Development and validation of machine learning methods to improve the diagnosis of respiratory diseases
The most commonly used test for diagnosing respiratory diseases is spirometry. This test involves a patient breathing in to a spirometer, which measures how much air a patient can hold in their lungs, how quickly they can force the air out of their lungs, and other aspects of lung health. Spirometry tests are commonly used by GPs in primary care for diagnosis of respiratory diseases such as asthma and chronic obstructive pulmonary disease (COPD).
It is important that spirometry data are of good enough quality that the measurements can be trusted to show the true state of a patient's lung health. Guidelines have been established for evaluating spirometry data quality, but sometimes the doctors evaluating the spirometry data quality in primary care do not have the required training to perform this evaluation.
Spirometry data alone are not usually enough to definitively diagnose respiratory diseases, particularly rare diseases such as idiopathic pulmonary fibrosis (IPF). Additional tests, such as CT scans, full lung function tests, blood tests, etc. are often required to make a final diagnosis. Getting access to these additional tests can take a long time and delay treatment for the underlying disease. Patients may benefit from an improved quality of life if their diagnosis would be made earlier.
Therefore, we aim to develop and validate artificial intelligence software to help doctors evaluate the quality of spirometry data, and make accurate diagnoses of respiratory diseases using only spirometry data.
We will use spirometry and health records data from UK Biobank to train a machine learning model to predict whether a patient has a respiratory disease, based on spirometry test results and basic demographic data. A similar artificial intelligence software has been shown to outperform specialist lung doctors in diagnosing respiratory diseases using full lung function test data (i.e. not just spirometry data) (Topalovic et al, ERJ 2019).
By the end of the project, we expect to have built clinical decision support tools that can help doctors in the diagnosis and management of patients with respiratory diseases. These tools will be particularly useful for doctors working in primary care, which is where the majority of patients suffering from respiratory diseases will be managed.