The Proportional Hazards Model with Linearly Time-dependent Covariates and Interval-censored Data

The semi-parametric estimation under the proportional hazards (PH) model with a linearly time-dependent covariates and with interval-censored data has not been investigated before. The partial likelihood approach does not work and one has to use the generalized likelihood function (GLF). There is a challenge in this problem. The GLF must be in the form of the baseline hazard function, rather than the baseline survival function as in the PH model with time-independent covariates, and a feasible way to specify the hazard function is a piece-wise constant function. However, several naive ways do not yield a consistent estimator. We propose proper modifications of the GLF. Simulation results suggest that our method works. The generalization to other types of time-dependent covariates is also discussed.