Influence of missing explanatory variables and longitudinal assessments in breast cancer clinical trials

Clinical trials in breast cancer assess treatment regimens based on a balance of efficacy and adverse effects. To achieve high-quality evidence for these assessments, it is important to minimise potential sources of bias. Therefore, potential bias in the parameter estimates resulting from missing observations is an important concern.In this thesis, the influence of missing data on explanatory variables in time-dependent Cox model analysis is explored, with application to breast cancer clinical trials. In particular, imputation in the context of time-dependent covariates that may be informative missing data which is described has not been studied in detail in the statistical literature. Standard imputation methods from the statistical literature are described, which involve assumptions about the missing data mechanism. Missing observations of quality of life (QoL) are imputed by standard methods before analysis of disease-free survival (DFS) and the performance of the imputation methods is considered. Then the influence of missing observations of an outcome variable assessing safety is considered. Repeated measures analysis of a safety assessment is performed. The insights into the influence of missing data could be generalised.Two clinical trials are considered; the International Breast Cancer Study Group (IBCSG) Trials VI and VII and the Herceptin Adjuvant (HERA) trial. Both investigated adjuvant treatment in breast cancer. There was no evidence in Trials VI and VII that the patient's QoL is related to the patient's DFS, though such a relationship could be masked by the missing observations. Simulation was performed in the context of a positive relationship between QoL and DFS. The simulation study suggested that the performance of the standard imputation methods was influenced by the missing data mechanism. There was no benefit from imputing LVEF values in the HERA trial. It was appropriate to perform the repeated measures analysis of LVEF values using observed LVEF values only.

Name dc:type.qualificationname

phd

Level dc:type.qualificationlevel

doctoral-pg

Grantor dc:publisher.institution

University of Strathclyde

Year dc:date.issued

2016

Author dc:creator

• Procter, Marion J.

DOI

10.48730/j5rq-qt56

Identifier

T14317

Author Identifier

200387168

OAI identifier oai:identifier

oai:strathclyde:hx11xf24j