Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.728035
Title: Methodological development to support clinical prediction modelling within local populations : applications in transcatheter aortic valve implantation and an analysis of the British Cardiovascular Interventional Society national registry
Author: Martin, Glen
ISNI:       0000 0004 6497 1092
Awarding Body: University of Manchester
Current Institution: University of Manchester
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
There is growing interest in using large-scale observational data collected through national disease registries to develop clinical prediction models (CPMs) that use the experiences of past patients to make predictions about risks of outcome in future patients. CPMs are often developed in isolation across different populations, with repetitive de novo development a common modelling strategy. However, this fails to utilise all available information and does not respond to changes in health processes through time/space. Using the UK transcatheter aortic valve implantation (TAVI) registry as motivation, this thesis aimed to develop methods that improve the development of CPMs within local populations. Three research questions (RQs) were considered: (1) what are the challenges of mortality risk prediction in TAVI due to changes in procedure knowledge and the patient population? (2) Can we use a combination of baseline patient characteristics to predict the risk of mortality post TAVI? (3) How can we exploit multi-dimensional information about patients to inform clinical decision-making at a local-level? Chapter 2 demonstrates potential to simplify the procedure by removing pre-dilation of the aortic valve, thereby altering the underlying treatment pathway, and Chapter 3 shows that mortality rates from registries should be reported in the context of the underlying patient population. Despite Chapter 2 and 3 presenting potential challenges to TAVI risk prediction (RQ 1), CPMs are fundamental to support benchmarking/audit analyses. To this end, Chapter 4 found that the performance of existing TAVI CPMs was inadequate for use in UK patients. Through the discovery of new risk factors (e.g. frailty) in Chapter 5, the thesis derived a UK-TAVI CPM for audit analyses within the UK cohort (Chapter 6). While Chapters 4-6 present the classic framework of CPM development (RQ 2), this cannot overcome the challenges of mortality prediction in the TAVI setting (RQ 1) and is not suited to support local healthcare decision-making (RQ 3). Thus, Chapter 7 found that local model development could be supported through aggregating existing models rather than re-development. Existing methods of model aggregation were extended in Chapter 8 to allow prior research and new data to be utilised within the modelling strategy; application of the herein derived method to the UK TAVI registry indicated that it could facilitate the choice between model aggregation and de novo CPM derivation. Generally, this thesis has the potential to improve the implementation of CPMs within local populations by moving away from the iterative process of re-development. Practically, the thesis derived a UK-TAVI CPM for audit analyses, using classic and novel methodology.
Supervisor: Sperrin, Matthew Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.728035  DOI: Not available
Keywords: Clinical Prediction Modelling ; Logistic Regression ; transcatheter aortic valve implantation
Share: