A multivariate risk prediction model of malignant partially cystic thyroid nodules / 中华内分泌外科杂志

ABSTRACT

Objective To develop a multivariate logistic regression model , and to predict the risk of ma-lignant partially cystic thyroid nodules .Methods 470 patients(662 nodules)treated with surgery and confirmed by pathological diagnosis were screened out .Their ultrasonographic morphology and vascularity of thyroid nod-ules, thyroid stimulating hormone(TSH)and clinical information were collected and analyzed retrospectively .The model was developed to calculate the individual risk and ROC curve was used to evaluate the predictive index . Results The regression model wasZ=-3.60+1.40X2+2.47X3+1.05X4+0.57X9+0.07X10+1.02X12 ( X2 represents eccentric acute-angle configuration , X3 represents microcalcification , X4 represents cystic-solid margin,X9 represents echogenecity of solid portion , X10 represents TSH,X12 represents gender );M=eZ/1+eZ (M represents probability of malignancy , e represents natural constant 2.72).When applied the model to the ver-ification group, the accuracy, sensitivity, specificity, misdiagnosis rate, missed diagnosis rate, negative likeli-hood ratio(LR-)and positive likelihood ratio(LR+)was 96.95%,100%, 96.68%, 3.32%, 0%, 0 and 30.12 respectively.The largest area under the receiver-operating characteristics curve (AUC)was 0.88,which proved the model has high diagnostic value .Conclusions The model has high accuracy to predict the risk of malignan-cy.M is closely related to malignant risk of partially cystic thyroid nodules .