Thirty-five studies (5,892 women) were included.
Prevalence of disease was 13% for endometrial cancer and 40% for endometrial polyps or hyperplasia.
The number of women unable to tolerate EVUS was reported in 16 studies; 14 studies reported the number of women who had a non- diagnostic EVUS.
Mean endometrial thickness according to pathology: normal, 4 mm (standard deviation, SD=1); endometrial polyp, 10 mm (SD=3); hyperplasia, 14 mm (SD=4); and cancer, 20 mm (SD=6).
There was no correlation between sensitivity and specificity within any of the individual thickness measurements used to define abnormal in the individual studies.
The sensitivity and specificity were reported for endometrial disease using different endovaginal thickness measurements to define an abnormal result. Using a 5-mm thickness threshold: the sensitivity for endometrial disease (1,306 women) was 92% (95% CI: 90, 93) and the specificity (2,137 women) was 81% (95% CI: 79, 83); the sensitivity for cancer (457 women) was 96% (95% CI: 94, 98) and the specificity (2,986 women) was 61% (95% CI: 59, 63).
The accuracy of detecting endometrial disease, stratified by HRT status, was reported for different endovaginal thickness thresholds. Using a 5-mm threshold, with no HRT, the sensitivity (423 women) was 95% (95% CI: 93, 97), the specificity (593 women) 92% (95% CI: 90, 94), the positive likelihood 11.9, and the negative likelihood 0.05. With HRT, the sensitivity (883 women) was 91% (95% CI: 89, 93), the specificity (1,544 women) 77% (95% CI: 75, 79), the positive likelihood 4.0, and the negative likelihood 0.12.
Homogeneity. For sensitivity, using a 5-mm thickness threshold, the studies were homogeneous for cancer, while 2 out of 20 were heterogeneous for any endometrial disease. For specificity, the results were inconsistent across the studies. Using a 5-mm thickness threshold, 7 out of 20 studies were heterogeneous for cancer and 8 out of 20 were heterogeneous for any endometrial disease. The estimates for specificity were less heterogeneous after stratifying by the use of HRT, with heterogeneity improving among women not using HRT but remaining in women using HRT. Stratification by other factors did not improve consistency across the studies.
The results confirmed the trade-off between sensitivity and specificity obtained by changing the threshold thickness used to define abnormality. At 3 mm, the sensitivity was 98% and the specificity 38%; at 10 mm, the sensitivity was 66% and the specificity 79%. Summary ROC curves were significantly different between women using HRT and those not using HRT (P=0.02).
In the sensitivity analysis, the inclusion of excluded studies had little effect. After the exclusion of one study that contributed 25% of the patients to the pooled results, the sensitivity for endometrial disease at a 5-mm thickness threshold decreased from 92 to 88%; sensitivity for cancer alone decreased from 96 to 95%.