In this thesis the bootstrap method is introduced into broader eld of survival
analysis. In this branch of statistics incomplete data is usually present which is
also called censored data. Four different bootstrap methods for survival analysis
are gathered and accurately described. Their adjustment for competing risks
where different types of events are analysed is also included. Furthermore, the
concept of the number of years lost for simple data and data with competing risks
is introduced.
Results were obtained by three sets of simulations: simulations of simple data,
simulations of data with competing risks and imitation of real data. Real data
requires working with population tables from which survival of the population
is calculated. Results suggest that the simple and the conditional bootstrap
are the best bootstrapping methods for censored data. We show why the use
of bootstrap is useful in practice. Bootstrapping number of life years lost also
takes the variability of the demographic variables into account in addition to the
variability of the sample. Demographic variables such as age, year and gender are
required when working with population tables. Without bootstrapping condence
intervals can be oddly shaped, especially when there are only few events in the
sample. This drawback can be improved by using this resampling method.
|
