The use of modern modelling methods for the statistical analysis of microbiological data in clinical trials of antimicrobial stewardship interventions

Lau, Tin Man Mandy ORCID: https://orcid.org/0000-0001-5894-570X 2024. The use of modern modelling methods for the statistical analysis of microbiological data in clinical trials of antimicrobial stewardship interventions. PhD Thesis, Cardiff University. Item availability restricted.

	PDF - Accepted Post-Print Version Restricted to Repository staff only until 17 January 2026 due to copyright restrictions. Download (5MB)
	PDF (Cardiff University Electronic Publication Form) Restricted to Repository staff only Download (105kB)

Abstract

Collecting microbiological data as part of a clinical trial is a complicated and costly process. The current microbiological outcome measures defined in randomised controlled trials (RCTs) of antimicrobialstewardship interventions(ASIs)simplify these data into binary or categorical variables. Inappropriate use of these data may result in a waste of resources and misinterpretation of research findings. These simplified variables reduce classification accuracy, and reduce statistical power to detect a relation between variables. Misuse of these data poses various methodological challenges and difficulties. I investigated the use of advanced statistical techniques to improve the analysis of microbiological data in ASI trials. I illustrated the application of these techniques using data from two RCTs of interventions that sought to promote antimicrobial stewardship. A group of experts in high-dimensional data, RCTs, and ASIs, including clinicians, microbiologists, and statisticians, were invited to discussthe challenges of using and analysing microbiological data in trials of ASIs. Following their suggestions and recommendations, the latent class and structural equation models were chosen to analyse microbiological data and define microbiological outcome measures. The problem of missing data was addressed, which is commonly seen in microbiological data. Additionally, the estimands framework was used to describe the considerationsinvolved in using microbiological data to define microbiological outcome measures and align analysis accordingly. This thesis emphasises the importance of collaborating with professionals to interpret the microbiological results throughout the study accurately. This collaborative effort involves determining appropriate estimands, understanding the microbiological data, interpreting the microbiological outcomes, and providing clear and meaningful labelling to the latent subgroups. Compared to the traditional binary outcome, the microbiological outcome measures contained more insightful information that can be used to guide antibiotic choice and prescribing decisions. Future research focusing on identifying and validating diseasespecific manifest variables would be beneficial to drive this work forward.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Medicine
Date of First Compliant Deposit:	17 January 2025
Last Modified:	17 Jan 2025 14:57
URI:	https://orca.cardiff.ac.uk/id/eprint/175360

Actions (repository staff only)

Edit Item