In many applications, such as medicine, biology or actuarial science, the primary interest lies in the time until a certain event occurs. One typical feature of survival data is that the study population is only followed up over a finite time frame. Therefore, the event of interest may not be observed for all individuals, and we have to deal with so-called fixed right-censored data. Further, it might be convenient to consider only individuals that have not experienced the event of interest prior to a prespecified time. In this case, we speak of left-truncation. Handling the resulting lack of information about the real data becomes particularly challenging when we aim to model dependence patterns within clusters of several survival times.
Vine-copulas allow highly flexible modeling of high-dimensional dependence structures. Since vine-copula theory has only been developed for complete data so far, we extend existing concepts to make parametric maximum likelihood inference for three and four dimensional fixed right-censored and possibly left-truncated survival data. This allows to model flexible pairwise association patterns within the considered clusters.
We thoroughly establish the modified likelihood function in terms of vine-copula components. Thereafter, the small sample performance of the estimator is evaluated through extensive simulations. Further, two real-data applications in tumor research and veterinary medicine are presented demonstrating the viability of the developed models and their estimation.
«
In many applications, such as medicine, biology or actuarial science, the primary interest lies in the time until a certain event occurs. One typical feature of survival data is that the study population is only followed up over a finite time frame. Therefore, the event of interest may not be observed for all individuals, and we have to deal with so-called fixed right-censored data. Further, it might be convenient to consider only individuals that have not experienced the event of interest prior...
»