1. Personalisation of heart failure care using clinical trial data
- Author
-
Adamson, Carly
- Subjects
R Medicine (General) - Abstract
Heart failure is a common, debilitating and life limiting disease, resulting in a large burden for both the individual patient and healthcare provision. Therefore, optimisation of treatments for these patients is of prime importance. Heart failure with reduced ejection fraction has a large evidence base for effective treatments, and more recently effective treatments have started to be identified for those with preserved ejection fraction. The effectiveness of these treatments is calculated at a population level, and there is a great deal of interest to try and identify if different patients may benefit more from certain treatments. In addition, we wish to understand more about different phenotypes in heart failure, to help understand what the patient might expect for the trajectory of their illness and potentially develop targeted treatments. To explore these issues further, this thesis presents several approaches using heart failure clinical trial data to try and further understand the patient journey and explore how treatment may be delivered in a more personalised fashion. The first analyses look at the patterns of heart failure hospitalisations, including the timing of admissions, and the relationship with different modes of death. This was examined in both heart failure with preserved and reduced ejection fraction. The accepted trajectory of recurrent admissions falling closer together over time was confirmed, and admissions closer together were linked to a higher risk of cardiovascular death, particularly due to progressive pump failure. Sudden death did appear to be truly sudden and not strongly linked to hospitalisations. The next approach was to perform latent class analysis to try and identify clusters of patients, or phenotypes, within heart failure with preserved and reduced ejection fraction separately using a data driven method. Phenotypes were identified with consistency across different data and using different approaches. These phenotypes were clinically recognisable. Identifying phenotypes in this way may be a route to looking for differential responses to treatments. Lastly, supervised machine learning methods were used to predict outcomes in patients with heart failure and reduced ejection fraction. These techniques provide more analytical flexibility, but did not show performance benefit compared with prognostic models based on survival analysis methods. Overall, the predictive abilities were modest. In conclusion, several avenues were explored to help understand the patient journey in heart failure, aiming to give more detail about the expected patient trajectory and exploring methods to examine for differential treatment responses in phenotypes of patients in heart failure.
- Published
- 2023
- Full Text
- View/download PDF