Skip to main content

Table 8 Precision and recall rates from the machine learning approach and CRIS’ pattern matching approach

From: Development and evaluation of a de-identification procedure for a case register sourced from mental health electronic records

Types of PI MIST performance CRIS performance
Total Number of Notes scanned 20 20
Total number of PI instances 191 191
Number of PIs correctly identified and masked (True Positives) 154 169
Number of PIs that should have been masked (False Negatives) 43 22
Number of instances masked that should not have been masked (False Positives) 8 0
Precision 95.1% 100%
Recall 78.1% 88.5%