2.5. Statistical analysis

All analyses were calculated based on 2 types of comparison method with statistical software MetaDisc version1.4 (Unit of Clinical Biostatistics, Ramo‘n y Cajal Hospital, Madrid, Spain); one is patient-based analysis, and the other is node-based analysis.

A patient-based data analysis uses the pathologically proven positive node in the same patient who had been identified to have metastatic lymph nodes by preoperative imaging, while node-based date analyses use the pathologically proven positive node in the corresponding node which had been described as containing positive node by preoperative imaging.

The pooled sensitivity, specificity, positive likelihood ratio (LR+), negative likelihood ratio (NL-), diagnostic odds ratio (DOR) with the respective 95% confidence intervals (95% CI) were calculated of each imaging technology. We also calculated summary receiver operating characteristics curves (SROC) and the area under the curve (AUC) to assess the interaction between sensitivity and specificity.

Finally, a 2-sample Z-test was performed to evaluate a significant difference in sensitivity, specificity, DOR, AUC or the Q∗ index between these 2 techniques. A P value <.05 was considered to be statistically significant. All of the statistical analyses were performed using Meta-DiSc version 1.4^[16] or SPSS 13.0 (SPSS Inc., Chicago, IL).

The I-square (I²) tests was used to evaluate the statistical heterogeneity. It was considered significant study heterogeneity if the I² value was >50%. When the I² value was >50%, the pooled estimates were carried out by random-effects model, otherwise performed by fixed-effects model for the meta-analysis.