How well are human-related hazards captured by multi-agent dynamic risk modelling?