Table 6.

The performance of i2b2 PHI subcategories (micro-averaged on 2016 test set, at strict entity level). Only the categories that appear in the test set are shown in the table. Number of training instances for each sub category is also shown.

Main category	Subcategory	Precision	Recall	F1-measure	# training instances (%)
CONTACT	PHONE	98	96	97	143 (0.69)
	FAX	50	60	55	4 (0.02)
	EMAIL	100	60	75	2 (0.01)
	URL	25	33	29	5 (0.02)
NAME	DOCTOR	95	96	96	2,396 (11.49)
	PATIENT	93	85	89	1,270 (6.09)
DATE	DATE	97	95	96	5,723 (27.46)
AGE	AGE	96	94	95	3,637 (17.45)
PROFESSION	PROFESSION	86	64	74	1,471 (7.06)
ID	HEALTHPLAN	0	0	0	0 (0.00)
	LICENSE	95	95	95	38 (0.18)
	MEDICALRECORD	0	0	0	4 (9.09)
	IDNUM	0	0	0	2 (0.01)
LOCATION	HOSPITAL	87	82	84	2,196 (10.53)
	STREET	92	68	78	46 (0.22)
	ORGANIZATION	82	61	70	1,113 (5.34)
	CITY	88	89	88	1,394 (6.69)
	STATE	94	95	94	662 (3.18)
	COUNTRY	95	88	92	666 (3.20)
	ZIP	100	88	94	23 (0.11)
	OTHER	67	11	18	25 (0.12)