Figure 3. Causal diagram of the underlying mechanism of MVJM for the longitudinal and survival processes: Rectangles denote observed data; ellipse denotes unobserved quantities; circles denote un-observed (including subject-specific random bi) terms and unknown parameters.

The arrows direct statistical dependencies. The causal chain of interest runs from longitudinal exposure yik(k = 1, 2, 3) to hazard λi(ti) for event, whereas yik(k = 1, 2, 3) does not “cause”event, but is statistically related to event through its dependence on the unobserved random quantities bi.

Huang et al.Journal of Medical Statistics and Informatics  2021 9:2DOI : 10.7243/2053-7662-9-2