In the general population, obesity (high BMI) is often considered as an important risk factor for many diseases and premature death. However, in patients with a certain disease, the risk of death in people with obesity was lower than that in people with normal weight. This abnormal correlation has caused widespread discussion, and different explanations have been given by physiology and epidemiology. Based on causal inference theory, this study used directed acyclic graphs to introduce the collision bias (also called collision-stratification bias). Through example demonstrations, this study explored whether and why the abnormal correlation between obesity and higher survival rates in patients with hyperglycemia exists. Finally, it was concluded that this abnormal correlation among patients with a certain disease (e.g., hyperglycemia in the current analysis) was partly or even totally caused by collider bias.

在一般人群中，肥胖（高BMI）通常被认为是许多疾病和早逝的重要风险因素。然而，在某些患有特定疾病的患者中，肥胖者的死亡风险反而低于正常体重的个体。这种异常关联引起了广泛的讨论，并且从生理学及流行病学等不同角度提出了不同的解释。基于因果推断理论，本研究利用有向无环图（DAG）介绍了碰撞偏倚（也称为碰撞分层偏倚），并通过实例演示探讨了高血糖患者中肥胖与较高生存率之间的异常关联是否存在以及为何存在。最终结论认为，在患有特定疾病的人群（例如本次分析中的高血糖人群）中，这种异常关联在一定程度上甚至完全是由碰撞偏倚导致的。