The review included 14 diagnostic accuracy studies with data from 21,839 patients.
The prevalence of rates of detected cancers ranged from 1.2 to 7.3% between the studies. No significant association between sensitivity and (1 minus specificity) was seen on the Spearman's rank test (p=0.12).
When the results of all 14 studies were pooled, the sensitivity was 59% (95% confidence interval, CI: 51, 67) and ranged from 38% (specificity 95%) to 79% (specificity 97%). The specificity was 94% (95% CI: 91, 96) and ranged from 69% (sensitivity 67%) to 99% (sensitivity 57 to 65%). The PPV was 28% (95% CI: 20, 36) and ranged from 8% (NPV 98%) to 68% (NPV 99%). The NPV was 99% (95% CI: 98, 99) and ranged from 96% (PPV 21%) to 100% (PPV 22%).
When only the 5 'good quality' studies were included in the analysis, the values were somewhat higher: the sensitivity was 64% (95% CI: 47, 80) and ranged from 38% (specificity 95%) to 79% (specificity 97%). The specificity was 97% (95% CI: 95, 99) and ranged from 90% (sensitivity 74%) to 99% (sensitivity 57 to 65%). The PPV was 47% (95% CI: 29, 64) and ranged from 21% (NPV 98%) to 68% (NPV 99%). The NPV was 99% (95% CI: 98, 99) and ranged from 98% (PPV 21 to 38%) to 99% (PPV 55 to 68%).
Study heterogeneity was highly significant for almost all indicators in both groups.
Linear regression indicated that none of the independent variables showed any significant relation with any of the diagnostic indicators that were studied.