Modeling heterogeneity via Multi-state models

Abstract

In this thesis a dataset of 271 intensive-care patients with automatic respiratory distress syndrome that are CMV seropositive is analyzed. The data, obtained from two different hospitals, is analyzed using multi-state models. The first thing we calculated was if there was heterogeneity on the hospital level when it comes to dying and the effect of covariates. To do this we used a frailty-cox model. We found a frailty variance of 1.98 ∗ 10^−16, which implies there is no noticeable heterogeneity on hospital level. After this we used a different multi-state model to calculate the population attributable fraction of CMV reactivation on dying, and found a value of about 0.23, which implies a significant effect.