Addressing methodological issues associated with using the Kidney Disease Quality of Life (KDQoL) questionnaire in a trial measuring the clinical and cost-effectiveness of nocturnal haemodialysis

BackgroundThis thesis addresses methodological issues that arise during clinical trials measuring quality of life using the Kidney Disease Quality of Life (KDQoL) questionnaire. This thesis focuses on the use of statistical methods to account for dropouts when analysing the KDQoL and the use of statistical methods to allow for the KDQoL to be used in a cost-utility analysis.MethodsA systematic review assesses the current practice of trials using the KDQoL in their handling of dropouts and the implementation and reporting of the KDQoL as a quality of life measure. A patient level simulation demonstrates how the methods currently used to estimate the KDQoL scores respond to the presence of random and non-random dropouts. Novel methods including a standard joint model and joint model with competing risks of dropouts is then applied to two respective datasets. A novel prediction model for the estimation of EQ-5D-5L utility values from KDQoL responses is developed.ResultsThe systematic review highlights the underreporting of dropouts and an inconsistency in the reporting of the KDQoL. The simulation highlights the inadequacy of the current statistical methods used when non-random dropouts are present. A standard joint model presents issues when applied to KDQoL data due to the unique problem of dropouts due to transplants and death. A competing risk joint model is provided as an alternative. A comparison of statistical methods to predict EQ-5D-5L utility values supports the use of mixture modelling over standard linear models.ConclusionThis thesis demonstrates the inadequacy of current methods for handling dropouts in KDQoL data and proposes a competing risks joint model as a solution. While effective these methods require sufficient sample sizes and event numbers due to computational complexity. This thesis also provides researchers with the ability to estimate EQ-5D-5L utility values from KDQoL responses using a robust methodological framework.