Fig. 6.

Illustration of the potential issue of collider bias in a study of disease progression/prognosis. The black box around fracture indicates that this variable has been conditioned on by restricting the prognosis (fracture‐related mortality) analysis to individuals who have experienced a fracture. In this hypothetical example, by conditioning on having a fracture, an inverse association between experiencing a fall and level of physical activity may be induced, with those experiencing a fall measuring a higher level of physical activity than fracture cases who have not experienced a fall. If physical inactivity also causes fracture‐related mortality, experiencing a fall may appear protective against mortality. This will also be the case for any instrument associated with experiencing falls. The magnitude and direction of bias will depend on the interaction between the causes of fracture on probability of experiencing a fracture.^{( 100 )}