Paper detail

Bayesian Quantile-Based Joint Modelling of Repeated Measurement and Time-to-Event data, with an Application to Lung Function Decline and Time to Infection in Patients with Cystic Fibrosis

Background: The most widely used approach to joint modelling of repeated measurement and time to event data is to combine a linear Gaussian random effects model for the repeated measurements with a log-Gaussian frailty model for the time-to-event outcome, linking the two through some form of correlation structure between the random effects and the log-frailty. In this approach, covariates are assumed to affect the mean response profile of the repeated measurement data. Objectives: Some applications raise substantive questions that cannot be captured by this structure. For example, an important question in cystic fibrosis (CF) research is to understand the impact of a patient's lung function trajectory on their risk of acquiring a variety of infections, and how this varies at different quantiles of the lung function distribution. Methods: Motivated by this question, we develop a joint quantile modelling framework in this paper with an associated Markov Chain Monte Carlo algorithm for Bayesian inference. Results: The translation from the common joint model towards quantile regression succeeds and is applied to CF data from the United Kingdom. The method helps detecting an overall difference in the relation between lung function decline and onset of infection in the different quantiles. Conclusions: Joint modelling without taking into account the special heteroscedastic structure is not sufficient in certain research question and the extensions towards models beyond the mean is necessary.