Lin Lin, Chen Chang-zhong, Yu Xiao-dan
Department of Epidemiology, School of Binzhou Medical College, Yantai 264003, China.
Dana-Farber Cancer Institute, Harvard Medical School, Boston, USA;
Zhonghua Liu Xing Bing Xue Za Zhi. 2013 Nov;34(11):1139-41.
In many studies about biomedical research factors influence on the outcome variable, it has no influence or has a positive effect within a certain range. Exceeding a certain threshold value, the size of the effect and/or orientation will change, which called threshold effect. Whether there are threshold effects in the analysis of factors (x) on the outcome variable (y), it can be observed through a smooth curve fitting to see whether there is a piecewise linear relationship. And then using segmented regression model, LRT test and Bootstrap resampling method to analyze the threshold effect. Empower Stats software developed by American X & Y Solutions Inc has a threshold effect analysis module. You can input the threshold value at a given threshold segmentation simulated data. You may not input the threshold, but determined the optimal threshold analog data by the software automatically, and calculated the threshold confidence intervals.
在许多关于生物医学研究因素对结果变量影响的研究中，它在一定范围内没有影响或有积极作用。超过某个阈值，效应的大小和/或方向就会改变，这称为阈值效应。在分析因素（x）对结果变量（y）的影响时是否存在阈值效应，可以通过平滑曲线拟合来观察是否存在分段线性关系。然后使用分段回归模型、似然比检验（LRT）和自助重采样方法来分析阈值效应。美国X&Y解决方案公司开发的Empower Stats软件有一个阈值效应分析模块。你可以在给定阈值分割的模拟数据中输入阈值。也可以不输入阈值，而是由软件自动确定最佳阈值模拟数据，并计算阈值置信区间。
