. 2019 Jul 2;26(11):1297–1304. doi: 10.1093/jamia/ocz096

Table 5.

Performance of each label category with pretrained MIMIC models on i2b2 2012 task

	word2vec	GloVe	fastText	ELMo	BERT_BASE	BERT_LARGE
PROBLEM	76.49	77.83	75.35	84.1	85.91	86.1 ^a
TEST	78.12	81.26	76.94	84.76	86.88 ^a	86.56
TREATMENT	76.22	78.52	76.88	83.9	84.27	85.09 ^a
CLINICAL DEPT	78.18	77.92	77.27	83.71 ^a	77.92	78.23
EVIDENTIAL	73.14	74.26	72.94	72.95	74.21	74.96 ^a
OCCURRENCE	64.77	64.19	61.02	66.27 ^a	62.36	65.65

MIMIC: Medical Information Mart for Intensive Care;

^aThe best performing result in the respective task.