Joint modelling of longitudinal and survival (time-to-event) data has received significant attention in recent years, however much of the literature in this area has been methodological in nature. The use of joint models in applied research has been somewhat less evident. In this study we used a joint modelling approach to investigate the association between individual-level exposures to a natural disaster such as winter storm, flood, etc. and subsequent changes to physical disability, whilst accounting for non-random dropout due to death. Data for the study was based on a linked dataset containing 27,790 individuals who were interviewed at least once between 1st January 2000 and 1st December 2010 as part of the longitudinal ‘Health and Retirement Study’ in the United States. Disability was assessed using activities of daily living, measured on a discrete 12-point scale. Individual-level exposure to a natural disaster was identified at the county-level based on disaster funding received from the Federal Emergency Management Agency. Our joint model consisted of two submodels: (i) a negative-binomial mixed effects model with a log-link function for modelling the repeatedly measured disability scores and (ii) a proportional hazards model for time to death. The association between the two submodels can then be parameterised in various ways. We investigate the association between disaster exposure and disability using a time-varying exposure covariate which can be included in either one or both of the submodels. We fit the joint models using a Bayesian approach, since this provides the greatest flexibility.