Title: The analysis of diabetic retinopathy data: A conditional copula approach
Authors: Candida Geerdens - Hasselt University (Belgium) [presenting]
Elif Acar - University of Manitoba (Canada)
Paul Janssen - Hasselt University (Belgium)
Abstract: In many studies the response of interest is time to a predefined event (e.g. time to blindness). Such event times are often subject to right-censoring. Further, event times can be grouped into clusters and therefore be correlated (e.g. time to blindness in both eyes). Copulas provide a popular tool to account for the within-cluster association in time-to-event data. However, a further complexity arises when the study contains not only the observed event times, but also a covariate. The question is then whether or not the covariate has an effect on (1) the time to event and/or (2) the within-cluster association. We propose, based on a conditional copula model, an estimation and a testing strategy to infer the impact of a continuous cluster-level covariate on the association in clustered right-censored event time data. A local likelihood approach is used to estimate the functional form of the copula parameter and a generalized likelihood ratio test is described to assess its constancy. Data on diabetic retinopathy (blindness) are used as motivation and illustration. We investigate the association between the time to blindness in both eyes correcting for a possible effect of age at onset of diabetes.