Twenty one studies were included in the review (n=1,273). The authors stated that all 21 studies were level IIb evidence exploratory cohort studies with good reference standards. Five studies included a comparison group of women who had not received preoperative chemotherapy.
Pooled IR was 90% (95% CI: 88, 91). Significant heterogeneity was noted in the result (p<0.001). IRs were found to be lower when 'mixed' sentinel node biopsy techniques were used (pooled IR 87%, 95% CI: 85, 90) and higher when dye and/ or isotope techniques were used (pooled IR ranged from 92% to 95%). In the five comparative studies, there was no statistically significant difference in pooled IR between those who received preoperative chemotherapy and those who did not. The Bayesian model estimated pooled IR as 91% (95% CrI: 88, 94).
Pooled sensitivity was 88% (95% CI: 85, 90). The authors did not report whether there was any heterogeneity in the result. In the five comparative studies, there was no statistically significant difference in pooled sensitivity between those who received preoperative chemotherapy and those who did not. The Bayesian model also estimated pooled sensitivity as 88% (95% CrI: 84, 91).
Pooled negative predictive value was 90% (95% CI not reported). The authors did not report whether there was any heterogeneity in the result.
Pooled overall accuracy was 94% (95% CI not reported). The authors did not report whether there was any heterogeneity in the result.
Funnel plots for IR and sensitivity indicated no publication bias.