ABSTRACT
In many applications in the health and social sciences, the response of interest is duration to a certain event, such as age at first maternity, survival time after diagnosis, or times spent in different jobs or places of residence. A particular form of heterogeneity may arise when permanent survival from an event is possible. The issue is then to identify latent subpopulations in the censored group, namely to distinguish a permanent survival subgroup from a subgroup liable or susceptible to experience the event, but exhibiting extended survival. The statistical model applied to such data needs to account for the intra-cluster or inter-event correlation. The multivariate lognormal is another possibility, which adapts to the situation of conditional multivariate data, when durations on a second event are obtained conditional on the duration in a first event. Competing risks models involve the tracking of multiple durations corresponding to different types of exit or transition.
