Table 3.
Overall diagnostic performance of chest radiography and CT by rib fracture position: anterior, posterior, and lateral
| True positive/false positive | False negative/true negative | Sensitivity (%, 95% CI) | Specificity (%, 95% CI) | Positive predictive value (%, 95% CI) | Negative predictive value (%, 95% CI) | |
|---|---|---|---|---|---|---|
| Anterior (n=111, 82%) | ||||||
| Chest radiography (n=22 800) | 733/871 | 3485/17 711 | 15·8% (8·0 to 29·9) | 97·2% (95·9 to 98·3) | 14·7% (3·6 to 41·6) | 98·2% (91·7 to 99·7) |
| CT (n=21 000) | 1769/1332 | 2116/15 783 | 51·4% (32·8 to 70·0) | 95·2% (92·7 to 97·0) | 24·2% (6·2 to 56·6) | 98·9% (94·7 to 99·8) |
| Difference | .. | .. | 35·5% (24·5 to 41·7; p<0·001) | −2·0% (−3·2 to −1·3; p<0·001) | 9·5% (2·5 to 16·7; p<0·01) | 0·7% (0·1 to 3·0; p<0·05) |
| Posterior (n=15, 11%) | ||||||
| Chest radiography (n=22 800) | 88/683 | 482/21 547 | 27·3% (14·5 to 47·0) | 98·4% (97·6 to 99·0) | 1·6% (0·3 to 6·2) | 99·9% (99·3 to 100) |
| CT (n=21 000) | 164/1081 | 1011/6479 | 60·2% (40·3 to 78·0) | 97·7% (96·5 to 98·5) | 2·8% (0·6 to 10·7) | 99·9% (99·4 to 100) |
| Difference | .. | .. | 33·0% (23·5 to 39·4; p<0·001) | −0·7% (−1·2 to −0·4; p<0·001) | 1·2% (0·1 to 4·7; p<0·05) | 0 (0·0 to 0·1; p≥0·05) |
| Lateral (n=10, 7%) | ||||||
| Chest radiography (n=22 800) | 91/837 | 289/21 583 | 0·8% (0·0 to 2·5) | 98·0% (97·0 to 98·8) | 1·0% (0·2 to 4·3) | 99·9% (99·6 to 100·0) |
| CT (n=21 000) | 164/901 | 361/19 574 | 4·0% (1·1 to 10·4) | 97·7% (96·5 to 98·6) | 1·4% (0·3 to 5·4) | 99·9% (99·7 to 100·0) |
| Difference | .. | .. | 3·2% (0·6 to 8·5; p<0·05) | −0·3% (−0·6 to −0·1; p<0·01) | 0·4% (−0·2 to 1·6; p≥0·05) | 0 (0·0 to 0·1; p≥0·05) |
Data are shown with 95% CI and p values when available. Difference in diagnostic statistics calculated as CT minus chest radiography. Diagnostic data derived according to the definition that an observation is successful if the fracture is detected on the correct rib in the correct location. The positive and negative results are the total frequencies across all reporters, cases, and locations. Estimates are derived from multilevel models and therefore differ from the raw estimates that could be calculated from number of true positive, false positive, false negative, and true negative events.