Abstract: A
crucial task in a competing risks framework is how to evaluate dependence among
several causes of failure acting simultaneously. No independent censoring
mechanism is clinically acceptable. For instance, in oncology since cancers are
usually monoclonal the rates of growth of tumor cells in local and distant sites
are likely to be related, thus introducing dependence between local and distant
relapses. The case study illustrated in the paper highlights the importance of
determining the contribution of each j
type of failure, in
evaluating the overall risk of failing. In practice, we suggest to use a risk
indicator based on the joint estimation of the relative and of the overall risks
of recurrence and
These
describe the Pr (failure j occurs by least
one failure occurred) and the Pr (failure j
occurs by t due to any cause), respectively. Based on an esophageal cancer
study we show how different diagnostic conclusions may arise when comparing the
results from analyses with and without independence assumed.
