Pan Jin-ren, Chen Kun
Department of Epidemiology and Health Statistics, School of Medicine, Zhejiang University, Hangzhou 310058, China.
Zhonghua Liu Xing Bing Xue Za Zhi. 2010 Jul;31(7):808-11.
Interaction assessment is an important step in epidemiological analysis. When etiological study is carried out, the logarithmic models such as logistic model or Cox proportional hazard model are commonly used to estimate the independent effects of the risk factors. However, estimating interaction between risk factors by the regression coefficient of the product term is on multiplicative scale, and for public-health purposes, it is supposed to be on additive scale or departure from additivity. This paper illustrates with a example of cohort study by fitting Cox proportional hazard model to estimate three measures for additive interaction which presented by Rothman. Adopting the S-Plus application with a built-in Bootstrap function, it is convenient to estimate the confidence interval for additive interaction. Furthermore, this method can avoid the exaggerated estimation by using ORs in a cohort study to gain better precision. When using the complex combination models between additive interaction and multiplicative interaction, it is reasonable to choose the former one when the result is inconsistent.
交互作用评估是流行病学分析中的重要一步。在进行病因学研究时，常用逻辑模型或Cox比例风险模型等对数模型来估计危险因素的独立效应。然而，通过乘积项的回归系数来估计危险因素之间的交互作用是基于乘法尺度的，而从公共卫生目的出发，应该基于加法尺度或偏离加法性。本文通过一个队列研究的例子，拟合Cox比例风险模型来估计Rothman提出的三种加法交互作用的度量。采用具有内置Bootstrap函数的S-Plus应用程序，可以方便地估计加法交互作用的置信区间。此外，该方法可以避免在队列研究中使用比值比（OR）进行估计时的夸大，从而获得更好的精度。当使用加法交互作用和乘法交互作用之间的复杂组合模型时，结果不一致时选择前者是合理的。
