Seventy-two studies (n=5,042) were included in the review and some studies reported on more than one index test: 24 studies provided data on MRI; 32 provided data on CT; eight provided data on PET; and 31 provided data on SLNB. Reporting of index test, reference standard and clinical demographic details was generally poor. Blinding of interpreters was unclear for most data sets. Approximately half of the data sets had potential for verification bias. Three quality criteria (adequate description of index test, reporting of withdrawals, and less than four weeks between index test and reference standard) remained significant in the multivariate model described above and were used to grade study quality (each index test appeared to have been treated as a separate study for quality assessment): 32 studies were graded high quality, 52 medium and 11 low.
SLNB (31 studies, n=1,140): Pooled estimates of accuracy measures were sensitivity 91.4% (95% CI 87.1% to 94.6%) and specificity 100% (95% CI 99.6% to 100%). Positive likelihood ratio was 40.8 (95% CI 24.6 to 67.6) and negative likelihood ratio was 0.18 (95% CI 0.14 to 0.24). The pooled failure rate of SLNB was 8.4% (95% CI 3.3% to 15.5%) when blue dye was used alone and 4.4% (95% CI 2.0 to 7.7%) for combined blue dye and technetium 99m colloidal albumin.
PET (eight studies, n=445): Pooled estimates of accuracy measures were sensitivity 74.7% (95% CI 63.3% to 84.0%) and specificity 97.6% (95% CI 95.4% to 98.9%). Positive likelihood ratio was 15.3 (95% CI 7.9 to 29.6) and negative likelihood ratio was 0.27 (95% CI 0.11 to 0.66).
MRI (24 studies, n=1,206): Pooled estimates of accuracy measures were sensitivity 55.5% (95% CI 49.2% to 61.7%) and specificity 93.2% (95% CI 91.4% to 94.0%). Positive likelihood ratio was 6.4 (95% CI 4.9 to 8.3) and negative likelihood ratio was 0.50 (95% CI 0.39 to 0.64).
CT (32 studies, n=2,640): Pooled estimates of accuracy measures were sensitivity 57.5% (95% CI 53.5% to 61.4%) and specificity 92.3% (95% CI 91.1% to 93.5%). Positive likelihood ratio was 4.3 (95% CI 3.0 to 6.2) and negative likelihood ratio was 0.58 (95% CI 0.48 to 0.70).
Using a 27% pre-test probability of lymph node metastases in all cases (regardless of stage), a positive SLNB result increased post-test probability to 94% (95% CI 90% to 96%) and a positive PET result increased post-test probability to 85% (95% CI 75% to 92%). Full results for positive and negative post-test probabilities by index test and disease stage were reported in the article.
Results of the bivariate analysis were reported as a summary receiver operating characteristic (SROC) plot only.
When univariate regression analyses were used to investigate potential sources of heterogeneity, index test type, study quality grade, reference standard histological method, disease stage and lymph node type were found to have a significant effect on diagnostic odds ratio. Only index test type (SLNB versus MRI and PET versus MRI) remained significant in the multivariate model.