Critique of impure reason: Unveiling the reasoning behaviour of medical large language models | Synapse