0
$\begingroup$

I have a large dataset of within-individual repeated measures, where each participant has varying number of observations (multiple rows of data for each participant). Data include a time-dependent marker variable (dynamic risk), time-to-event data (recidivism), as well as demographic information and a separate risk score based on historical factors (static risk)

I am using JointLCMM to model trajectories of a marker variable (dynamic risk) while accounting for data missing due to survival events (recidivism). I am a bit confused about whether the variable used in the survival argument must be time independent.

As I had not been including covariates in the mixed-effects model

m<-JointLCMM(dynamic_risk~time,random = ~ time, subject = 'ID', mixture = ~ time,ng = 2, idiag = TRUE, data = data, link = "linear")

I hadn't thought to include them in the survival argument

survival=Surv(entry,exit, recidivism_flag)~dynamic_risk

Reading more on JointLCMM, it seems that this may be problematic, as my dynamic_risk variable is time-dependent.

Is it essential that the variables in the JointLCMM survival term be time-independent?

If time-independent covariates are essential for the survival term, should they be included in the mixed model as well?

$\endgroup$
1
$\begingroup$

With the JointLCMM() function you fit a latent class joint model. This models assumes that the association between the longitudinal outcome dynamic_risk and the risk of recidivism_flag is explained via latent classes. In other words, given that you know the class in which a subject belong, his/her dynamic_risk and risk of recidivism_flag are assumed to be independent.

Given this underlying assumption (which is also known as the conditional independence assumption), you would include the dynamic_risk again into your relative risk model as a time-varying covariate. If you have indications that this conditional independence assumption may not hold, you could fit a shared parameter joint model (that makes another type of conditional independence assumption) or include the linear predictor of the mixed model for dynamic_risk as a time-varying covariate in the relative risk model. The former can be done in the R packages JM and JMbayes. The latter is not yet available in a software packages as far as I know.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.