Transvaginal ultrasound assessment of myometrial and cervical stromal invasion in women with endometrial cancer: interobserver reproducibility among ultrasound experts and gynecologists.

OBJECTIVES To assess interobserver reproducibility among ultrasound experts and gynecologists in the prediction by transvaginal ultrasound of deep myometrial and cervical stromal invasion in women with endometrial cancer. METHODS Sonographic videoclips of the uterine corpus and cervix of 53 women with endometrial cancer, examined preoperatively by the same ultrasound expert, were integrated into a digitalized survey. Nine ultrasound experts and nine gynecologists evaluated presence or absence of deep myometrial and cervical stromal invasion. Histopathology from hysterectomy specimens was used as the gold standard. RESULTS Compared with gynecologists, ultrasound experts showed higher sensitivity, specificity and agreement with histopathology in the assessment of cervical stromal invasion (42% (95% CI, 31-53%) vs 57% (95% CI, 45-68%), P < 0.01; 83% (95% CI, 78-86%) vs 87% (95% CI, 83-90%), P = 0.02; and kappa, 0.45 (95% CI, 0.40-0.49) vs 0.58 (95% CI, 0.53-0.62), P < 0.001, respectively) but not of deep myometrial invasion (73% (95% CI, 66-79%) vs 73% (95% CI, 66-79%), P = 1.0; 70% (95% CI, 65-75%) vs 69% (95% CI, 63-74%), P = 0.68; and kappa, 0.48 (95% CI, 0.44-0.53) vs 0.52 (95% CI, 0.48-0.57), P = 0.11, respectively). Though interobserver reproducibility (in the context of test proportions 'good' and 'very good', according to kappa) regarding deep myometrial invasion did not differ between the groups (experts, 34% vs gynecologists, 22%, P = 0.13), ultrasound experts assessed cervical stromal invasion with significantly greater interobserver reproducibility than did gynecologists (53% vs 14%, P < 0.001). CONCLUSION Preoperative ultrasound assessment of deep myometrial and cervical stromal invasion in endometrial cancer is best performed by ultrasound experts, as, compared with gynecologists, they showed a greater degree of agreement with histopathology and greater interobserver reproducibility in the assessment of cervical stromal invasion.