The impact of exposure categorisation for grouped analyses of cohort data