Analyse zeitveraenderlicher Kovariablen und rekurrenter Ereignisse am Beispiel einer Studie zur prophylaktischen Behandlung von Oesophagusvarizen

To assess the effect of prophylactic sclerotherapy on variceal hemorrhage and survival of patients with liver cirrhosis and esophageal varices, a randomized study had been carried out. We analysed the data from different points of view. The time-dependent Cox model and the linear counting process of Aalen are applied allowing for the time-dependent covariate ``variceal bleeding`` -- that switches up to three times -- in a multivariate analysis of the remaining life time. A model for the times to and between the recurrent events of bleeding including unobserved heterogeneity is estimated by a distribution-free and by a parametric method where the latter also admits time-dependent covariates such as repeated measurements of laboratory data. We find that high age, high Child-Pugh score and especially the first occurence of variceal bleeding have a statistically significant negative effect on survival whereas patients with fundic varices and/or alcoholic cirrhosis have a significant higher risk of bleeding. In both analyses, inclusion of time-dependent covariates does not change the estimation substantially. In particular, prophylactic sclerotherapy is not shown to reduce the risk of bleeding nor dying significantly.