To personalise care, we need to accurately estimate an individual’s risk of developing a disease so that we can then tailor interventions to prevent ill health.
While most countries conduct screening for several major diseases, risk assessing populations at scale remains challenging, such that we fall short in providing the preventive care from which individuals would benefit. For example, about 1.5 million adults in the UK are not currently receiving preventive treatments to stop heart attacks, and another million have undiagnosed diabetes or kidney disease.
With this work, we aim to use machine learning to help us build tools to predict your risk of leading chronic conditions. Behind most chronic conditions are a common set of risk factors, like age, body mass index (BMI), and blood pressure. A key question of this work is what predictors can be used in multiple models and how we can then put these together to make it easier to identify who would benefit from preventive treatments.
For most people, recent work has shown that some conditions, such as lung cancer, can be predicted with very few risk factors. We know though that some of these predictions will be quite uncertain, and that this can impact who gets what treatment. Building on this, we also plan to analyse the use of different risk prediction models in subgroups of patients, for example men and women separately.
In summary, we aim to build risk prediction models for common chronic diseases such as heart disease, diabetes, and major cancers. We aim to look at common features in these models, how we can bring these models together in a simple-to-implement fashion, and in what subgroups of patients we need more complex risk models. Further, we aim to analyse how state-of-the-art machine learning techniques can help with these risk prediction tasks. Improved prediction may be through better use of existing information (such as age or sex) or by identifying what other measurements should be made.
The public health benefit of this is to facilitate personalised screening and disease early detection, allowing us to prevent disease where we can, and diagnose it early in other cases to prevent it from becoming more severe.
The project duration is estimated to be 36 months.