Nineteen studies (3,228 patients) were included in the review: seven prospective (1,276 patients); three retrospective (497 patients); and nine with unspecified study design. Twelve studies reported blinding.
Carcinoembryonic antigen (15 studies): Carcinoembryonic antigen had a pooled sensitivity of 45.9% (95% CI: 43.2 to 48.5) and a pooled specificity of 97.0% (95% CI: 96.0% to 97.8%). The sensitivity of carcinoembryonic antigen in individual studies ranged from 28.8% to 82.4% and the specificity for individual studies ranged from 76.7% to 100%. The Spearman correlation of the logit of sensitivity and the logit of 1-sensitivity was 0.268, which suggested a threshold effect. The area under the curve was 0.7691.
CYFRA-21-1 (12 studies): CYFRA 21-1 had a pooled sensitivity of 47.3% (95% CI: 44.0% to 50.6%) and a pooled specificity of 91.8% (95%: 89.5% to 93.7%). The sensitivity of individual studies ranged from 21.8% to 90.9% and the specificity of individual studies ranged from 7.7% to 100%. The Spearman correlation of the logit of sensitivity and the logit of 1-sensitivity was 0.644, which suggesting a threshold effect. The area under the curve was 0.8213.
There was no significant difference between the area under the curve of carcinoembryonic antigen and the area under the curve of CYFRA 21-1 (p>0.05). There was evidence of significant statistical heterogeneity for both specificity and sensitivity of carcinoembryonic antigen and CYFRA 21-1 (Χ2 ranged from 71.53 to 152.10, p=0.000). In metaregression analyses, the different assays used accounted for the greatest source of heterogeneity (diagnostic odds ratio 2.60 for carcinoembryonic antigen and diagnostic odds ratio 2.32 for CYFRA 21-1).