University of Edinburgh

N/A

DL_2023_032

Lung conditions like asthma and COPD can be incredibly unpredictable, and it can be very hard to foretell when an attack is likely to occur. Inconsistent use of an inhaler (or other treatment), smoking, obesity, history of respiratory infections, and more, are associated with higher risk of attacks. Despite knowing so many risk factors, identifying who is actually going to have an attack has proven a challenging task. Our risk prediction uses machine learning algorithms, applied to routine health data, like a GPs record of a consultation, to estimate asthma attacks. Machine learning models are able to account for interactions – specific combinations of risk factors which are treated as more than the sum of their parts. These methods often estimate outcomes very accurately, but they require a lot of data and a lot of computing power. This makes them highly suitable for analysis of routine data, which is collected every day on a large scale. In this project, we aim to test a previously developed machine learning model in a new population, and ensure that it works well for all patients, including young people and people with multiple health conditions.

Asthma and chronic obstructive pulmonary disease (COPD) are common chronic lung diseases which are characterised by restricted airflow due to inflammation of the airways. Exacerbations of these conditions result in extreme distress, the need for emergency intervention, and can be fatal. Primary care consultations provide the opportunity for patients and clinicians to assess changes to symptoms, lung function, and corresponding risk of exacerbation. Accurate prediction of risk can instigate timely primary care intervention, prompt more frequency primary care visits, promote risk-reducing lifestyle choices, and encourage patients to seek emergency care following symptom deterioration. Furthermore, highlighting periods when risk is lower can avoid unnecessary additional treatments if an attack is unlikely. This could reduce the amount of steroids that someone with a high recurring risk of exacerbations might need over a lifetime, which could reduce the likelihood of steroid related side effects. This project has four primary objectives. Firstly, it will allow the validation of an existing asthma attack prediction model (based on a random forest algorithm) in a new population, providing evidence for its generalisability outside of the original training data. Secondly, it will allow for extension into COPD patients – looking at differences and similarities in the two conditions and their early exacerbation determinants. Third, the model can be expanded into younger people, and the validity in this population can be assessed. Finally, it will allow the impact of COVID-19 to be evaluated, including infections, long-COVID, and changes to health care practice. With further validation and sub-group discrimination testing, we move closer to being able to design an intervention prototype to test in the primary care setting for usability and patient benefit.

Public Health Research

15/03/2024

De-Personalised

TRE