Pooled estimates of sensitivity, specificity, likelihood ratios and diagnostic odds ratios, with 95% CIs, were calculated using both Mantel-Haenszel fixed-effect model and DerSimonian Laird random-effects model. Separate analyses were performed for EUS and EUS with FNA. A correction factor of 0.5 was added to zero value cells.
Heterogeneity of sensitivities and specificities was tested using the likelihood ratio test and heterogeneity of likelihood ratios and diagnostic odds ratios was assessed using Cochran’s Q test. Summary receiver operating characteristic (SROC) curves were constructed and the area under the curve (AUC), an indicator of overall test performance, was calculated.
Studies were grouped into time periods (1988 to 1994, 1995 to 1999 and 2000 to 2006) to assess the impact on test performance of changing technology over time.
The effect of publication and selection bias on the summary estimates was tested by Harbord-Egger bias indicator and Begg-Mazumdar indicator. Funnel plots were constructed to evaluate potential publication bias using standard error and diagnostic odds ratio.