One of the characteristics on cohort studies is that exposures may change over time. The full use of information related to time-updated exposures, time-dependent covariates and their relationships to estimate the association between exposures and outcomes has become the hotspot of research. In this paper, the Kailuan cohort is used as an example to explore the association between fasting blood-glucose and hepatocellular carcinoma, based on different Cox regression models. Cox or time-dependent Cox regression models can be used to estimate the impact of exposure at baseline or on the time-updated exposures. When time-dependent confounders exist, marginal structure model is recommended. We also summarize the basic principles, conditions of applications, effect estimates, and results interpretation for each model, in this paper.
队列研究的特点之一是暴露因素会随时间而改变，如何充分利用暴露因素及其协变量的变化及其相互关系，从而获得更真实的暴露因素与结局关系是目前的研究热点。本研究以开滦队列为例，探讨基于基线暴露状态、随时间变化的暴露信息以及同时控制依时混杂因素时，如何利用Cox比例风险回归及其拓展模型，包括依时Cox回归及边际结构模型，探讨FPG与肝癌的关系，概述并比较了上述拓展模型的基本原理、应用条件、估计结果及结果解释。.
Keywords: Cohort studies; Cox proportional hazard model; Marginal structure model; Time-dependent confounding; Time-updated exposures.