Ljevar, Vanja
(2022)
Exploring the impact of socio-cognitive factors on adherence to asthma medication using traditional mixed methods and machine learning.
PhD thesis, University of Nottingham.
Abstract
Asthma self-management and regular use of medication are crucial in preventing asthma attacks, which can be fatal. However, existing research usually does not take into consideration the impact that perceptions about asthma, of both patients and the general public have on adherence to treatment. Both methodological and empirical contributions are made, combining the approaches from data science and psychology. The thesis proposes new forms of triangulation of alternative research methods; introduces a novel process of identifying perceptions in text, as well as the use of Grouped Model Class Reliance, which is capable of assessing the most predictive feature group while accounting for non-linear relationships. Empirical contributions are reflected in identification of the leading perceptions, that include stigmatization and its underlying mechanisms, as well as the perceived sense of community.
Using a mixed method approach, this work combines traditional Psychological quantitative and qualitative methods and predictive algorithms/machine learning techniques. Using interviews, exploratory qualitative research identified patients’ internal perceptions (about asthma) and external perceptions (what they considered were others’ perceptions). Qualitative analysis of Twitter content formed the second part of this study, identifying several themes in perceptions expressed in tweets. Convergence analysis revealed mutual topics from tweets and interviews: self-pity, humour, disclosure, lack of understanding - topics that reflect stigmatiization; attachment to inhaler and perceived sense of community. However, the presence of negative humour and self-pity was much more prominent on Twitter than in interviews, signifying that some perceptions are more freely expressed on social media such as Twitter, than in the laboratory setting. Conversely, interviews provide more context for stigmatization, though examples. Having recognised the value of Twitter as a naturalistic setting for observing perception, this work created a novel procedure for analysing perceptions in ‘big data’ comprising of filtering, activation, evaluation and modality, that can be used in asthma non-related domains. The results indicate that perceptions related to stigma are the most prevalent negative perceptions about asthma held by both asthma patients and non-patients (following the Twitter analysis).
The next stage of research expanded on this initial conclusion by assessing the impact that negative perceptions have on adherence to medication by people with asthma. The third study measured the impact stigma-related factors have on adherence to asthma medication, concluding that denial was the strongest factor. Mediation analysis using coping mechanisms as mediators also highlighted the non-atomic nature of stigma, identifying different underlying mechanisms by which factors relating to stigmatisation of asthma impact patients’ adherence to medication. However, this work also indicated there is potential information sharing and non-linear interactions occurring across factors. This led to the final study that mitigated the effects of non-linear relationships, using a first Grouped Model Class Reliance (Group-MCR) to compare and quantify the importance of several groups of factors in predicting adherence (including perceptions, demographics, lifestyle, coping, emotions, asthma and psychology traits). This final, fourth study was important in linking up the work of this thesis as it established that perceptions are not just important in predicting adherence - they are the strongest set of predictors of adherence when compared to other factors considered in the literature.
This thesis takes advantage of a mixed-method approach, highlighting the value of the exploratory nature of qualitative work that provided the context and enabled identification of relevant perceptions; the strength of traditional statistics in describing effects, which was evident in the mediation analysis that implied different stigma mechanisms; and the predictive power of machine learning when dealing with complex, non-linear relationships and large amounts of data. This work indicates that public health interventions should focus on patients’ perceptions as an important component of treatment. In addition, the non-atomic and intrinsic nature of stigma identified within patients with asthma and the general public, underlines the importance of not only changing the negative perceptions of patients in the development of future interventions, but also engagement about asthma with the wider public, with the ultimate aim of reducing stigma.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
