI for Personalised respiratory health and pollution (AI-Respire)

We propose to set up the basis for an AI-based digital tool for adaptation/mitigation to the impacts of climate change and pollution on respiratory health in an urban setting. This will enable users to explore interactions between exposure to pollutants, changing weather patterns and their effect on respiratory health, accounting for the complex interactions between environment and health. The project has two coupled aspects:
1. AI model to create a digital twin to establish this interaction using asthmatic and healthy subjects as group test case. This will incorporate big data from health cohorts as well as other studies linking exposure to respiratory outcomes and cell response to pollution, as well as air quality and weather data.
2. Building on this exposure-response model, develop AI-based personalised models using deep learning techniques to include individual circumstances (e.g., age, sex, lifestyle, medical history), combined with air pollution exposure to give a prediction of individual respiratory health.

Up to 90% of the world's population breathe air with high levels of both indoor and outdoor pollution which takes ~7 million lives each year worldwide. In the UK, it is rated as one of the most serious threats to public health with only cancer, obesity and heart disease eclipsing it. The health risks associated with fine and ultrafine particulate matter (PM2.5 and PM0.1) include development and exacerbating respiratory diseases such as chronic obstructive lung diseases including asthma, respiratory infections and lung cancer. While measures are being taken to curb pollution levels, it is essential for individuals to reduce their personal exposure and abate the ill-health effects of pollution. One way of doing this would be to predict who are those individuals who would be at most risk of developing health ill-effects in the long-term. There is virtually no information of this kind of risk assessment at an individualised level and the most available information at the moment is that those at risk are children, the elderly and those already suffering from chronic lung and cardiovascular disease.

The integrated AI modelling will also represent various intervention scenarios (e.g. avoiding certain more polluted travel routes for at-risk people such as asthmatics) to assess reduced exposure and corresponding changes in health outcomes. These biologic parameters of exposure will be integrated with the respiratory responses to pollution in individuals using a combination of cardio-respiratory, physical activity and personal fine particles exposure data from satellite to personal monitors e.g. smart watches. We will also integrate cellular, biochemical and biomarker personal data with the other parameters. We will numerically model the pollution and air flows at the neighbourhood scale and apply an approach centred on the impact of pollution on health to all aspects of modelling, sensor placement and management of the environment as well as the individuals. Thus, any mitigation strategies can be designed to minimize the impact of pollution on health. We develop two unique AI capabilities (1) a new AI method for solving differential equations that we call AI4HFM that can determine the dispersion of pollution through the air and (2) a unique generative method to predict health impacts from pollution levels as well as a level of uncertainty associated with this. This will be combined with reinforcement learning to tailor the AI model for an individual based on information obtained from that individual. Thus the approach may be used to guide healthy activity, prevention, diagnosis and management of respiratory diseases. It will also empower individuals so they can make informed decisions that will influence their health.