NErler / JointAI

Joint Analysis and Imputation of generalized linear models and linear mixed models with missing values
https://nerler.github.io/JointAI
28 stars 4 forks source link

JM_imp() for binary longitudinal data and right-censored multistate survival data #6

Closed axinaxia closed 2 years ago

axinaxia commented 2 years ago

Dear Dr. Nicole Erler,

I am trying to understand if I can use JM_imp() for my study, which investigates the relationship between a time-varying binary variable (exposure) and right-censored survival outcomes with missing values in the binary variable.

I didn't find an example for binary/categorical time-varying covariates and survival data in this paper https://www.jstatsoft.org/article/view/v100i20. Is it possible to do? In addition, is it possible to use JM_imp() for multistate data as JMbayes2 does?

Thanks a lot!

Xin

NErler commented 2 years ago

Dear Xin, You can use JM_imp() with a binary time-varying variable and right censored time-to-event outcome. The specification is the same as for a continuous longitudinal variable. The type of mixed model that is used for the longitudinal variable is determined based on how the variable is coded, i.e., for a factor with 2 levels a logistic mixed model is chosen. You can change that by using the models argument of JM_imp(). Note that the default association structure for categorical longitudinal variables is to use the observed/imputed value and not the underlying value. This can be changed by setting the assoc_type argument (i.e., assoc_type = c(<variable name> = "underl.value")).

It is not (yet) possible to use multi-state models.

Best, Nicole