Comparison of IOTA three-step strategy and logistic regression model LR2 for discriminating between benign and malignant adnexal masses

Abstract

Aims: To compare the diagnostic performance of two ultrasound-based diagnostic systems for the classification of benign or malignant adnexal masses, the three-step strategy and the predictive logistic regression model LR2, both proposed by the International Ovarian Tumour Analysis (IOTA) Group. Material and methods: Prospective observational study at a single centre that included patients diagnosed with a persistent adnexal mass by transvaginal ultrasound over a period of two years. They were evaluated by a non-expert sonographer by applying the three-step diagnostic strategy and the LR2 predictive model to classify the masses as benign or malignant. Patients were treated surgically or followed up for at least one year, taking as the standard reference for benignity or malignancy the histological diagnosis of the lesion or ultrasound changes suggestive of malignancy during the follow-up period. Sensitivity, specificity, positive and negative likelihood ratios and overall accuracy of both systems was calculated and compared. Results: One hundred patients were included, with a mean age of 50.6 years (range 18-87). Surgery was performed on 62 (62%) patients and 38 (38%) were managed expectantly. Eighty-three (83%) lesions were benign and 17 (17%) were malignant. The IOTA three-step strategy presented sensitivity of 94.1% (95%CI, 86.7-98.3%) and specificity 97.6% (95%CI, 94.8-99%). The LR2 logistic regression model showed sensitivity 94.1% (95%CI, 73-98.9%) and specificity 81.9% (95%CI 72.3-88.7%). Comparison of the two systems showed a statistically significant dif-ference in specificity in favour of the three-step strategy. Conclusions: The IOTA three-step strategy, in addition to being sim-ple to use in clinical practice, has a high diagnostic accuracy for the classification of benignity and malignancy of the adnexal masses, overtaking that of other predictive models such as the LR2 logistic regression model.