甲状腺乳头状癌中央区淋巴结转移预测模型的建立及临床验证

Establishment and clinical validation of a prediction model for central cervical lymph node metastasis in papillary carcinoma of the thyroid

目的:以单中心资料为研究对象,建立包含超声声像图特征的甲状腺乳头状癌(papillary carcinoma of the thyroid,PTC)患者中央区淋巴结转移(central lymph node metastasis,CLNM)预测模型,以辅助临床进行术前评估、针对患者选择个体化的治疗方案。方法:研究共纳入912例经术后病理学检查证实的PTC患者,收集一般情况、实验室指标和超声声像图,将其随机分为训练组(n=727)和验证组(n=185)。在训练组中研究CLNM的相关因素,采用logistic回归分析法建立预测模型,绘制列线图预测CLNM的风险,在验证组中对其效能进行验证,并利用临床决策曲线对其临床应用进行评价。结果:本研究建立了包含性别、年龄、病灶总直径(total tumor diameter,TTD)、促甲状腺激素(thyroid stimulating hormone,TSH)、钙化、病灶与甲状腺包膜的关系和桥本甲状腺炎的预测模型,在验证组和训练组中得到的曲线下面积(area under curve,AUC)分别为0.746和0.826。与术前超声相比,预测模型提高了诊断的灵敏度,并且通过临床决策曲线体现了其临床收益。结论:利用logistic回归建立的列线图预测模型,在预测PTC患者CLNM中体现出了较好的效能,能够补充超声诊断的不足。

Objective:To develop a prediction model for central cervical lymph node metastasis(CLNM)in papillary carcinoma of the thyroid(PTC)by combining the patients’basic clinical information,laboratory indexes,and ultrasound features.This model assists clinical preoperative evaluation and selects personalized treatment plans for patients.Methods:A total of 912 patients who underwent surgery for PTC were enrolled in the study.The enrolled patients were randomly divided into a training dataset(n=727)and a validation dataset(n=185).Univariate and multivariate analysis were performed to examine risk factors associated with CLNM.A nomogram comprising the prognostic model to predict the CLNM was established,and validation was performed in the validation cohort.Finally,clinical decision curve was used to evaluate its clinical application.Results:We developed a prediction model that included gender,age,total tumor diameter(TTD),thyroid stimulating hormone(TSH),calcification,and the thyroid lesion in contact with the thyroid capsule and Hashimoto’s thyroiditis.The nomogram showed area under curve(AUC)of 0.746 and 0.826 in the training cohort and validation cohort.Therefore,the model had higher sensitivity than preoperative ultrasound.Conclusion:The nomogram prediction model developed by logistic regression showed a good performance in predicting CLNM in PTC patients,and addressed the shortcomings of ultrasound diagnosis.