Development and external validation of a risk prediction model for asthma attacks in primary care: a retrospective cohort study

PI Title: Dr Matthew Martin

Lead member: Nottingham

"Asthma is a common health problem that affects 1/10 people in the UK and costs the NHS over £1 billion per year. The most serious consequence of asthma are asthma attacks which are flare-ups of asthma during which symptoms quickly get worse. Asthma attacks occur every 10 seconds in the UK and deaths from attacks increased by a third in the last decade. Because of the serious and harmful consequences of attacks their prevention is recognised as a key priority by medical experts and patients.

In other areas of medicine tools, on computers, are available to determine the risk of serious illness such as heart attacks based on risk factors already recorded in patients’ electronic health records. Detection of patients at high risk of serious illness allows the targeting of treatment to reduce this risk or prevent the illness. Medical experts in asthma have called for the development of a similar risk-scoring tool for asthma attacks to identify those at the highest risk of future attacks and try to reduce this risk with appropriate treatment. Previous attempts to develop such a tool unfortunately have not proven effective enough to use in everyday practice. This is because the tools were designed using small amounts of data or did not consider important factors such as how often patients are using their asthma medication.

We want to develop a new effective risk-scoring tool for asthma attacks. We will do this by:
1) Identifying known risk factors for attacks from previous research and from discussion with panels of asthma patients and asthma specialist doctors.
2) Using new approaches based on artificial intelligence (‘machine learning’) to assess and rank the usefulness of these identified risk factors for attacks compared with any new factors identified by machine learning. We will use data from a large database of patients’ GP health records, linked to their hospital records (the information is anonymized so individual patients cannot be identified).
3) Presenting the most important risk factors identified by machine learning to our asthma patient and asthma specialist panels to decide on the factors most important to include in the final ‘model’ of the tool, which will then be tested on a separate health record database.

Patients with asthma have been involved in designing this study and will be involved throughout in advising on important risk factors for attacks, approach to analysis, understanding and circulating the results to others.

The results will be used to develop a new computer-based tool for GPs to assess patients’ future risk of asthma attacks, allowing them to target treatment to reduce this risk in high-risk individuals."

Amount Awarded: £49,908