Figure 3. Graph depicts the receiver operating characteristic curve for endometrial thickness at transvaginal US in the diagnosis of endometrial abnormality. This curve was generated by using multiple cutoff points of 1-mm endometrial thickness, from 1 to 24 mm. The numbers on the graph represent endometrial thicknesses, in millimeters. The area under the curve is 0.78 ± 0.04 (standard error). The cutoff point of endometrial thickness that maximized the accuracy of diagnosis was determined to be 6 mm. When an endometrial thickness greater than 6 mm was considered abnormal, transvaginal US had a sensitivity of 84.1%, specificity of 58.2%, positive predictive value of 56.9%, and negative predictive value of 84.8%. Thirty-seven cases were true-positive; seven, false-negative; 28, false-positive; and 39 true-negative, on the basis of combined hysteroscopic-histopathologic diagnoses.