Figure 1.

A causal directed acyclic graph representing some key causal features of the data structure. Smoking represents the exposure status (quit smoking vs. continue smoking), Death (19) and Dementia (20) represent indicators of death by 19 years of follow-up and dementia by 20 years of follow-up, respectively. C represents a vector of possible shared causes of dementia and death (such as cardiovascular comorbidities), that may or may not be measured. The key relationships are: 1) Smoking may independently affect the risk of both dementia and death over time through different mechanisms; 2) dying over the first 19 years of follow-up (without prior onset of dementia) determines that the indicator of dementia at 20 years of follow-up is zero (the bold arrow representing this key determinism induced by competing events); and 3) dementia and death can have shared causes.