Missing data in longitudinal studies

Specialeforsvar ved Nanna Leonora Lausvig

Titel: Missing data in longitudinal studies

Resume: Missing data due to dropout is common in longitudinal studies. The presence of missing observations complicates the analysis and introduces the risk of bias. In this thesis, I investigate ways of handling monotonically missing data. The complete case analysis and the last observation carried forward are easy to apply, but require strong assumptions on the missing data mechanism. The linear mixed model can be applied to longitudinal data sets where the number of measurements is not the same for all subjects - e.g. because of dropout. Less strong assumptions are required to get valid results. However, these assumptions cannot be verified from the observed data, so it is important to evaluate the sensitivity of the results towards deviations from the assumptions. Multiple imputation can be used to implement specific assumptions about the mechanism causing data to be missing, and it is a useful tool when conducting a sensitivity analysis. All analyses are illustrated via application to a real life data set from a study in schizophrenic patients. A comparison of the methods is made through simulation. 

Vejledere:   Helle Sørensen, Mette Krog Josiassen, H. Lundbeck A/S
Censor:       Søren Andersen, Novo