甲状腺囊实性结节良恶性鉴别的超声预估模型建立

Risk prediction model of malignant partially cystic thyroid nodules

目的建立Logistic回归模型，预测超声诊断甲状腺囊实性结节恶性病变的风险。方法回顾性分析766个甲状腺囊实性结节的内部结构、结节内微钙化、结节囊实交界面、结节囊性部分胶质结晶、结节实性部分回声等超声图像信息，建立Logistic回归模型，与病理金标准对照，验证并评价模型的诊断效能。结果建立的模型为：Z=-2．30＋1．42X2＋2．39X3＋1．17X4—1．35X7＋0．62X9；P=e^z／1＋e^2（P为患病概率，e为自然数2．72）。该预测模型的ROC曲线下面积为0．86，说明其具有较高的诊断效能。用该模型对验证组266个结节进行预测，准确性为93．61％，敏感性为83．33％，漏诊率为16．67％，特异性为94．63％，误诊率为5．37％，阳性似然比为15．52，阴性似然比为0．18。结论该回归模型对于预测甲状腺囊实性结节的良恶性具有较高的准确性。

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 - 1.35X7 ＋ 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.