Pre-trained language models in medical code prediction – An empirical analysis