A Two-Step Regression Model for Hazard Functions
将Cox比例危险模型扩展为允许协变量系数随时间变化，适用于癌症死亡率研究，并通过惩罚最大似然估计得到平滑危险函数，案例显示扩展模型拟合更好。
SUMMARY The regression model for proportional hazards (Cox, 1972) is extended to allow (approximately) for time-varying covariate coefficients. This is particularly appropriate for cancer mortality studies where the effect of a covariate measured at one point of time decays as time progresses. A smoothed hazard function estimate for this extended model is derived using penalized maximum likelihood. A case-history is given concerned with recurrence of carcinoma of the vulva. The basic model does not fit very well, but the extended model fits well.