Advancing Asthma Care: Predictive Modelling and Community Perspectives in New Zealand

Abstract

Asthma is a chronic respiratory disease that affects people of all ages worldwide. This medical condition places a significant burden on individuals, healthcare systems and communities. Over 300 million people are affected globally by this respiratory condition. Among regions with a high prevalence of asthma, the Western Pacific region ranks among the highest. New Zealand (NZ), which lies within this region, has demonstrated a high prevalence of asthma, particularly among different ethnic groups, including the Indigenous Māori population and Pacific Peoples (also known as Pasifika). Worsening of asthma symptoms may lead to asthma attacks, resulting in emergency department visits and hospital admissions. In some cases, it may also be fatal. Globally, asthma causes approximately 450,000 deaths each year, while 101 people have died in the year 2019 from asthma in NZ. Therefore, asthma management is critical. Supporting earlier identification of individuals at risk of severe asthma outcomes is therefore critical for improving asthma management and reducing preventable harm.

Recent advances in Machine Learning (ML) offer opportunities to improve asthma management through predictive analytics. While these technological developments have the potential to enhance the quality of life and reduce healthcare costs, existing research remains limited in several key areas, including the development of robust and generalisable prediction models for imbalanced asthma data, the prediction of hospital length of stay following asthma admissions, and the consideration of how such technologies are perceived by the most vulnerable communities in NZ. Addressing these gaps is essential to ensure that predictive tools are not only technically sound but also meaningful and acceptable in real-world healthcare settings. Accordingly, this thesis adopts a mixed-methods approach to advance data-driven, equitable, and community-informed asthma care.

The quantitative components of the thesis followed the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology to develop the ML-based prediction models. The findings demonstrated that ML models can effectively predict both future asthma attacks and hospital length of stay using routinely collected hospital and pharmaceutical data in NZ, even in the presence of substantial data imbalance. Across these analyses, prior asthma attacks and medication use consistently emerged as key predictors of impending asthma attacks, showing that reliable predictions can be achieved using a smaller number of clinically meaningful features. These findings contribute to improved understanding of predictive performance, generalisability, and efficiency in asthma attack prediction modelling. Furthermore, it provided new insights into asthma hospitalisation patterns in NZ. The results showed that demographic factors and admission time affect the length of stay for asthma.

Complementing the predictive modelling, the qualitative components were conducted using the inductive thematic analysis methodology through semi-structured interviews. The thesis explored the perspectives of different ethnic groups in NZ, including the Indigenous Māori population and Pasifika, on the use of Artificial Intelligence and digital technology for asthma management, specifically for predicting asthma attack risk. The findings highlighted openness to the potential benefits, alongside key concerns related to trust, data privacy, accessibility, and cultural alignment. The results showed that technical accuracy alone is insufficient for successful adoption, and that cultural values and expectations of these technology-based solutions play a central role in enhancing the acceptability. The thesis also provides a framework for Pasifika asthma care, serving as a guide for enhancing asthma awareness and designing future technology-based solutions for the community. The thesis further identifies and reflects on future research directions informed by the results and their broader implications.