Joint Models vs the 'usual' time-dependent Cox regression for time-varying predictors I've got a methodological question, and no data set attached.
Suppose I aim to fit a proportional hazards model (Cox) for survival data.
I have multiple observations for each individual (data in long format).
Particular interest lies within one continuous predictor (such as blood lipid levels) and I'm examining the association/effect on risk of myocardial infarction.
I'm a newbie to this, but I use time-dependent Cox regression (without clustering since only one event is analyzed [repeated events are not of interest]). It appears to me that this is the standard method.
Now, the packages JMbayes, JM, joinR, lcmm can fit joint models (http://www.r-bloggers.com/joint-models-for-longitudinal-and-survival-data/) which appear to be a fusion between mixed effect models and Cox regression.
This seems like a nice idéa, to combine to robust methods... A couple of advantages are reported for Joint Models, of which the "precision in each patients trajectory" is repeatedly mentioned. Howeer, I searched pubmed, google scholar and google for publications using this approach and did not find much.
Should I stick to the "usual" time-dependent (counting process) Cox regression?
Advantages? Drawbacks?
 A: The major breakthrough of the joint models relative to the time-dependent Cox model is that they allow one to deal with the error measurements in the time dependent variables (longitudinal variable in this case). In a Cox model with time dependent covariates we assume that the variables are measured without error.
Some references:
Tsiatis, A. A. e M. Davidian (2004). Joint modeling of longitudinal and time-to-event data: An overview. Statistica Sinica 14 (3),809-834. 
Rizopoulos, D. (2012b). Joint Models for Longitudinal and Time-toEvent Data With Applications in R. Chapman and Hall/CRC
Henderson, R., P. Diggle, e A. Dobson (2000, Dec). Joint modelling
of longitudinal measurements and event time data. Biostatistics 1 (4), 465-480.
A: The major problem is that they are time consuming (computational speaking) and the software is not yet very friendly. There are several approaches and you can let all covariates to be time-dependent, but in this case you should consider to take a multivariate longitudinal model before. In addition for each situation you have to think what is the best thing to share between the longitudinal and survival models - sometimes is the random effects (if you use a mixed effects model for the longitudinal data) and sometimes the expected value of the longitudinal variable at the event time. Sometimes it is some other characteristics of the longitudinal trajectory like the variance, the cumulative effect of the longitudinal data and so on. JM should be adapted to each situation and to each type of data. It's not mandatory that the longitudinal model for HIV data be the same for PSA for example. 
