Thirteen studies were included in the review. The median number of patients in each study was 44 (range: 10 to 183).
Studies met between 12 to 23 of the 34 quality criteria.
High-quality studies (5 studies).
The median prevalence of mediastinal metastases was 34% (range: 29 to 60); prevalence was significantly lower than in low-quality studies (p=0.002). The pooled sensitivity was 39% (95% confidence interval, CI: 17, 61) and the pooled specificity was 99% (95% CI: 96, 100). The pooled kappa-1 coefficient showed poor to fair accuracy for false-negative results (30%, 95% CI: 15, 46).
Low-quality studies (8 studies).
The median prevalence of mediastinal metastases was 81% (range: 55 to 100) for the 6 studies that did not use real-time imaging and 83% for the 3 studies that did. These studies did not verify positive results, thus no data on false-positive results were available. The pooled sensitivity was 78% (95% CI: 71, 84). The pooled kappa-1 coefficient showed fair accuracy for false-negative results (40%, 95% CI: 19, 62).
Complications (12 studies).
Two studies reported major complications; these included two major bleeds and one pneumothorax requiring a chest tube. The overall rate of major complications was 0.26% (95% CI: 0.01, 4).
The discriminant function analysis showed a statistically significant difference in the joint sensitivity and specificity between the high- and low-quality studies (p=0.002).
This analysis showed that sensitivity was higher in studies with a higher prevalence of lymph node metastasis (less than 60% versus at least 60%): the difference was 60% (95% CI: 51, 69). Sensitivity was also higher in more recently published studies (before 1995 versus 1995 onwards): the difference was 10% (95% CI: 1, 18). However, it should be noted that only one of the high-quality studies showed a disease prevalence that was 60% or more, so this difference could be a result of study quality rather than an association with prevalence.