Modelling MRI enhancing lesion counts in multiple sclerosis using a negative binomial model: implications for clinical trials