Table 2.
Comparison of Methods to Detect Colorectal Cancer (CRC) Tests Compared to Adjudicated Reference Standard in 200 Patients
| All CRC Tests (n=265) | Colonoscopy (n=190) | FSIG (n=10) | DCBE (n=19) | FOBT (n=46) | |
|---|---|---|---|---|---|
|
| |||||
| NLP system | |||||
| Recall | 93% * | 92% | 100% | 100% | 96% |
| Precision | 94% †,‡ | 94% | 91% | 100% | 92% |
| F-measure | 94% | 93% | 95% | 100% | 94% |
|
| |||||
| Chart review | |||||
| Recall | 74% | 72% | 80% | 53% | 91% |
| Precision | 98% | 98% | 89% | 100% | 98% |
| F-measure | 84% | 83% | 84% | 69% | 94% |
|
| |||||
| Billing records | |||||
| Recall | 44% | 56% | 20% | 42% | 2% |
| Precision | 83% | 99% | 67% | 100% | 4% |
| F-measure | 58% | 71% | 31% | 59% | 3% |
*
p<0.001 when comparing recall of NLP to both chart review and billing records
†
p=0.1 when comparing precision of NLP to chart review
‡
p=0.001 when comparing precision of NLP to billing records
NLP=natural language processing; CRC = colorectal cancer; FSIG = flexible sigmoidoscopy; DCBE = double contrast barium enema; FOBT = fecal occult blood testing