Risk prediction model of malignant partially cystic thyroid nodules / 中华超声影像学杂志

ABSTRACT

Objective A multivariate logistic regression model was built to estimate the risk of malignant partially cystic thyroid nodules.Methods Data of ultrasonographic morphology,microcalcifications,margin,colloid crystals and echogenicity of 766 thyroid nodules (of 506 patients) were analyzed retrospectively.All patients underwent surgery and the diagnosis were confirmed by pathology.The model was built to calculate the individual risk and evaluate the predictive index.Results The regression model was Z =-2.30 + 1.42X2 + 2.39X3 + 1.17X4-1.35 X7 + 0.62X9 ; P =eZ/1 + ez (P represents probability of malignancy,e represents natural constant 2.72).The largest area under the receiver-operating characteristics curve (AUC) was 0.86.When apply the model to the verification group(266 nodules),the accuracy,sensitivity,specificity,misdiagnosis rate,missed diagnosis rate,negative likelihood ratio and positive likelihood ratio were 93.61％,83.33％,94.63％,5.37％,16.67％、0.18 and 15.52 respectively.Conclusions This model based on the ultrasonographic variables increases the diagnostic accuracy of malignancy in patients with partially cystic thyroid nodules.