“…By contrast, semi‐parametric joint models assume that the visit and outcome processes are conditionally independent given random effects and covariates, that is, that where η i =( η i 1 , η i 2 ) is a vector of random effects. Generally, methods assume that Z i ( t ) is a baseline covariate or else a subset of X i ( t ), for example, the Liang model where W i is a subset of the covariates X i , Z i is a baseline auxiliary covariate and for identifiability, we assume that E ( η i 1 ∣ X i ( t )) = 0 and E ( η i 2 | Z i ) = 1∀ t . This approach is helpful when there are unobserved time‐invariant patient factors that influence both the visit and outcome processes; for example, socioeconomic status might influence both outcomes and the ability to attend follow‐up appointments.…”