Eighteen studies (1,201 patients) were included.
Sixteen studies were prospective, two were retrospective database reviews, and three enrolled consecutive patients. Only 1 study fulfilled all quality criteria. Ten studies used histopathology alone as the reference standard; other studies used a combined reference standard of histopathology or adequate clinical follow-up, and 1 study used surgery as the reference standard. All patients underwent the reference standard in 10 studies, at least 80% received it in 7 studies, and in 1 study only 51% of eligible patients received the reference standard.
Sensitivity ranged from 36 to 100% and the pooled sensitivity was 83% (95% confidence interval, CI: 78, 87%). Specificity was 100% in all but one of the studies; this was the only study to verify positive EUS-FNA findings by histopathology and it found two false-positive findings. For patients with abnormal mediastinal lymph nodes on CT scans, the pooled sensitivity was 90% (95% CI: 84, 94) based on 8 studies (560 patients). For patients without abnormal mediastinal lymph nodes on CT scans, the pooled sensitivity was 58% (95% CI: 39, 75) based on 4 studies (175 patients). Only one value for heterogeneity was given (no evidence of heterogeneity, p=0.77); it was unclear which measure of accuracy this referred to.
Studies that used a compound reference standard reported significantly higher pooled sensitivity (93%) than those that used histopathology exclusively (pooled sensitivity 79%)(p=0.01).
Minor adverse events relating to EUS-FNA were reported in 10 cases; there were no major adverse events.
The funnel plot showed no evidence of publication bias.