랜드마크 분석과 시간종속 공변량 콕스모형의 비교 : 중간사건이 있는 자료

Abstract

When comparing the survival time according to time-dependent covariate, there may be a guarantee-time bias problem. To overcome this bias, time-dependent covariate Cox model and landmark method have been used. The relationship between the regression coefficients from both models can be quantified. This study evaluated the performance of landmark method over the time-dependent covariance Cox model, assuming that the time-dependent covariate Cox model is a true model. Multiple landmark method pooling multiple landmark times was also evaluated. Theory, simulation, and the actual data confirm that the effect size favored null hypothesis in the landmark analysis as compared with the time-dependent covariate cox model. This tendency was more apparent as the rate of censoring decreased and intermediate events increased. In the multiple landmark method, the performance was improved in terms of MSE, but the bias and power were not improved greatly. In order to make a non-biased estimate, it will be necessary to identify distribution of the censoring time and intermediate event time to set the landmark time.