Understanding Collider Bias in Biostatistical Analysis

Music City Center 

Room: CC-110A 

In biostatistical analysis, adjustment for potential confounding is a common practice to determine the effect of an exposure or treatment on health outcomes. However, little attention has been given to the potential distortion of the association between exposure and outcome caused by collider bias. A collider refers to a variable that is caused by both the exposure and the outcome or risk factors of the outcome. It is often overlooked but can introduce bias into data analysis, leading to erroneous conclusions. Various statistical approaches commonly employed to control potential confounding, such as restriction, stratification, or adjustment for the collider in regression models, can inadvertently introduce collider bias. This short course aims to equip researchers, biostatisticians, and data analysts in the fields of public health or clinical research with valuable knowledge to increase their awareness of collider bias. Throughout the course, participants will learn how to identify and interpret causal diagrams and directed acyclic graphs (DAGs), which serve as powerful tools for assessing and understanding collider bias. Real-world examples and case studies will be utilized to illustrate the potential impact of collider bias on study results, emphasizing the importance of accurately addressing this bias in data analysis and study design.