Discussion on ‘Joint modeling of survival and longitudinal non‐survival data’ by Gould et al.

First, we would like to congratulate the authors on a very accurate and clearly written work. The paper provides a thorough literature review on various aspects of the joint model and certainly provides a strong motivation to the use of joint modeling techniques in medical applications. Jointly modelling two or more processes together has been an active area of research for quite sometime now. This is particularly useful when the processes have different distributions, as in mixed data frameworks. A particularly interesting feature is the possibility to predict one event conditional on the others. We believe three extremely interesting experimental designs arise which are amenable to joint modeling techniques. First of all, joint models must be used when the focus is on the longitudinal outcome(s), but there is informative drop-out. Secondly, joint models must be used when the focus is on the survival outcome, and there are time-varying covariates which might be measured with error [1]. Finally, in many applications both aspects of the joint model might be of interest, where one might discover that a treatment is beneficial for survival but detrimental for a longitudinal outcome measuring quality of life or occurrence of a side effect. While much research happened on this topic, it is yet to become popular in the applied field despite the fact that, as surveyed by the authors, there are now many software options for easy implementation of joint models. Reasons in our opinion range from lack of user-friendly software to lack of understanding of the the advantages of the joint model framework.