Comparative Diagnostic Performance of IOTA Simple Rules, the ADNEX Model, and ACR O-RADS in a Selected Surgical...

Background/Objectives: The accurate preoperative risk stratification of adnexal masses is clinically important. This study compared the diagnostic performance of the International Ovarian Tumor Analysis (IOTA) Simple Rules (SR), the Assessment of Different NEoplasias in the adneXa (ADNEX) model, and the American College of Radiology Ovarian-Adnexal Reporting and Data System (ACR O-RADS). Methods: This retrospective comparative study analyzed 429 women from a selected surgical dataset with pathology restrictions at a tertiary hospital in Nanjing, China, between January 2022 and October 2024. Two experienced gynecologic ultrasonographers independently applied SR, ADNEX (test-positive at a malignancy risk ≥ 10%), and O-RADS (test-positive at categories 4-5). Surgical histopathology served as the reference standard. Diagnostic performance was evaluated using exact binomial 95% confidence intervals (CIs), paired comparisons, receiver operating characteristic (ROC) analysis, ADNEX calibration, observed malignancy proportions across O-RADS categories, and interobserver agreement. Results: Among the 429 selected index adnexal masses, 114 (26.6%) were malignant. SR produced inconclusive classifications for 81 masses (18.9%). When SR-malignant and SR-inconclusive classifications were both regarded as positive indications for referral, SR achieved a sensitivity of 83.3% (95% CI, 75.2-89.7%) and a specificity of 84.1% (95% CI, 79.6-88.0%). ADNEX achieved a sensitivity of 86.0% (95% CI, 78.2-91.8%) and a specificity of 81.9% (95% CI, 77.2-86.0%), whereas O-RADS achieved a sensitivity of 89.5% (95% CI, 82.3-94.4%) and a specificity of 83.5% (95% CI, 78.9-87.4%). The AUCs for ADNEX and O-RADS were 0.92 and 0.90, respectively, with no statistically significant difference identified by paired DeLong analysis (p = 0.16). After Holm adjustment, none of the pairwise comparisons of sensitivity or specificity reached statistical significance. ADNEX exhibited apparent overall underprediction (calibration intercept, 0.865; calibration slope, 1.022; Brier score, 0.096). For O-RADS categories 2 and 4, the point estimates of observed malignancy proportions fell outside the corresponding predefined risk ranges. Interobserver agreement was almost perfect for O-RADS (κ = 0.89) and substantial for ADNEX (κ = 0.79) and SR (κ = 0.68). Conclusions: Within this selected surgical dataset subject to pathology-based restrictions, SR, ADNEX, and O-RADS demonstrated distinct diagnostic performance profiles. Following Holm adjustment, no statistically significant differences in sensitivity or specificity were detected among the three systems. Analyses of masses receiving an inconclusive SR classification were exploratory and therefore do not establish the validity of a sequential diagnostic pathway. Prospective multicenter investigations involving unselected populations are warranted before these findings can be generalized to broader clinical populations.