As the 'treatment' annotation is not specific to drugs, it would be interesting to know what other kinds of entities are appearing there. Identify a subset of the treatments with high levels of inter-annotator agreement, manually inspect them to make a statement about what they contain. If there are many, perform an automated analysis using a tool (e.g. Annotator, MetaMap) to map them to high-level semantic groups. This is potentially important particularly because there are not already tools for finding these other kinds of treatments in the literature. It provides a way that the human contributors could generate something unique that machines would miss.
As the 'treatment' annotation is not specific to drugs, it would be interesting to know what other kinds of entities are appearing there. Identify a subset of the treatments with high levels of inter-annotator agreement, manually inspect them to make a statement about what they contain. If there are many, perform an automated analysis using a tool (e.g. Annotator, MetaMap) to map them to high-level semantic groups. This is potentially important particularly because there are not already tools for finding these other kinds of treatments in the literature. It provides a way that the human contributors could generate something unique that machines would miss.