The diagnostic accuracies of the three new combinations and the standard combination in distinguishing between NPC cases and controls
| Combination | New cut-off values† | Sensitivity (%) | Specificity (%) | AUC (95% CI) | p Value§ | ||
| Early stage (95% CI) | Advanced stage (95% CI)‡ | Total (95% CI) | Control (95% CI) | ||||
| BB+EBNA1-IgA | 0.258 | 97.0 (85.0 to 100) | 92.8 (89.2 to 96.4) | 93.5 (90.1 to 96.9) | 95.0 (92.0 to 98.0) | 0.977 (0.963 to 0.991) | <0.001 |
| HA+EBNA1-IgA | 0.379 | 97.0 (85.0 to 100) | 86.2 (81.5 to 91.0) | 88.0 (83.5 to 92.5) | 94.0 (90.7 to 97.3) | 0.961 (0.943 to 0.979) | <0.001 |
| KSB+EBNA1-IgA | 0.191 | 93.9 (80 to 100) | 94.6 (91.5 to 97.7) | 94.5 (91.3 to 97.7) | 87.0 (82.3 to 91.7)* | 0.964 (0.947 to 0.981) | <0.001 |
| Standard combination | 0.998 | 97.0 (85 to 100) | 88.6 (84.2 to 93.0) | 90.0 (85.8 to 94.2) | 95.5 (92.6 to 98.4) | 0.970 (0.956 to 0.985) | |
†New cut-off value for NPC diagnosis was defined as the value with the largest Youden Index chosen from each ROC.
‡Differences in the sensitivity of early-stage and advanced-stage NPC were compared by Pearson’s χ2 tests. p<0.05 was considered as statistically significant.
§p Values were estimated by non-inferiority tests based on the bootstrap approach for AUC between new combinations and the standard combination. p<0.05 was considered as statistically significant, whereas p>0.05 was consider to be inferior to the standard kit.