Table 4:
Error Analysis
| Error class | Error type | Count |
|---|---|---|
| Patient history of cancer | False positive | 3 |
| Family history of cancer | False positive | 1 |
| Cancer not reportable by UAB – ICD-9 codes | False positive | 4 |
| Cancer not reportable by UAB – NLP + ICD-9 codes | False positive | 6 |
| Cancer at this body location not reportable | False positive | 2 |
| Uncertain language means not reportable | False positive | 2 |
| Negation of cancer mention not detected | False positive | 1 |
| Secondary tumor is not reportable | False positive | 1 |
| No NLP CUIs and coarse ICD-9 codes | False negative | 5 |
| Coarse NLP CUIs and no ICD-9 codes | False negative | 6 |
| Coarse NLP CUIs and coarse ICD-9 codes | False negative | 5 |
| NLP or machine learning failures | False negative | 4 |
CUI, Concept Unit Identifier; ICD-9, International Classification of Diseases – 9; NLP, natural language processing; UAB, University of Alabama at Birmingham.