VP63.17: A retrospective study on the diagnostic accuracy of transvaginal sonography of adenomyosis in a single tertiary hospital

To be able to evaluate the diagnostic performance of transvaginal sonography in the diagnosis of adenomyosis from March 2017 to September 2019. A two-year retrospective study who had transvaginal ultrasound diagnosis of adenomyosis and underwent hysterectomy and histopathologic confirmation of adenomyosis. Frequency and proportion were used for nominal variables, median and range for ordinal variables, and mean and standard deviation for interval/ratio variables. Sensitivity, specificity, predictive value, and likelihood ratio of the transvaginal sonography in detecting adenomyosis was performed; 95% CI intervals were reported. 242 women were included, mean age was 45.69 ± 5.76 years and 54% had adenomyosis by transvaginal ultrasound. Institutional use of TVS demonstrated low to moderate sensitivity (67.1%), specificity (70.2%) and overall accuracy (68.18%) in diagnosing adenomyosis. The PPV was good at 80.9% but the NPV was relatively low at 53.2%. The positive and negative likelihood ratios predict approximately 15% increase and 15% decrease in pre-test probabilities of disease, respectively. Patients who had adenomyosis on histopathology, 67.1% probability that TVS will turn positive for adenomyosis (sensitivity). Patients without adenomyosis on histopathology, we have 70.2% probability that TVS will be negative for adenomyosis (specificity). Patients who had adenomyosis on histopathology are 2.25 times more likely to yield a positive result on TVS and 0.47 times less likely to yield a negative result on TVS for adenomyosis (LR-). When TVS is positive for adenomyosis, we have 80.9% probability that the patient has adenomyosis on histopathology (PPV). When TVS is negative for adenomyosis, we have 53.2% probability that the patient doesn't have adenomyosis on histopathology (NPV). Overall, the accuracy of TVS in detecting adenomyosis is 68.18%, using histopathology as a reference standard.