Model Development To Predict Central Lymph Node Metastasis In CN0 Papillary Thyroid Microcarcinoma

【Background】Thyroid cancer is the most common endocrine tumor,among which papillary thyroid carcinoma(PTC)is the main one.Papillary thyroid microcarcinoma(PTMC)is defined as PTC with a tumor diameter less than or equal to 1 cm.Its incidence is increasing year by year and its proportion in papillary thyroid carcinoma is also increasing.Lymphatic metastasis is the main way of metastasis of papillary thyroid carcinoma.The preoperative evaluation of cervical lymph nodes mainly relies on ultrasonography,but its accuracy is not enough.Whether prophylactic central lymph node(CLN)resection is necessary for c N0 PTMC is still controversial.Machine learning is more and more widely used in medicine,and it has good performance in the diagnosis and prediction of diseases.The purpose of this study was to explore the risk factors affecting central lymph node metastasis(CLNM)in c N0 PTMC patients,and to develop and validate a model for predicting CLNM risk using machine learning algorithms.【Methods】This study reviewed the clinical data of 1121 c N0 PTMC patients who underwent primary thyroidectomy in our hospital between January 2014 and December 2018.Univariate and multivariate analyzes were performed to explore risk factors associated with CLNM.Six machine learning algorithms and logistic regression model for predicting CLNMs were built and internally validated.We calculate the area under the receiver operating characteristic(AUROC),sensitivity,specificity,accuracy,positive predictive value(PPV)and negative predictive value(NPV)for checking the performance of the model.【Results】Results showed that 33.5%(376 of 1,121 patients)had CLNM.In multivariate regression analysis,it was found that male(OR =2.06,95% CI: 1.53–2.79,P<0.001),tumor size(OR =3.58,95% CI: 1.86–6.88),multifocal(OR = 1.56,95% CI: 1.19–2.03,P=0.001)and extrathyroidal extension(OR =1.85,95% CI: 1.03–3.31,P=0.039)were independent risk factors for CLNM;while patient age(OR =0.96,95% CI: 0.94–0.97,P<0.001)were independent protective factors.The AUROC predictions of the six machine learning algorithms ranged from 0.664 to 0.794,with the random forest model performing best with an AUROC of 0.794.Finally,we used the random forest model and uploaded the results to a web-based risk calculator to predict the patient’s CLNM probability(https://xijing-thyroid.shinyapps.io/ptmc＿clnm).【Conclusions】Gender,age,tumor size,multifocal lesions,and extrathyroidal extension(ETE)were all independent predictors of CLNM.Using machine learning algorithms to develop a CLNM predictive model in c N0 PTMC patients is a feasible approach.Our online risk calculator based on the random forest model promises to be a useful tool for surgical decision-making.