A comparison of different joint models for longitudinal and competing risks data: with application to an epilepsy drug randomised control trial