Ling, Xiaoxiao; (2025) Item-Level Imputation in Trial-Based Economic Evaluations. Doctoral thesis (Ph.D), UCL (University College London).

Text Thesis_Revised_Final.pdf - Submitted Version Access restricted to UCL open access staff until 1 May 2026. Download (18MB)

Abstract

Cost-effectiveness analysis (CEA) alongside randomised controlled trials typically collects individual-level cost and effectiveness data via multi-item questionnaires at different time points and aggregates the data collected over the entire follow-up to create two summary outcomes of interest: total costs and total benefits, e.g. quality-adjusted life-years (QALYs). Therefore, when some items in the questionnaires are not completed even at a single time point, the resulting total costs or QALYs will be missing. When item-missingness occurs in questionnaires, imputation at item-level can make full use of observed information and generate more accurate estimates for parameters of interest than imputation at more aggregated levels. However, the application of item-level imputation in CEA has received relatively little attention in published research and faces challenges due to the complexities of CEA data (e.g. data skewness, longitudinal structure and correlation between outcomes). This thesis focuses on item-level imputation in trial-based CEAs with multi-item questionnaires and longitudinal follow-up using a Bayesian approach. Bayesian method has been chosen due to its flexibility to account for the typical complexities of item-level imputation in CEA simultaneously and its benefits of incorporating external evidence or expert opinions into the decision-making via prior distributions. The thesis is structured as follows: Chapter 2 reviews the current application of item-level imputation in existing trial-based CEAs. Chapter 3 describes a real case study utilised by the thesis, and based on that, Chapter 4 introduces a full Bayesian model to impute item-level missing data in trial-based CEAs using multi-item questionnaires. Chapter 5 investigates the impact of performing imputation at different aggregation levels of missingness (i.e. the total, timepoint and item level) by comparing the CEA results based on the three modelling approaches to handle missing data. Chapter 6 discusses the potential influence resulting from the choice of ``minimally informative'' prior distributions for common models in Bayesian CEAs.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item