Nineteen studies (n=966 patients) were include in the review. Seven studies assessed endoscopic ultrasonography, 15 studies assessed fluorodeoxyglucose positron emission tomography (FDG-PET), and three studies assessed both imaging modalities. All studies met at least 10 of the 14 QUADAS criteria. Most studies (17) did not report uninterpretable/intermediate test results. Partial verification bias could not be excluded for eight studies. Interpretation of reference standard blind to index test results was absent or unclear in 13 studies. Participant selection criteria were unclear for 11 studies.
The diagnostic sensitivity of endoscopic ultrasonography ranged from 20 to 100%; the sensitivity of FDG-PET ranged from 42 to 100%. The specificity of endoscopic ultrasonography ranged from 36 to 100%; the specificity of FDG-PET ranged from 27% to 100%. The overall accuracy, as indicated by area under the summary receiver-operating characteristic curve, was similar for the two modalities (endoscopic ultrasonography 0.86, 95% CI 0.77 to 0.96; FDG-PET 0.80, 95% CI 0.72 to 0.89); Q* index values were also similar.
There were no significant differences in test performance for any of the subgroup analyses. Multi-variate regression modelling indicated that prospective studies gave higher diagnostic accuracy than retrospective studies (relative diagnostic odds ratio 8.51, 95% CI 1.2 to 60.2); no other co-variate was significant.