135 Views
Sunday Poster Session
Category: Colorectal Cancer Prevention
Artin Galoosian, MD, MA
David Geffen School of Medicine at UCLA
Los Angeles, CA, United States
Table 1A: Characteristics of the validation cohort, N=495
| Total | Gender | p-value* | |
Male | Female | |||
All patients (N, %) | 495 (100.0%) | 250 (50.5%) | 245 (49.5%) | N/A |
Age in years (mean, SD) | 58.0 + 7.4 | 57.5 + 7.2 | 58.4 + 7.5 | p=0.162 |
ADR detected by NLP (N, %) | 166 (33.5%) | 100 (40.3%) | 66 (26.9%) | p=0.004 |
ADR detected by manual chart abstraction (N, %) | 181 (36.6%) | 111 (44.6%) | 70 (28.6%) | p< 0.001 |
Table 1B: Performance of the NLP algorithm
ADR detected by NLP | ADR detected by manual chart abstraction (gold standard) | Total | |
Adenoma detected | No adenoma detected | ||
Adenoma detected | 165 | 1 | 166 |
No adenoma detected | 16 | 313 | 329 |
Total | 181 | 314 | 495 |
Test characteristics | Sensitivity (165/181) | 91.2% | |
Specificity (313/314) | 99.7% | ||
PPV (165/166) | 99.4% | ||
F-score | 95.1% | ||
Accuracy [(313+165)/495] | 96.6% | ||
Kappa coefficient | 93% (90% in males, 96% in females) |