甲状腺微小乳头状癌中央区淋巴结转移的术前超声预测模型建立

目的采用二元Logistic回归分析建立术前超声预测模型，探讨其对甲状腺微小乳头状癌中央区淋巴结转移风险的预测能力。 方法选取2014年1月至2016年7月就诊于浙江省人民医院，经术后病理确诊为甲状腺微小乳头状癌（PTMC）的患者352例，共有413个腺叶与同侧中央区淋巴结。对其术前超声图像进行回顾性分析，采用二元Logistic回归分析对影响中央区淋巴结转移病理结果的各变量进行逻辑向前回归分析，建立超声预测模型。模型建立后，于2017年1月至2017年10月连续纳入186例PTMC患者作为模型的验证组，共有229个腺叶与同侧中央区淋巴结。应用模型预测186例验证组患者中央区淋巴结转移风险，与术后病理比较，并绘制ROC曲线，以评价模型的预测效果。 结果以术前超声特征为自变量，中央区淋巴结转移（有/无）为因变量，Logistic回归分析结果显示最终进入模型的变量为X₁（癌灶最大径）、X₂（癌灶个数）、X₆（微钙化占癌灶面积）以及X₉（甲状腺包膜外侵犯），建立的风险预测模型为：Y=-2.52+1.36X₁+0.63X₂+2.06X₆+2.19X₉；P=e^Z /1＋e^Z（P是转移概率，e是自然数2.72）。应用模型预测验证组转移风险，与术后病理对比，其准确性为89.08%，敏感度为91.36%，漏诊率为8.64%，特异度为87.84%，误诊率为12.16%，阴性似然比0.098，阳性似然比7.513，ROC曲线下面积为0.931。 结论超声预测模型对PTMC患者中央区淋巴结转移具有较好的预测能力，有助于提高中央区淋巴结转移的术前诊断率，可在一定程度上为临床治疗方案的合理选择提供帮助。

Central lymph node metastasis of papillary thyroid microcarcinoma: development of an ultrasonographic risk-prediction model

ObjectiveTo establish a preoperative ultrasound prediction model by binary Logistic regression analysis to explore its predictive ability for central lymph node metastasis risk of papillary thyroid microcarcinoma. MethodsFrom January 2014 to July 2016, 352 patients diagnosed with PTMC by postoperative pathology at Zhejiang Provincial People's Hospital were selected, with a total of 413 thyroid lobes and ipsilateral central lymph nodes. The preoperative ultrasound images were retrospectively analyzed, and the variables affecting the pathological results of central lymph node metastasis were analyzed by forward Logistic regression to establish an ultrasonic prediction model. After the model was established, 186 consecutive PTMC patients were included from January 2017 to October 2017 as a model validation group, with a total of 229 glandular lobes and ipsilateral central lymph nodes. The model was used to predict the risk of central lymph node metastasis in the validation group, and the result was compared with postoperative pathology. The receiver operating characteristic (ROC) curve was plotted to evaluate the prediction effect of the model. ResultsUsing preoperative ultrasonographic features as independent variables and central lymph node metastasis (with or without) as dependent variables, Logistic regression analysis showed that the variables that finally entered the model were X₁ (maximum diameter of cancer lesion: 5~10 mm), X₂ (multiple cancer lesions), X₆ (micro-calcification area ≥ 1/2 of the cancer lesion area), and X₉ (extra-thyroid invasion). The model was: Y=-2.52+ 1.36X₁+ 0.63X₂+ 2.06X₆+ 2.19X₉; P=e^Z/1+ e^Z (P represents the probability of central lymph node metastasis, and e is the natural constant 2.72). When applying the model to the verification group, the accuracy, sensitivity, missed diagnosis rate, specificity, misdiagnosis rate, positive likelihood ratio, and negative likelihood ratio were 89.08%, 91.36%, 8.64%, 87.84%, 12.16%, 7.513, and 0.098, respectively. The largest area under the ROC curve was 0.931. ConclusionThe ultrasonic prediction model developed in this study has a good prediction ability for central lymph node metastasis in PTMC patients, which is, to a certain extent, helpful to improve the preoperative diagnosis rate and select clinical treatment scheme reasonably.