甲状腺囊实性结节恶性概率的多因素回归分析

A multivariate risk prediction model of malignant partially cystic thyroid nodules

目的：建立以超声、检验和临床等多项指标诊断甲状腺囊实性结节恶性病变的综合预估模型。方法对662个经手术证实病理的甲状腺囊实性结节的超声、检验及临床信息进行Logistic回归分析，建立风险预估模型，并应用ROC曲线评价模型预测的诊断效能。结果建立的模型为：Z＝－3.60＋1.40X2＋2.47X3＋1.05X4＋0.57X9＋0.07X10＋1.02X12（ X2为结节内部偏心锐角结构，X3为微钙化， X4为囊实交界面，X9为实性部分回声，X10为促甲状腺激素，X12为性别）；M＝eZ／1＋eZ （ M为结节恶性概率，e为自然常数2.72）。用该模型对262个验证组进行预测，准确率为96.95％，敏感度为100％，漏诊率为0％，特异度为96.68％，误诊率为3.32％，阳性似然比为30.12，阴性似然比为0。该预测模型的ROC曲线下面积为0.88，说明其具有较高的诊断效能。结论该预估模型对预测甲状腺囊实性结节的良恶性具有较高的准确率。据M值（M≥0.5预测结节为恶性，＜0.5为良性）对甲状腺囊实性结节的良恶性进行评估。

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 was：Z=-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 .