This paper examines and critically assesses the possible contribution to discourse studies of topic modelling, a group of machine learning methods which have been employed with the goal of automatically detecting thematic information in large text collections. Explored data stem from a big collection of online patient comments about the National Health Service (NHS) in England.