Table 4.
Test corpus | Reviewer #1 (abstractor) | Reviewer #3 (informaticist) | Both Reviewers Combined* | ||||||||||||
PHI type | N PHI instances | N residual PHI | Expected precision† | Predic-tions | Correct | Recall | Preci-sion | Predic-tions | Correct | Recall | Preci-sion | N predic-tions | N correct | Recall | Precis. |
A | B | C | D | E | F | G | H | I | J | K | L | M | N | O | P |
HIPAA | |||||||||||||||
Pat. name | 59 | 32 | 0.54 | 2 | 1 | 0.03 | 0.50 | 5 | 4 | 0.13 | 0.80 | 6 | 4 | 0.13 | 0.67 |
Date | 228 | 34 | 0.15 | 0 | 0 | 0.00 | 0.00 | 8 | 2 | 0.06 | 0.25 | 8 | 2 | 0.06 | 0.25 |
ALL | 287 | 66 | 0.23 | 2 | 1 | 0.02 | 0.50 | 13 | 6 | 0.09 | 0.46 | 14 | 6 | 0.09 | 0.43 |
Unduplicated count of N predictions and N correct across the two reviewers.
Defined as the number of residual PHI instances (col. C) divided by the total number of PHI instances (col. B).