A stepwise analysis of the diagnostic algorithm for the prediction of malignancy in thyroid nodules

Abstract

BackgroundThe evaluation of the malignancy risk of thyroid nodules involves clinical factors, sonographic characteristics, cytopathology, and molecular profiling. Altogether, this algorithm can be costly and time consuming. We evaluated the stepwise contribution of each diagnostic step toward an accurate prediction of malignancy. MethodsA retrospective study of dominant nodules of 137 patients who underwent surgical excision was performed. A baseline logistic regression model for predicting malignancy was regressed on clinical factors. In a stepwise fashion, the 2015 American Thyroid Association ultrasound risk stratification, Bethesda classification of fine-needle aspiration biopsies, and molecular profiling were added to the baseline model and the significance of each step analyzed using likelihood ratio test. Receiver operating characteristic curves were calculated for each model. ResultsThe addition of American Thyroid Association risk stratification and Bethesda classification to preceding models were statistically significant (P < .001). The addition of molecular profiling (as a strategy independent of a particular test) was not significant (P = .812). The areas under the curve of the baseline model and models sequentially including American Thyroid Association stratification, cytopathology, and molecular profiling were 0.76, 0.85, 0.91, and 0.91, respectively. ConclusionClinical factors, sonographic characteristics, and cytopathology are sufficiently accurate in predicting malignancy risk of most thyroid nodules.