Table 3.

Recall per PHI category for both the simple PHI versus non-PHI task and the same task with at least one token per PHI span needing to be detected

PHI category	Recall for PHI versus non-PHI	Recall for PHI versus non-PHI with at least one token detected per span
All PHI	97.7	99.1
Macro-averaged over PHI categories	93.1	96.4
Dates	98.6	99.5
Provider names	98.4	100
Locations	88.6	97.8
Vendors and softwares	67.5	77.8
IDs	98.8	100
Patient names	100	100
Phone numbers	100	100

Notes: The first score accounts for misclassifications between PHI categories and the second score for mislabelings of PHI prefixes or suffixes. Scores were computed on Steinkamp Penn test set.

PHI: protected health information.