| Summary: | Missing data are a problem for any clinical trial outcome but are particularly an issue for quality of life (QoL) outcomes. To investigate the problem of missing data and methods to deal with it, this thesis uses a novel approach, illustrated using seven completed trials. Data from postal reminders were used to investigate the missing data mechanism and test the accuracy of imputation procedures (as the true value was in fact known). The previously analysis for five of the seven example trials was an analysis of covariance adjusting for baseline QoL and other patient characteristics. Alternative analysis strategies taking account of other interim responses are considered and contrasted with the published analyses. The economic impact of the different data collection methods is explored using two economic decision rules. Different analysis strategies were shown to have an impact on the result of the trial. There is no single best way of dealing with missing data, but some recommendations for researchers are provided. The role of reminders is shown to be extremely important as the reminder system is a cost-effective use of resources to maintain the sample size, decreasing the amount of missing data and reducing the threat of bias. Data collected by reminders can be used to inform the selection of potential imputation methods, again reducing bias. The aim of any trial is to obtain an unbiased as possible estimate of treatment difference to help inform and improve clinical practice to the benefit of patients; the use of reminders may be pivotal in this.
|
|---|
