Seventeen diagnostic accuracy studies (n=883) were included in the review.
All of the studies used a matched design to minimise differences between patients who received testing, described the scanner model, and defined clinical criteria for disease confirmation and test interpretation. However, none reported that the radiologist was blinded. There was generally a lack of clear recruitment procedures and poor descriptions of the study populations. No publication bias was detected.
Across all studies, the pooled sensitivity for CT was 81% (95% CI: 72, 88) and specificity was 66% (95% CI: 53, 77). Based on the pooling of 9 studies with appropriate data, the pooled sensitivity and specificity for PET were 92% (95% CI: 87, 95) and 68% (95% CI: 51, 81), respectively, in those with a positive CT, and 73% (95% CI: 50, 88) and 86% (95% CI: 75, 93) in those with a negative CT. In a single study, the sensitivity and specificity for PET were 100% and 68% in those with an indeterminate CT. Since was a trend towards poorer test performance for FDG-PET in individuals with a negative CT, it was considered inappropriate to combine all these studies for an overall estimate for FDG-PET.
Based on the subgroup analysis, there were higher than average pooled estimates for studies with a lower quality score (3 studies) and for those who were referred for FDG-PET regardless of the results of conventional imaging (4 studies). There was no evidence of statistical heterogeneity.