Table 6.

End-to-end system micro F1 scores on the withheld CMED test and on two subsets of the test set with gold standard and predicted medication mentions. The Clinical BERT medication mention extractor was used for all the systems to extract medications.

Model	All		Multiple medications		Multiple medications-different labels
	(n = 1755)		(n = 1462)		(n = 280)
	Gold Med. Mentions	End-to-End	Gold Med. Mentions	End-to-End	Gold Med. Mentions	End-to-End
MedSingleTask	0.718	0.693	0.685	0.667	0.614	0.594
MedMultiTask	0.748	0.725	0.728	0.711	0.605	0.586
MedSpan	0.806	0.787	0.815	0.799	0.806	0.785
MedIdentifiers	0.827	0.799	0.834	0.815	0.826	0.804
MedIDTyped	0.827	0.803	0.829	0.810	0.824	0.800