Even though AI in general, and NLP in particular, has made a lot of progress in recent years, the impact on the processing of medical written data has so far been limited. We argue that this is mainly because publicly available data is scarce in the medical domain and thus provide an overview of available data sources as well as strategies to overcome data scarcity. We also discuss de-identification approaches and possible challenges when working with de-identified data. Finally, we give an overview of available German NLP models for the medical domain and discuss domain adaptation as a way to transfer models from a specific application area to another.