TESTING THE INDEPENDENCE OF COMPETING RISKS - AN APPLICATION TO THE ANALYSIS OF BREAST-CANCER RECURRENCE

As evidenced by the results of data analysis the contralateral breast cancer recurrence is most likely to originate from subclinical tumor foci that preexist at the time of treatment. This raises the natural question as to whether the recurrences in the contralateral breast evolve independently of those in the ipsilateral breast. It is common knowledge in the theory of competing risks that this problem can not be solved within the nonparametric framework. To overcome the obstacle one needs to proceed from a biologically-based parametric model of tumor latency. We explore this possibility with a model proposed earlier (HOANG et al., 1992). When incorporated into a family of joint distributions with given marginal distributions for the tumor latency times, the model allows testing the independence hypothesis by the likelihood ratio test. Our analysis of data on the ipsilateral and contralateral breast recurrence, though being conclusive only under the model assumptions, suggests that the dependence of risks is negligible for the two types of local failure.