From the aDepartment of Epidemiology, Columbia University, Mailman School of Public Health, New York, NY; and bEpidemiology, Worldwide Safety and Regulatory, Pfizer Inc., New York, NY.
Epidemiology. 2015 Mar;26(2):216-22. doi: 10.1097/EDE.0000000000000224.
Epidemiology textbooks typically divide biases into 3 general categories-confounding, selection bias, and information bias. Despite the ubiquity of this categorization, authors often use these terms to mean different things. This hinders communication among epidemiologists and confuses students who are just learning about the field. To understand the sources of this problem, we reviewed current general epidemiology textbooks to examine how the authors defined and categorized biases. We found that much of the confusion arises from different definitions of "validity" and from a mixing of 3 overlapping organizational features in defining and differentiating among confounding, selection bias, and information bias: consequence, the result of the problem; cause, the processes that give rise to the problem; and cure, how these biases can be addressed once they occur. By contrast, a consistent taxonomy would provide (1) a clear and consistent definition of what unites confounding, selection bias, and information bias and (2) a clear articulation and consistent application of the feature that distinguishes these categories. Based on a distillation of these textbook discussions, we provide an example of a taxonomy that we think meets these criteria.
流行病学教材通常将偏倚分为 3 大类——混杂、选择偏倚和信息偏倚。尽管这种分类无处不在，但作者经常用不同的含义来使用这些术语。这阻碍了流行病学家之间的交流，并使刚刚接触该领域的学生感到困惑。为了了解这个问题的根源，我们回顾了当前的一般流行病学教材，以检查作者如何定义和分类偏倚。我们发现，大部分混淆来自于“有效性”的不同定义，以及在定义和区分混杂、选择偏倚和信息偏倚时，将 3 个重叠的组织特征混合在一起：后果，问题的结果；原因，导致问题产生的过程；和治疗，一旦出现这些偏倚，如何解决这些偏倚。相比之下，一致的分类法将提供（1）一个明确和一致的定义，将混杂、选择偏倚和信息偏倚统一起来，以及（2）一个明确的阐述和一致的应用，将这些类别区分开来。基于对这些教材讨论的提炼，我们提供了一个我们认为符合这些标准的分类法示例。
