Twenty-seven studies (917 participants, range eight to 97) were included in the review. QUADAS scores ranged from 8 to 12 out of 14. All papers were scored highly for reporting quality.
Primary site recurrence/residual disease: Twenty studies (27 data sets) assessed recurrence/residual disease at the primary site. Pooled estimate of sensitivity was 94% (95% CI 87% to 97%) and pooled estimate of specificity was 82% (95% CI 76% to 86%). Pooled estimate of positive predictive value was 75% (95% CI 68 to 82%) and the pooled estimate of negative predictive value was 95% (95% CI 92 to 97%). QUADAS score did not effect the estimate of sensitivity, but higher QUADAS scores were correlated with lower specificity (p=0.04).
Recurrence/residual disease of nodal metastasis: Thirteen studies (14 data sets) assessed recurrence/residual disease in the neck. Pooled estimate of sensitivity was 74% (95% CI 50% to 89%), and pooled estimate of specificity was 88% (95% CI 74 to 95%). Pooled estimate of positive predictive value was 49% (95% CI 29% to 70%) and pooled estimate of negative predictive value was 96% (95% CI 84 to 99%). Exclusion of the outlying study made no significant difference to sensitivity and specificity.
The bivariate model indicated that estimates of sensitivity were significantly higher where FDG-PET was performed more than 10 weeks after treatment (p=0.002); no difference in specificity was identified. There were insufficient data to investigate other possible sources of heterogeneity (radiotherapy versus chemoradiotherapy, type of reference standard).