Fifty-one studies met the inclusion criteria (n=3,857 participants; 28 studies had sample sizes over 50). Study quality was considered to be good. Six studies reported avoiding potential for progression bias. Fifteen studies avoided differential verification bias. Nineteen studies reported blinding of interpreters of tests. Mort than 80% of studies met the other QUADAS criteria.
PET: Pooled sensitivity was 88.4% (95% CI 86.3 to 90.3), specificity was 83.1% (95% CI 79.6 to 86.3) and DOR was 32.778 (95%CI 24.107 to 44.570).
PET/CT: Pooled sensitivity was 90.1% (95% CI 85.5 to 93.6), specificity was 80.1% (95% CI 73.1 to 86.0) and DOR was 27.105 (95% CI 15.307 to 47.998).
Endoscopic ultrasonography: Pooled sensitivity was 81.2% (95% CI 78.7 to 83.5), specificity was 93.2% (95% CI 91.7 to 94.5) and DOR was 49.774 (95% CI 25.756 to 96.189).
Sensitivity was significantly greater for PET/CT than for PET and endoscopic ultrasonography (p<0.001). Specificity was significantly greater for endoscopic ultrasonography than PET or PET/CT (p<0.001).
Significant heterogeneity was observed for all three measures for endoscopic ultrasonography and PET and for specificity of PET/CT. Patient spectrum was identified as the most important variable that explained the heterogeneity across studies.
The area under the curve was 0.9324 for PET, 0.9414 for PET/CT and 0.9387 for endoscopic ultrasonography; the *Q index for PET/CT and endoscopic ultrasonography were significantly higher than PET (p<0.05).
Results for a range of subgroup analyses were presented.