Abstract:
|
In time-to-event data analysis, statistical inference is usually made based on the cumulative information up to a certain time point by using the hazard function, survival probability, or quantile survival. Balmert and Jeong (2016) proposed a new concept of life lost to compare time-to-event distributions among K-samples. In this talk, we extend our previous work to a regression setting, where the quantile of the distribution of years lost on a log-scale is linear in a vector of covariates. An estimating equation is proposed to estimate the quantile life lost, adjusting for confounding factors. Consistency and asymptotic normality of the regression parameter estimators are established. Simulation results show that the operation characteristics behave reasonably well with finite samples. The proposed method is illustrated with a real dataset from a phase III clinical trial on breast cancer.
|