Identification of novel predictors of asthma exacerbations through multimodal data analysis

<p dir="ltr">Asthma is a complex, heterogeneous disorder of the airways. There is wide variability in the manifestations of asthma that are experienced by patients, giving rise to different disease sub-types. This makes it difficult to identify personalised biomarkers that are indicative or predictive of asthma exacerbation risk. The wide range of data modalities recorded in asthma studies underlines the importance of using the appropriate mathematical methods to infer meaningful results. This thesis aims to identify novel population- and patient-level risk factors of asthma exacerbations that relate to (1) small airways dysfunction and (2) multivariate temporal scaling behaviour of diary variables.</p><p dir="ltr">Data from an international, multi-centre cohort study and seven previously conducted randomised controlled trials were analysed to explore the two aims, respectively. Several statistical methods were employed to derive clinically meaningful insights from the data, depending on its modality and nature. Methods include factor analysis, generalised linear mixed-effect modelling, mediation analysis, spectral clustering, and a multivariate extension of detrended fluctuation analysis.</p><p dir="ltr">Population-level risk factors for asthma exacerbations were identified, over a 12-month follow-up. These relate to lung density-based imaging biomarkers and physiological indices of small airways dysfunction. Specifically, lung ventilation gradients along the inferior-superior and anterior-posterior axes of the lungs are associated with prospective small airways dysfunction and are additive to risk of asthma exacerbations when contextualised alongside other risk factors.</p><p dir="ltr">Building on previous work identified and summarised in a systematic review, I developed personalised predictive biomarkers for asthma exacerbations, using multivariate detrended fluctuation analysis applied to time series of diary variables. This biomarker, although not a strong generalisable predictor of asthma exacerbations in the entire population, can predict asthma exacerbations prior to their occurrence in a subset of patients. Additionally, the biomarker was able to predict exacerbations up to 9 days before its onset giving patients more time to seek medical help to prevent the occurrence of the exacerbation.</p>