A1074
Title: A new EM algorithm for Cox models with partly interval censoring
Authors: Jun Ma - Macquarie University (Australia) [presenting]
Abstract: Partly interval censoring is a common feature in medical datasets. Existing estimation methods for semi-parametric Cox models with partly interval-censored event times include EM algorithms and direct constrained optimization techniques. A new EM approach tailored to this context is presented. The method treats left- and interval-censored data as missing event times. In each iteration, the E-step computes the conditional expectation of the complete-data log-likelihood. In the M-step, a profile likelihood is used to efficiently update the regression coefficients, followed by an update of the baseline hazard function, which is approximated using a piecewise constant function. This iterative procedure converges fast. A simulation study is presented, along with a real-data example to illustrate the method.
