Automated de-identification of free-text medical records

Table 5 Categorization of algorithm false negatives by PHI type on test corpus.

PHI Type	# False negatives in 296,400 words/1,836 nursing notes	# False negatives per 100,000 words	Recall
Full name	4 †	1
Last name	14 †	5
First name	31 †	11
Location (not street address)	7	2
Full date	2	1	Unknown
Partial date	9	3
Year	8	3
Age over 89	3	1
Overall	78	27	0.94 (estimated)

† None of these names were actually patient names, and therefore were non-critical PHI.

ISSN: 1472-6947